Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
d6d48196 JA |
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> | |
fe45e630 | 9 | #include <linux/blk-integrity.h> |
d6d48196 | 10 | #include <linux/scatterlist.h> |
82d981d4 | 11 | #include <linux/part_stat.h> |
6b2b0459 | 12 | #include <linux/blk-cgroup.h> |
d6d48196 | 13 | |
cda22646 MK |
14 | #include <trace/events/block.h> |
15 | ||
d6d48196 | 16 | #include "blk.h" |
2aa7745b | 17 | #include "blk-mq-sched.h" |
8e756373 | 18 | #include "blk-rq-qos.h" |
a7b36ee6 | 19 | #include "blk-throttle.h" |
d6d48196 | 20 | |
ff18d77b CH |
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 | ||
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 | ||
e9907009 CH |
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); | |
df376b2e | 69 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
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); | |
200a9aff | 83 | if (biovec_phys_mergeable(q, &pb, &nb)) |
e9907009 CH |
84 | return false; |
85 | return __bvec_gap_to_prev(q, &pb, nb.bv_offset); | |
86 | } | |
87 | ||
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 | ||
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 | ||
54efd50b KO |
98 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
99 | struct bio *bio, | |
bdced438 ML |
100 | struct bio_set *bs, |
101 | unsigned *nsegs) | |
54efd50b KO |
102 | { |
103 | unsigned int max_discard_sectors, granularity; | |
104 | int alignment; | |
105 | sector_t tmp; | |
106 | unsigned split_sectors; | |
107 | ||
bdced438 ML |
108 | *nsegs = 1; |
109 | ||
54efd50b KO |
110 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
111 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
112 | ||
1adfc5e4 ML |
113 | max_discard_sectors = min(q->limits.max_discard_sectors, |
114 | bio_allowed_max_sectors(q)); | |
54efd50b KO |
115 | max_discard_sectors -= max_discard_sectors % granularity; |
116 | ||
117 | if (unlikely(!max_discard_sectors)) { | |
118 | /* XXX: warn */ | |
119 | return NULL; | |
120 | } | |
121 | ||
122 | if (bio_sectors(bio) <= max_discard_sectors) | |
123 | return NULL; | |
124 | ||
125 | split_sectors = max_discard_sectors; | |
126 | ||
127 | /* | |
128 | * If the next starting sector would be misaligned, stop the discard at | |
129 | * the previous aligned sector. | |
130 | */ | |
131 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
132 | ||
133 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
134 | tmp = sector_div(tmp, granularity); | |
135 | ||
136 | if (split_sectors > tmp) | |
137 | split_sectors -= tmp; | |
138 | ||
139 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
140 | } | |
141 | ||
885fa13f CH |
142 | static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, |
143 | struct bio *bio, struct bio_set *bs, unsigned *nsegs) | |
144 | { | |
d665e12a | 145 | *nsegs = 0; |
885fa13f CH |
146 | |
147 | if (!q->limits.max_write_zeroes_sectors) | |
148 | return NULL; | |
149 | ||
150 | if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) | |
151 | return NULL; | |
152 | ||
153 | return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); | |
154 | } | |
155 | ||
9cc5169c BVA |
156 | /* |
157 | * Return the maximum number of sectors from the start of a bio that may be | |
158 | * submitted as a single request to a block device. If enough sectors remain, | |
159 | * align the end to the physical block size. Otherwise align the end to the | |
160 | * logical block size. This approach minimizes the number of non-aligned | |
161 | * requests that are submitted to a block device if the start of a bio is not | |
162 | * aligned to a physical block boundary. | |
163 | */ | |
d0e5fbb0 ML |
164 | static inline unsigned get_max_io_size(struct request_queue *q, |
165 | struct bio *bio) | |
166 | { | |
3ee16db3 | 167 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector, 0); |
9cc5169c BVA |
168 | unsigned max_sectors = sectors; |
169 | unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT; | |
170 | unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT; | |
171 | unsigned start_offset = bio->bi_iter.bi_sector & (pbs - 1); | |
d0e5fbb0 | 172 | |
9cc5169c BVA |
173 | max_sectors += start_offset; |
174 | max_sectors &= ~(pbs - 1); | |
175 | if (max_sectors > start_offset) | |
176 | return max_sectors - start_offset; | |
d0e5fbb0 | 177 | |
e4b469c6 | 178 | return sectors & ~(lbs - 1); |
d0e5fbb0 ML |
179 | } |
180 | ||
429120f3 ML |
181 | static inline unsigned get_max_segment_size(const struct request_queue *q, |
182 | struct page *start_page, | |
183 | unsigned long offset) | |
dcebd755 ML |
184 | { |
185 | unsigned long mask = queue_segment_boundary(q); | |
186 | ||
429120f3 | 187 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
188 | |
189 | /* | |
190 | * overflow may be triggered in case of zero page physical address | |
191 | * on 32bit arch, use queue's max segment size when that happens. | |
192 | */ | |
193 | return min_not_zero(mask - offset + 1, | |
194 | (unsigned long)queue_max_segment_size(q)); | |
dcebd755 ML |
195 | } |
196 | ||
708b25b3 BVA |
197 | /** |
198 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
199 | * @q: [in] request queue associated with the bio associated with @bv | |
200 | * @bv: [in] bvec to examine | |
201 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
202 | * by the number of segments from @bv that may be appended to that | |
203 | * bio without exceeding @max_segs | |
204 | * @sectors: [in,out] Number of sectors in the bio being built. Incremented | |
205 | * by the number of sectors from @bv that may be appended to that | |
206 | * bio without exceeding @max_sectors | |
207 | * @max_segs: [in] upper bound for *@nsegs | |
208 | * @max_sectors: [in] upper bound for *@sectors | |
209 | * | |
210 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
211 | * big to fit in a single segment and hence that it has to be split in the | |
212 | * middle. This function verifies whether or not that should happen. The value | |
213 | * %true is returned if and only if appending the entire @bv to a bio with | |
214 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
215 | * the block driver. | |
dcebd755 | 216 | */ |
af2c68fe BVA |
217 | static bool bvec_split_segs(const struct request_queue *q, |
218 | const struct bio_vec *bv, unsigned *nsegs, | |
708b25b3 BVA |
219 | unsigned *sectors, unsigned max_segs, |
220 | unsigned max_sectors) | |
dcebd755 | 221 | { |
708b25b3 BVA |
222 | unsigned max_len = (min(max_sectors, UINT_MAX >> 9) - *sectors) << 9; |
223 | unsigned len = min(bv->bv_len, max_len); | |
dcebd755 | 224 | unsigned total_len = 0; |
ff9811b3 | 225 | unsigned seg_size = 0; |
dcebd755 | 226 | |
ff9811b3 | 227 | while (len && *nsegs < max_segs) { |
429120f3 ML |
228 | seg_size = get_max_segment_size(q, bv->bv_page, |
229 | bv->bv_offset + total_len); | |
dcebd755 ML |
230 | seg_size = min(seg_size, len); |
231 | ||
ff9811b3 | 232 | (*nsegs)++; |
dcebd755 ML |
233 | total_len += seg_size; |
234 | len -= seg_size; | |
235 | ||
236 | if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) | |
237 | break; | |
238 | } | |
239 | ||
ff9811b3 | 240 | *sectors += total_len >> 9; |
dcebd755 | 241 | |
708b25b3 BVA |
242 | /* tell the caller to split the bvec if it is too big to fit */ |
243 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
244 | } |
245 | ||
dad77584 BVA |
246 | /** |
247 | * blk_bio_segment_split - split a bio in two bios | |
248 | * @q: [in] request queue pointer | |
249 | * @bio: [in] bio to be split | |
250 | * @bs: [in] bio set to allocate the clone from | |
251 | * @segs: [out] number of segments in the bio with the first half of the sectors | |
252 | * | |
253 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
254 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
255 | * following is guaranteed for the cloned bio: | |
256 | * - That it has at most get_max_io_size(@q, @bio) sectors. | |
257 | * - That it has at most queue_max_segments(@q) segments. | |
258 | * | |
259 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
260 | * the original bio. It is the responsibility of the caller to ensure that the | |
261 | * original bio is not freed before the cloned bio. The caller is also | |
262 | * responsible for ensuring that @bs is only destroyed after processing of the | |
263 | * split bio has finished. | |
264 | */ | |
54efd50b KO |
265 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
266 | struct bio *bio, | |
bdced438 ML |
267 | struct bio_set *bs, |
268 | unsigned *segs) | |
54efd50b | 269 | { |
5014c311 | 270 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 271 | struct bvec_iter iter; |
6869875f | 272 | unsigned nsegs = 0, sectors = 0; |
d0e5fbb0 | 273 | const unsigned max_sectors = get_max_io_size(q, bio); |
05b700ba | 274 | const unsigned max_segs = queue_max_segments(q); |
54efd50b | 275 | |
dcebd755 | 276 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
277 | /* |
278 | * If the queue doesn't support SG gaps and adding this | |
279 | * offset would create a gap, disallow it. | |
280 | */ | |
5014c311 | 281 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
282 | goto split; |
283 | ||
708b25b3 BVA |
284 | if (nsegs < max_segs && |
285 | sectors + (bv.bv_len >> 9) <= max_sectors && | |
286 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { | |
287 | nsegs++; | |
288 | sectors += bv.bv_len >> 9; | |
289 | } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs, | |
290 | max_sectors)) { | |
cf8c0c6a | 291 | goto split; |
e36f6204 KB |
292 | } |
293 | ||
54efd50b | 294 | bvprv = bv; |
578270bf | 295 | bvprvp = &bvprv; |
54efd50b KO |
296 | } |
297 | ||
d627065d CH |
298 | *segs = nsegs; |
299 | return NULL; | |
54efd50b | 300 | split: |
bdced438 | 301 | *segs = nsegs; |
cc29e1bf JX |
302 | |
303 | /* | |
304 | * Bio splitting may cause subtle trouble such as hang when doing sync | |
305 | * iopoll in direct IO routine. Given performance gain of iopoll for | |
306 | * big IO can be trival, disable iopoll when split needed. | |
307 | */ | |
6ce913fe | 308 | bio_clear_polled(bio); |
d627065d | 309 | return bio_split(bio, sectors, GFP_NOIO, bs); |
54efd50b KO |
310 | } |
311 | ||
dad77584 BVA |
312 | /** |
313 | * __blk_queue_split - split a bio and submit the second half | |
abd45c15 | 314 | * @q: [in] request_queue new bio is being queued at |
dad77584 BVA |
315 | * @bio: [in, out] bio to be split |
316 | * @nr_segs: [out] number of segments in the first bio | |
317 | * | |
318 | * Split a bio into two bios, chain the two bios, submit the second half and | |
319 | * store a pointer to the first half in *@bio. If the second bio is still too | |
320 | * big it will be split by a recursive call to this function. Since this | |
309dca30 CH |
321 | * function may allocate a new bio from q->bio_split, it is the responsibility |
322 | * of the caller to ensure that q->bio_split is only released after processing | |
323 | * of the split bio has finished. | |
dad77584 | 324 | */ |
abd45c15 JA |
325 | void __blk_queue_split(struct request_queue *q, struct bio **bio, |
326 | unsigned int *nr_segs) | |
54efd50b | 327 | { |
fa532287 | 328 | struct bio *split = NULL; |
54efd50b | 329 | |
7afafc8a AH |
330 | switch (bio_op(*bio)) { |
331 | case REQ_OP_DISCARD: | |
332 | case REQ_OP_SECURE_ERASE: | |
14ccb66b | 333 | split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a | 334 | break; |
a6f0788e | 335 | case REQ_OP_WRITE_ZEROES: |
14ccb66b CH |
336 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, |
337 | nr_segs); | |
a6f0788e | 338 | break; |
7afafc8a | 339 | default: |
14ccb66b | 340 | split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a AH |
341 | break; |
342 | } | |
bdced438 | 343 | |
54efd50b | 344 | if (split) { |
6ac45aeb | 345 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 346 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 347 | |
54efd50b | 348 | bio_chain(split, *bio); |
eb6f7f7c | 349 | trace_block_split(split, (*bio)->bi_iter.bi_sector); |
ed00aabd | 350 | submit_bio_noacct(*bio); |
54efd50b KO |
351 | *bio = split; |
352 | } | |
353 | } | |
14ccb66b | 354 | |
dad77584 BVA |
355 | /** |
356 | * blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
357 | * @bio: [in, out] bio to be split |
358 | * | |
359 | * Split a bio into two bios, chains the two bios, submit the second half and | |
360 | * store a pointer to the first half in *@bio. Since this function may allocate | |
309dca30 CH |
361 | * a new bio from q->bio_split, it is the responsibility of the caller to ensure |
362 | * that q->bio_split is only released after processing of the split bio has | |
363 | * finished. | |
dad77584 | 364 | */ |
f695ca38 | 365 | void blk_queue_split(struct bio **bio) |
14ccb66b | 366 | { |
859897c3 | 367 | struct request_queue *q = bdev_get_queue((*bio)->bi_bdev); |
14ccb66b CH |
368 | unsigned int nr_segs; |
369 | ||
abd45c15 JA |
370 | if (blk_may_split(q, *bio)) |
371 | __blk_queue_split(q, bio, &nr_segs); | |
14ccb66b | 372 | } |
54efd50b KO |
373 | EXPORT_SYMBOL(blk_queue_split); |
374 | ||
e9cd19c0 | 375 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 376 | { |
6869875f | 377 | unsigned int nr_phys_segs = 0; |
ff9811b3 | 378 | unsigned int nr_sectors = 0; |
e9cd19c0 | 379 | struct req_iterator iter; |
6869875f | 380 | struct bio_vec bv; |
d6d48196 | 381 | |
e9cd19c0 | 382 | if (!rq->bio) |
1e428079 | 383 | return 0; |
d6d48196 | 384 | |
e9cd19c0 | 385 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
386 | case REQ_OP_DISCARD: |
387 | case REQ_OP_SECURE_ERASE: | |
a958937f DJ |
388 | if (queue_max_discard_segments(rq->q) > 1) { |
389 | struct bio *bio = rq->bio; | |
390 | ||
391 | for_each_bio(bio) | |
392 | nr_phys_segs++; | |
393 | return nr_phys_segs; | |
394 | } | |
395 | return 1; | |
a6f0788e | 396 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 | 397 | return 0; |
a6f0788e | 398 | } |
5cb8850c | 399 | |
e9cd19c0 | 400 | rq_for_each_bvec(bv, rq, iter) |
ff9811b3 | 401 | bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors, |
708b25b3 | 402 | UINT_MAX, UINT_MAX); |
1e428079 JA |
403 | return nr_phys_segs; |
404 | } | |
405 | ||
48d7727c | 406 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
407 | struct scatterlist *sglist) |
408 | { | |
409 | if (!*sg) | |
410 | return sglist; | |
411 | ||
412 | /* | |
413 | * If the driver previously mapped a shorter list, we could see a | |
414 | * termination bit prematurely unless it fully inits the sg table | |
415 | * on each mapping. We KNOW that there must be more entries here | |
416 | * or the driver would be buggy, so force clear the termination bit | |
417 | * to avoid doing a full sg_init_table() in drivers for each command. | |
418 | */ | |
419 | sg_unmark_end(*sg); | |
420 | return sg_next(*sg); | |
421 | } | |
422 | ||
423 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
424 | struct bio_vec *bvec, struct scatterlist *sglist, | |
425 | struct scatterlist **sg) | |
426 | { | |
427 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 428 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
429 | |
430 | while (nbytes > 0) { | |
8a96a0e4 | 431 | unsigned offset = bvec->bv_offset + total; |
429120f3 ML |
432 | unsigned len = min(get_max_segment_size(q, bvec->bv_page, |
433 | offset), nbytes); | |
f9f76879 CH |
434 | struct page *page = bvec->bv_page; |
435 | ||
436 | /* | |
437 | * Unfortunately a fair number of drivers barf on scatterlists | |
438 | * that have an offset larger than PAGE_SIZE, despite other | |
439 | * subsystems dealing with that invariant just fine. For now | |
440 | * stick to the legacy format where we never present those from | |
441 | * the block layer, but the code below should be removed once | |
442 | * these offenders (mostly MMC/SD drivers) are fixed. | |
443 | */ | |
444 | page += (offset >> PAGE_SHIFT); | |
445 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
446 | |
447 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 448 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 449 | |
8a96a0e4 CH |
450 | total += len; |
451 | nbytes -= len; | |
862e5a5e ML |
452 | nsegs++; |
453 | } | |
454 | ||
455 | return nsegs; | |
456 | } | |
457 | ||
16e3e418 ML |
458 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
459 | struct scatterlist *sglist, struct scatterlist **sg) | |
460 | { | |
461 | *sg = blk_next_sg(sg, sglist); | |
462 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
463 | return 1; | |
464 | } | |
465 | ||
f6970f83 ML |
466 | /* only try to merge bvecs into one sg if they are from two bios */ |
467 | static inline bool | |
468 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
469 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
470 | { |
471 | ||
472 | int nbytes = bvec->bv_len; | |
473 | ||
f6970f83 ML |
474 | if (!*sg) |
475 | return false; | |
963ab9e5 | 476 | |
f6970f83 ML |
477 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
478 | return false; | |
479 | ||
480 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
481 | return false; | |
482 | ||
483 | (*sg)->length += nbytes; | |
484 | ||
485 | return true; | |
963ab9e5 AH |
486 | } |
487 | ||
5cb8850c KO |
488 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
489 | struct scatterlist *sglist, | |
490 | struct scatterlist **sg) | |
d6d48196 | 491 | { |
3f649ab7 | 492 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 493 | struct bvec_iter iter; |
38417468 | 494 | int nsegs = 0; |
f6970f83 | 495 | bool new_bio = false; |
5cb8850c | 496 | |
f6970f83 ML |
497 | for_each_bio(bio) { |
498 | bio_for_each_bvec(bvec, bio, iter) { | |
499 | /* | |
500 | * Only try to merge bvecs from two bios given we | |
501 | * have done bio internal merge when adding pages | |
502 | * to bio | |
503 | */ | |
504 | if (new_bio && | |
505 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
506 | goto next_bvec; | |
507 | ||
508 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
509 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
510 | else | |
511 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
512 | next_bvec: | |
513 | new_bio = false; | |
514 | } | |
b21e11c5 ML |
515 | if (likely(bio->bi_iter.bi_size)) { |
516 | bvprv = bvec; | |
517 | new_bio = true; | |
518 | } | |
f6970f83 | 519 | } |
d6d48196 | 520 | |
5cb8850c KO |
521 | return nsegs; |
522 | } | |
523 | ||
524 | /* | |
525 | * map a request to scatterlist, return number of sg entries setup. Caller | |
526 | * must make sure sg can hold rq->nr_phys_segments entries | |
527 | */ | |
89de1504 CH |
528 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
529 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 530 | { |
5cb8850c KO |
531 | int nsegs = 0; |
532 | ||
f9d03f96 | 533 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 534 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 535 | else if (rq->bio) |
89de1504 | 536 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 537 | |
89de1504 CH |
538 | if (*last_sg) |
539 | sg_mark_end(*last_sg); | |
d6d48196 | 540 | |
12e57f59 ML |
541 | /* |
542 | * Something must have been wrong if the figured number of | |
543 | * segment is bigger than number of req's physical segments | |
544 | */ | |
f9d03f96 | 545 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 546 | |
d6d48196 JA |
547 | return nsegs; |
548 | } | |
89de1504 | 549 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 550 | |
943b40c8 ML |
551 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
552 | { | |
553 | if (req_op(rq) == REQ_OP_DISCARD) | |
554 | return queue_max_discard_segments(rq->q); | |
555 | return queue_max_segments(rq->q); | |
556 | } | |
557 | ||
badf7f64 CH |
558 | static inline unsigned int blk_rq_get_max_sectors(struct request *rq, |
559 | sector_t offset) | |
560 | { | |
561 | struct request_queue *q = rq->q; | |
562 | ||
563 | if (blk_rq_is_passthrough(rq)) | |
564 | return q->limits.max_hw_sectors; | |
565 | ||
566 | if (!q->limits.chunk_sectors || | |
567 | req_op(rq) == REQ_OP_DISCARD || | |
568 | req_op(rq) == REQ_OP_SECURE_ERASE) | |
569 | return blk_queue_get_max_sectors(q, req_op(rq)); | |
570 | ||
571 | return min(blk_max_size_offset(q, offset, 0), | |
572 | blk_queue_get_max_sectors(q, req_op(rq))); | |
573 | } | |
574 | ||
14ccb66b CH |
575 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
576 | unsigned int nr_phys_segs) | |
d6d48196 | 577 | { |
6b2b0459 TH |
578 | if (!blk_cgroup_mergeable(req, bio)) |
579 | goto no_merge; | |
580 | ||
2705dfb2 | 581 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d MP |
582 | goto no_merge; |
583 | ||
2705dfb2 ML |
584 | /* discard request merge won't add new segment */ |
585 | if (req_op(req) == REQ_OP_DISCARD) | |
586 | return 1; | |
587 | ||
588 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) | |
13f05c8d | 589 | goto no_merge; |
d6d48196 JA |
590 | |
591 | /* | |
592 | * This will form the start of a new hw segment. Bump both | |
593 | * counters. | |
594 | */ | |
d6d48196 JA |
595 | req->nr_phys_segments += nr_phys_segs; |
596 | return 1; | |
13f05c8d MP |
597 | |
598 | no_merge: | |
14ccb66b | 599 | req_set_nomerge(req->q, req); |
13f05c8d | 600 | return 0; |
d6d48196 JA |
601 | } |
602 | ||
14ccb66b | 603 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 604 | { |
5e7c4274 JA |
605 | if (req_gap_back_merge(req, bio)) |
606 | return 0; | |
7f39add3 SG |
607 | if (blk_integrity_rq(req) && |
608 | integrity_req_gap_back_merge(req, bio)) | |
609 | return 0; | |
a892c8d5 ST |
610 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
611 | return 0; | |
f31dc1cd | 612 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 613 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 614 | req_set_nomerge(req->q, req); |
d6d48196 JA |
615 | return 0; |
616 | } | |
d6d48196 | 617 | |
14ccb66b | 618 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
619 | } |
620 | ||
eda5cc99 CH |
621 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
622 | unsigned int nr_segs) | |
d6d48196 | 623 | { |
5e7c4274 JA |
624 | if (req_gap_front_merge(req, bio)) |
625 | return 0; | |
7f39add3 SG |
626 | if (blk_integrity_rq(req) && |
627 | integrity_req_gap_front_merge(req, bio)) | |
628 | return 0; | |
a892c8d5 ST |
629 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
630 | return 0; | |
f31dc1cd | 631 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 632 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 633 | req_set_nomerge(req->q, req); |
d6d48196 JA |
634 | return 0; |
635 | } | |
d6d48196 | 636 | |
14ccb66b | 637 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
638 | } |
639 | ||
445251d0 JA |
640 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
641 | struct request *next) | |
642 | { | |
643 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
644 | ||
645 | if (segments >= queue_max_discard_segments(q)) | |
646 | goto no_merge; | |
647 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
648 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
649 | goto no_merge; | |
650 | ||
651 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
652 | return true; | |
653 | no_merge: | |
654 | req_set_nomerge(q, req); | |
655 | return false; | |
656 | } | |
657 | ||
d6d48196 JA |
658 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
659 | struct request *next) | |
660 | { | |
661 | int total_phys_segments; | |
d6d48196 | 662 | |
5e7c4274 | 663 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
664 | return 0; |
665 | ||
d6d48196 JA |
666 | /* |
667 | * Will it become too large? | |
668 | */ | |
f31dc1cd | 669 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 670 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
671 | return 0; |
672 | ||
673 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 674 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
675 | return 0; |
676 | ||
6b2b0459 TH |
677 | if (!blk_cgroup_mergeable(req, next->bio)) |
678 | return 0; | |
679 | ||
4eaf99be | 680 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
681 | return 0; |
682 | ||
a892c8d5 ST |
683 | if (!bio_crypt_ctx_merge_rq(req, next)) |
684 | return 0; | |
685 | ||
d6d48196 JA |
686 | /* Merge is OK... */ |
687 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
688 | return 1; |
689 | } | |
690 | ||
80a761fd TH |
691 | /** |
692 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
693 | * @rq: request to mark as mixed merge | |
694 | * | |
695 | * Description: | |
696 | * @rq is about to be mixed merged. Make sure the attributes | |
697 | * which can be mixed are set in each bio and mark @rq as mixed | |
698 | * merged. | |
699 | */ | |
700 | void blk_rq_set_mixed_merge(struct request *rq) | |
701 | { | |
702 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
703 | struct bio *bio; | |
704 | ||
e8064021 | 705 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
706 | return; |
707 | ||
708 | /* | |
709 | * @rq will no longer represent mixable attributes for all the | |
710 | * contained bios. It will just track those of the first one. | |
711 | * Distributes the attributs to each bio. | |
712 | */ | |
713 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
714 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
715 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
716 | bio->bi_opf |= ff; | |
80a761fd | 717 | } |
e8064021 | 718 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
719 | } |
720 | ||
b9c54f56 | 721 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
722 | { |
723 | if (blk_do_io_stat(req)) { | |
112f158f | 724 | part_stat_lock(); |
b9c54f56 | 725 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab JM |
726 | part_stat_unlock(); |
727 | } | |
728 | } | |
b9c54f56 | 729 | |
e96c0d83 EB |
730 | static enum elv_merge blk_try_req_merge(struct request *req, |
731 | struct request *next) | |
69840466 JW |
732 | { |
733 | if (blk_discard_mergable(req)) | |
734 | return ELEVATOR_DISCARD_MERGE; | |
735 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
736 | return ELEVATOR_BACK_MERGE; | |
737 | ||
738 | return ELEVATOR_NO_MERGE; | |
739 | } | |
26308eab | 740 | |
d6d48196 | 741 | /* |
b973cb7e JA |
742 | * For non-mq, this has to be called with the request spinlock acquired. |
743 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 744 | */ |
b973cb7e JA |
745 | static struct request *attempt_merge(struct request_queue *q, |
746 | struct request *req, struct request *next) | |
d6d48196 JA |
747 | { |
748 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 749 | return NULL; |
d6d48196 | 750 | |
288dab8a | 751 | if (req_op(req) != req_op(next)) |
b973cb7e | 752 | return NULL; |
f31dc1cd | 753 | |
79bb1dbd | 754 | if (rq_data_dir(req) != rq_data_dir(next)) |
b973cb7e | 755 | return NULL; |
d6d48196 | 756 | |
668ffc03 DLM |
757 | if (req->ioprio != next->ioprio) |
758 | return NULL; | |
759 | ||
d6d48196 JA |
760 | /* |
761 | * If we are allowed to merge, then append bio list | |
762 | * from next to rq and release next. merge_requests_fn | |
763 | * will have updated segment counts, update sector | |
445251d0 JA |
764 | * counts here. Handle DISCARDs separately, as they |
765 | * have separate settings. | |
d6d48196 | 766 | */ |
69840466 JW |
767 | |
768 | switch (blk_try_req_merge(req, next)) { | |
769 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
770 | if (!req_attempt_discard_merge(q, req, next)) |
771 | return NULL; | |
69840466 JW |
772 | break; |
773 | case ELEVATOR_BACK_MERGE: | |
774 | if (!ll_merge_requests_fn(q, req, next)) | |
775 | return NULL; | |
776 | break; | |
777 | default: | |
b973cb7e | 778 | return NULL; |
69840466 | 779 | } |
d6d48196 | 780 | |
80a761fd TH |
781 | /* |
782 | * If failfast settings disagree or any of the two is already | |
783 | * a mixed merge, mark both as mixed before proceeding. This | |
784 | * makes sure that all involved bios have mixable attributes | |
785 | * set properly. | |
786 | */ | |
e8064021 | 787 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
788 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
789 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
790 | blk_rq_set_mixed_merge(req); | |
791 | blk_rq_set_mixed_merge(next); | |
792 | } | |
793 | ||
d6d48196 | 794 | /* |
522a7775 OS |
795 | * At this point we have either done a back merge or front merge. We |
796 | * need the smaller start_time_ns of the merged requests to be the | |
797 | * current request for accounting purposes. | |
d6d48196 | 798 | */ |
522a7775 OS |
799 | if (next->start_time_ns < req->start_time_ns) |
800 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
801 | |
802 | req->biotail->bi_next = next->bio; | |
803 | req->biotail = next->biotail; | |
804 | ||
a2dec7b3 | 805 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 806 | |
2a5cf35c | 807 | if (!blk_discard_mergable(req)) |
445251d0 | 808 | elv_merge_requests(q, req, next); |
d6d48196 | 809 | |
42dad764 JM |
810 | /* |
811 | * 'next' is going away, so update stats accordingly | |
812 | */ | |
b9c54f56 | 813 | blk_account_io_merge_request(next); |
d6d48196 | 814 | |
a54895fa | 815 | trace_block_rq_merge(next); |
f3bdc62f | 816 | |
e4d750c9 JA |
817 | /* |
818 | * ownership of bio passed from next to req, return 'next' for | |
819 | * the caller to free | |
820 | */ | |
1cd96c24 | 821 | next->bio = NULL; |
b973cb7e | 822 | return next; |
d6d48196 JA |
823 | } |
824 | ||
eda5cc99 CH |
825 | static struct request *attempt_back_merge(struct request_queue *q, |
826 | struct request *rq) | |
d6d48196 JA |
827 | { |
828 | struct request *next = elv_latter_request(q, rq); | |
829 | ||
830 | if (next) | |
831 | return attempt_merge(q, rq, next); | |
832 | ||
b973cb7e | 833 | return NULL; |
d6d48196 JA |
834 | } |
835 | ||
eda5cc99 CH |
836 | static struct request *attempt_front_merge(struct request_queue *q, |
837 | struct request *rq) | |
d6d48196 JA |
838 | { |
839 | struct request *prev = elv_former_request(q, rq); | |
840 | ||
841 | if (prev) | |
842 | return attempt_merge(q, prev, rq); | |
843 | ||
b973cb7e | 844 | return NULL; |
d6d48196 | 845 | } |
5e84ea3a | 846 | |
fd2ef39c JK |
847 | /* |
848 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false | |
849 | * otherwise. The caller is responsible for freeing 'next' if the merge | |
850 | * happened. | |
851 | */ | |
852 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
853 | struct request *next) | |
5e84ea3a | 854 | { |
fd2ef39c | 855 | return attempt_merge(q, rq, next); |
5e84ea3a | 856 | } |
050c8ea8 TH |
857 | |
858 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
859 | { | |
e2a60da7 | 860 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
861 | return false; |
862 | ||
288dab8a | 863 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
864 | return false; |
865 | ||
050c8ea8 TH |
866 | /* different data direction or already started, don't merge */ |
867 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
868 | return false; | |
869 | ||
6b2b0459 TH |
870 | /* don't merge across cgroup boundaries */ |
871 | if (!blk_cgroup_mergeable(rq, bio)) | |
872 | return false; | |
873 | ||
050c8ea8 | 874 | /* only merge integrity protected bio into ditto rq */ |
4eaf99be | 875 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
876 | return false; |
877 | ||
a892c8d5 ST |
878 | /* Only merge if the crypt contexts are compatible */ |
879 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
880 | return false; | |
881 | ||
668ffc03 DLM |
882 | if (rq->ioprio != bio_prio(bio)) |
883 | return false; | |
884 | ||
050c8ea8 TH |
885 | return true; |
886 | } | |
887 | ||
34fe7c05 | 888 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 889 | { |
69840466 | 890 | if (blk_discard_mergable(rq)) |
1e739730 CH |
891 | return ELEVATOR_DISCARD_MERGE; |
892 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 893 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 894 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
895 | return ELEVATOR_FRONT_MERGE; |
896 | return ELEVATOR_NO_MERGE; | |
897 | } | |
8e756373 BW |
898 | |
899 | static void blk_account_io_merge_bio(struct request *req) | |
900 | { | |
901 | if (!blk_do_io_stat(req)) | |
902 | return; | |
903 | ||
904 | part_stat_lock(); | |
905 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
906 | part_stat_unlock(); | |
907 | } | |
908 | ||
eda5cc99 CH |
909 | enum bio_merge_status { |
910 | BIO_MERGE_OK, | |
911 | BIO_MERGE_NONE, | |
912 | BIO_MERGE_FAILED, | |
913 | }; | |
914 | ||
915 | static enum bio_merge_status bio_attempt_back_merge(struct request *req, | |
916 | struct bio *bio, unsigned int nr_segs) | |
8e756373 BW |
917 | { |
918 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
919 | ||
920 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 921 | return BIO_MERGE_FAILED; |
8e756373 | 922 | |
e8a676d6 | 923 | trace_block_bio_backmerge(bio); |
8e756373 BW |
924 | rq_qos_merge(req->q, req, bio); |
925 | ||
926 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
927 | blk_rq_set_mixed_merge(req); | |
928 | ||
929 | req->biotail->bi_next = bio; | |
930 | req->biotail = bio; | |
931 | req->__data_len += bio->bi_iter.bi_size; | |
932 | ||
933 | bio_crypt_free_ctx(bio); | |
934 | ||
935 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 936 | return BIO_MERGE_OK; |
8e756373 BW |
937 | } |
938 | ||
eda5cc99 CH |
939 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
940 | struct bio *bio, unsigned int nr_segs) | |
8e756373 BW |
941 | { |
942 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
943 | ||
944 | if (!ll_front_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 945 | return BIO_MERGE_FAILED; |
8e756373 | 946 | |
e8a676d6 | 947 | trace_block_bio_frontmerge(bio); |
8e756373 BW |
948 | rq_qos_merge(req->q, req, bio); |
949 | ||
950 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
951 | blk_rq_set_mixed_merge(req); | |
952 | ||
953 | bio->bi_next = req->bio; | |
954 | req->bio = bio; | |
955 | ||
956 | req->__sector = bio->bi_iter.bi_sector; | |
957 | req->__data_len += bio->bi_iter.bi_size; | |
958 | ||
959 | bio_crypt_do_front_merge(req, bio); | |
960 | ||
961 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 962 | return BIO_MERGE_OK; |
8e756373 BW |
963 | } |
964 | ||
eda5cc99 CH |
965 | static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
966 | struct request *req, struct bio *bio) | |
8e756373 BW |
967 | { |
968 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
969 | ||
970 | if (segments >= queue_max_discard_segments(q)) | |
971 | goto no_merge; | |
972 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
973 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
974 | goto no_merge; | |
975 | ||
976 | rq_qos_merge(q, req, bio); | |
977 | ||
978 | req->biotail->bi_next = bio; | |
979 | req->biotail = bio; | |
980 | req->__data_len += bio->bi_iter.bi_size; | |
981 | req->nr_phys_segments = segments + 1; | |
982 | ||
983 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 984 | return BIO_MERGE_OK; |
8e756373 BW |
985 | no_merge: |
986 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
987 | return BIO_MERGE_FAILED; |
988 | } | |
989 | ||
990 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
991 | struct request *rq, | |
992 | struct bio *bio, | |
993 | unsigned int nr_segs, | |
994 | bool sched_allow_merge) | |
995 | { | |
996 | if (!blk_rq_merge_ok(rq, bio)) | |
997 | return BIO_MERGE_NONE; | |
998 | ||
999 | switch (blk_try_merge(rq, bio)) { | |
1000 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 1001 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1002 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1003 | break; | |
1004 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1005 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1006 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1007 | break; | |
1008 | case ELEVATOR_DISCARD_MERGE: | |
1009 | return bio_attempt_discard_merge(q, rq, bio); | |
1010 | default: | |
1011 | return BIO_MERGE_NONE; | |
1012 | } | |
1013 | ||
1014 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1015 | } |
1016 | ||
1017 | /** | |
1018 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1019 | * @q: request_queue new bio is being queued at | |
1020 | * @bio: new bio being queued | |
1021 | * @nr_segs: number of segments in @bio | |
87c037d1 | 1022 | * from the passed in @q already in the plug list |
8e756373 | 1023 | * |
d38a9c04 JA |
1024 | * Determine whether @bio being queued on @q can be merged with the previous |
1025 | * request on %current's plugged list. Returns %true if merge was successful, | |
8e756373 BW |
1026 | * otherwise %false. |
1027 | * | |
1028 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1029 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1030 | * than scheduling, and the request, while may have elvpriv data, is not | |
1031 | * added on the elevator at this point. In addition, we don't have | |
1032 | * reliable access to the elevator outside queue lock. Only check basic | |
1033 | * merging parameters without querying the elevator. | |
1034 | * | |
1035 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1036 | */ | |
1037 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
0c5bcc92 | 1038 | unsigned int nr_segs) |
8e756373 BW |
1039 | { |
1040 | struct blk_plug *plug; | |
1041 | struct request *rq; | |
8e756373 BW |
1042 | |
1043 | plug = blk_mq_plug(q, bio); | |
bc490f81 | 1044 | if (!plug || rq_list_empty(plug->mq_list)) |
8e756373 BW |
1045 | return false; |
1046 | ||
5b205071 JA |
1047 | rq_list_for_each(&plug->mq_list, rq) { |
1048 | if (rq->q == q) { | |
1049 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == | |
1050 | BIO_MERGE_OK) | |
1051 | return true; | |
1052 | break; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Only keep iterating plug list for merges if we have multiple | |
1057 | * queues | |
1058 | */ | |
1059 | if (!plug->multiple_queues) | |
1060 | break; | |
8e756373 | 1061 | } |
8e756373 BW |
1062 | return false; |
1063 | } | |
bdc6a287 BW |
1064 | |
1065 | /* | |
1066 | * Iterate list of requests and see if we can merge this bio with any | |
1067 | * of them. | |
1068 | */ | |
1069 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1070 | struct bio *bio, unsigned int nr_segs) | |
1071 | { | |
1072 | struct request *rq; | |
1073 | int checked = 8; | |
1074 | ||
1075 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1076 | if (!checked--) |
1077 | break; | |
1078 | ||
7d7ca7c5 BW |
1079 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1080 | case BIO_MERGE_NONE: | |
bdc6a287 | 1081 | continue; |
7d7ca7c5 BW |
1082 | case BIO_MERGE_OK: |
1083 | return true; | |
1084 | case BIO_MERGE_FAILED: | |
1085 | return false; | |
bdc6a287 BW |
1086 | } |
1087 | ||
bdc6a287 BW |
1088 | } |
1089 | ||
1090 | return false; | |
1091 | } | |
1092 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); | |
eda5cc99 CH |
1093 | |
1094 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, | |
1095 | unsigned int nr_segs, struct request **merged_request) | |
1096 | { | |
1097 | struct request *rq; | |
1098 | ||
1099 | switch (elv_merge(q, &rq, bio)) { | |
1100 | case ELEVATOR_BACK_MERGE: | |
1101 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1102 | return false; | |
1103 | if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1104 | return false; | |
1105 | *merged_request = attempt_back_merge(q, rq); | |
1106 | if (!*merged_request) | |
1107 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
1108 | return true; | |
1109 | case ELEVATOR_FRONT_MERGE: | |
1110 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1111 | return false; | |
1112 | if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1113 | return false; | |
1114 | *merged_request = attempt_front_merge(q, rq); | |
1115 | if (!*merged_request) | |
1116 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
1117 | return true; | |
1118 | case ELEVATOR_DISCARD_MERGE: | |
1119 | return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; | |
1120 | default: | |
1121 | return false; | |
1122 | } | |
1123 | } | |
1124 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |