]>
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> | |
9 | #include <linux/scatterlist.h> | |
10 | ||
cda22646 MK |
11 | #include <trace/events/block.h> |
12 | ||
d6d48196 JA |
13 | #include "blk.h" |
14 | ||
e9907009 CH |
15 | /* |
16 | * Check if the two bvecs from two bios can be merged to one segment. If yes, | |
17 | * no need to check gap between the two bios since the 1st bio and the 1st bvec | |
18 | * in the 2nd bio can be handled in one segment. | |
19 | */ | |
20 | static inline bool bios_segs_mergeable(struct request_queue *q, | |
21 | struct bio *prev, struct bio_vec *prev_last_bv, | |
22 | struct bio_vec *next_first_bv) | |
23 | { | |
3dccdae5 | 24 | if (!biovec_phys_mergeable(q, prev_last_bv, next_first_bv)) |
e9907009 CH |
25 | return false; |
26 | if (prev->bi_seg_back_size + next_first_bv->bv_len > | |
27 | queue_max_segment_size(q)) | |
28 | return false; | |
29 | return true; | |
30 | } | |
31 | ||
32 | static inline bool bio_will_gap(struct request_queue *q, | |
33 | struct request *prev_rq, struct bio *prev, struct bio *next) | |
34 | { | |
35 | struct bio_vec pb, nb; | |
36 | ||
37 | if (!bio_has_data(prev) || !queue_virt_boundary(q)) | |
38 | return false; | |
39 | ||
40 | /* | |
41 | * Don't merge if the 1st bio starts with non-zero offset, otherwise it | |
42 | * is quite difficult to respect the sg gap limit. We work hard to | |
43 | * merge a huge number of small single bios in case of mkfs. | |
44 | */ | |
45 | if (prev_rq) | |
46 | bio_get_first_bvec(prev_rq->bio, &pb); | |
47 | else | |
48 | bio_get_first_bvec(prev, &pb); | |
df376b2e | 49 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
50 | return true; |
51 | ||
52 | /* | |
53 | * We don't need to worry about the situation that the merged segment | |
54 | * ends in unaligned virt boundary: | |
55 | * | |
56 | * - if 'pb' ends aligned, the merged segment ends aligned | |
57 | * - if 'pb' ends unaligned, the next bio must include | |
58 | * one single bvec of 'nb', otherwise the 'nb' can't | |
59 | * merge with 'pb' | |
60 | */ | |
61 | bio_get_last_bvec(prev, &pb); | |
62 | bio_get_first_bvec(next, &nb); | |
63 | if (bios_segs_mergeable(q, prev, &pb, &nb)) | |
64 | return false; | |
65 | return __bvec_gap_to_prev(q, &pb, nb.bv_offset); | |
66 | } | |
67 | ||
68 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) | |
69 | { | |
70 | return bio_will_gap(req->q, req, req->biotail, bio); | |
71 | } | |
72 | ||
73 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) | |
74 | { | |
75 | return bio_will_gap(req->q, NULL, bio, req->bio); | |
76 | } | |
77 | ||
54efd50b KO |
78 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
79 | struct bio *bio, | |
bdced438 ML |
80 | struct bio_set *bs, |
81 | unsigned *nsegs) | |
54efd50b KO |
82 | { |
83 | unsigned int max_discard_sectors, granularity; | |
84 | int alignment; | |
85 | sector_t tmp; | |
86 | unsigned split_sectors; | |
87 | ||
bdced438 ML |
88 | *nsegs = 1; |
89 | ||
54efd50b KO |
90 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
91 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
92 | ||
1adfc5e4 ML |
93 | max_discard_sectors = min(q->limits.max_discard_sectors, |
94 | bio_allowed_max_sectors(q)); | |
54efd50b KO |
95 | max_discard_sectors -= max_discard_sectors % granularity; |
96 | ||
97 | if (unlikely(!max_discard_sectors)) { | |
98 | /* XXX: warn */ | |
99 | return NULL; | |
100 | } | |
101 | ||
102 | if (bio_sectors(bio) <= max_discard_sectors) | |
103 | return NULL; | |
104 | ||
105 | split_sectors = max_discard_sectors; | |
106 | ||
107 | /* | |
108 | * If the next starting sector would be misaligned, stop the discard at | |
109 | * the previous aligned sector. | |
110 | */ | |
111 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
112 | ||
113 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
114 | tmp = sector_div(tmp, granularity); | |
115 | ||
116 | if (split_sectors > tmp) | |
117 | split_sectors -= tmp; | |
118 | ||
119 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
120 | } | |
121 | ||
885fa13f CH |
122 | static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, |
123 | struct bio *bio, struct bio_set *bs, unsigned *nsegs) | |
124 | { | |
125 | *nsegs = 1; | |
126 | ||
127 | if (!q->limits.max_write_zeroes_sectors) | |
128 | return NULL; | |
129 | ||
130 | if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) | |
131 | return NULL; | |
132 | ||
133 | return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); | |
134 | } | |
135 | ||
54efd50b KO |
136 | static struct bio *blk_bio_write_same_split(struct request_queue *q, |
137 | struct bio *bio, | |
bdced438 ML |
138 | struct bio_set *bs, |
139 | unsigned *nsegs) | |
54efd50b | 140 | { |
bdced438 ML |
141 | *nsegs = 1; |
142 | ||
54efd50b KO |
143 | if (!q->limits.max_write_same_sectors) |
144 | return NULL; | |
145 | ||
146 | if (bio_sectors(bio) <= q->limits.max_write_same_sectors) | |
147 | return NULL; | |
148 | ||
149 | return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs); | |
150 | } | |
151 | ||
d0e5fbb0 ML |
152 | static inline unsigned get_max_io_size(struct request_queue *q, |
153 | struct bio *bio) | |
154 | { | |
155 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); | |
156 | unsigned mask = queue_logical_block_size(q) - 1; | |
157 | ||
158 | /* aligned to logical block size */ | |
159 | sectors &= ~(mask >> 9); | |
160 | ||
161 | return sectors; | |
162 | } | |
163 | ||
54efd50b KO |
164 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
165 | struct bio *bio, | |
bdced438 ML |
166 | struct bio_set *bs, |
167 | unsigned *segs) | |
54efd50b | 168 | { |
5014c311 | 169 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 170 | struct bvec_iter iter; |
8ae12666 | 171 | unsigned seg_size = 0, nsegs = 0, sectors = 0; |
02e70742 ML |
172 | unsigned front_seg_size = bio->bi_seg_front_size; |
173 | bool do_split = true; | |
174 | struct bio *new = NULL; | |
d0e5fbb0 | 175 | const unsigned max_sectors = get_max_io_size(q, bio); |
54efd50b | 176 | |
54efd50b | 177 | bio_for_each_segment(bv, bio, iter) { |
54efd50b KO |
178 | /* |
179 | * If the queue doesn't support SG gaps and adding this | |
180 | * offset would create a gap, disallow it. | |
181 | */ | |
5014c311 | 182 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
183 | goto split; |
184 | ||
d0e5fbb0 | 185 | if (sectors + (bv.bv_len >> 9) > max_sectors) { |
e36f6204 KB |
186 | /* |
187 | * Consider this a new segment if we're splitting in | |
188 | * the middle of this vector. | |
189 | */ | |
190 | if (nsegs < queue_max_segments(q) && | |
d0e5fbb0 | 191 | sectors < max_sectors) { |
e36f6204 | 192 | nsegs++; |
d0e5fbb0 | 193 | sectors = max_sectors; |
e36f6204 | 194 | } |
cf8c0c6a | 195 | goto split; |
e36f6204 KB |
196 | } |
197 | ||
38417468 | 198 | if (bvprvp) { |
b4b6cb61 ML |
199 | if (seg_size + bv.bv_len > queue_max_segment_size(q)) |
200 | goto new_segment; | |
3dccdae5 | 201 | if (!biovec_phys_mergeable(q, bvprvp, &bv)) |
54efd50b KO |
202 | goto new_segment; |
203 | ||
204 | seg_size += bv.bv_len; | |
205 | bvprv = bv; | |
578270bf | 206 | bvprvp = &bvprv; |
52cc6eea | 207 | sectors += bv.bv_len >> 9; |
a88d32af | 208 | |
54efd50b KO |
209 | continue; |
210 | } | |
211 | new_segment: | |
212 | if (nsegs == queue_max_segments(q)) | |
213 | goto split; | |
214 | ||
6a501bf0 ML |
215 | if (nsegs == 1 && seg_size > front_seg_size) |
216 | front_seg_size = seg_size; | |
217 | ||
54efd50b KO |
218 | nsegs++; |
219 | bvprv = bv; | |
578270bf | 220 | bvprvp = &bvprv; |
54efd50b | 221 | seg_size = bv.bv_len; |
52cc6eea | 222 | sectors += bv.bv_len >> 9; |
02e70742 | 223 | |
54efd50b KO |
224 | } |
225 | ||
02e70742 | 226 | do_split = false; |
54efd50b | 227 | split: |
bdced438 | 228 | *segs = nsegs; |
02e70742 ML |
229 | |
230 | if (do_split) { | |
231 | new = bio_split(bio, sectors, GFP_NOIO, bs); | |
232 | if (new) | |
233 | bio = new; | |
234 | } | |
235 | ||
6a501bf0 ML |
236 | if (nsegs == 1 && seg_size > front_seg_size) |
237 | front_seg_size = seg_size; | |
02e70742 ML |
238 | bio->bi_seg_front_size = front_seg_size; |
239 | if (seg_size > bio->bi_seg_back_size) | |
240 | bio->bi_seg_back_size = seg_size; | |
241 | ||
242 | return do_split ? new : NULL; | |
54efd50b KO |
243 | } |
244 | ||
af67c31f | 245 | void blk_queue_split(struct request_queue *q, struct bio **bio) |
54efd50b | 246 | { |
bdced438 ML |
247 | struct bio *split, *res; |
248 | unsigned nsegs; | |
54efd50b | 249 | |
7afafc8a AH |
250 | switch (bio_op(*bio)) { |
251 | case REQ_OP_DISCARD: | |
252 | case REQ_OP_SECURE_ERASE: | |
338aa96d | 253 | split = blk_bio_discard_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a | 254 | break; |
a6f0788e | 255 | case REQ_OP_WRITE_ZEROES: |
338aa96d | 256 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, &nsegs); |
a6f0788e | 257 | break; |
7afafc8a | 258 | case REQ_OP_WRITE_SAME: |
338aa96d | 259 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
260 | break; |
261 | default: | |
338aa96d | 262 | split = blk_bio_segment_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
263 | break; |
264 | } | |
bdced438 ML |
265 | |
266 | /* physical segments can be figured out during splitting */ | |
267 | res = split ? split : *bio; | |
268 | res->bi_phys_segments = nsegs; | |
269 | bio_set_flag(res, BIO_SEG_VALID); | |
54efd50b KO |
270 | |
271 | if (split) { | |
6ac45aeb | 272 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 273 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 274 | |
947b7ac1 JA |
275 | /* |
276 | * Since we're recursing into make_request here, ensure | |
277 | * that we mark this bio as already having entered the queue. | |
278 | * If not, and the queue is going away, we can get stuck | |
279 | * forever on waiting for the queue reference to drop. But | |
280 | * that will never happen, as we're already holding a | |
281 | * reference to it. | |
282 | */ | |
283 | bio_set_flag(*bio, BIO_QUEUE_ENTERED); | |
284 | ||
54efd50b | 285 | bio_chain(split, *bio); |
cda22646 | 286 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
54efd50b KO |
287 | generic_make_request(*bio); |
288 | *bio = split; | |
289 | } | |
290 | } | |
291 | EXPORT_SYMBOL(blk_queue_split); | |
292 | ||
1e428079 | 293 | static unsigned int __blk_recalc_rq_segments(struct request_queue *q, |
07388549 ML |
294 | struct bio *bio, |
295 | bool no_sg_merge) | |
d6d48196 | 296 | { |
7988613b | 297 | struct bio_vec bv, bvprv = { NULL }; |
38417468 | 298 | int prev = 0; |
1e428079 | 299 | unsigned int seg_size, nr_phys_segs; |
59247eae | 300 | struct bio *fbio, *bbio; |
7988613b | 301 | struct bvec_iter iter; |
d6d48196 | 302 | |
1e428079 JA |
303 | if (!bio) |
304 | return 0; | |
d6d48196 | 305 | |
a6f0788e CK |
306 | switch (bio_op(bio)) { |
307 | case REQ_OP_DISCARD: | |
308 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 309 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
310 | return 0; |
311 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 312 | return 1; |
a6f0788e | 313 | } |
5cb8850c | 314 | |
1e428079 | 315 | fbio = bio; |
5df97b91 | 316 | seg_size = 0; |
2c8919de | 317 | nr_phys_segs = 0; |
1e428079 | 318 | for_each_bio(bio) { |
7988613b | 319 | bio_for_each_segment(bv, bio, iter) { |
05f1dd53 JA |
320 | /* |
321 | * If SG merging is disabled, each bio vector is | |
322 | * a segment | |
323 | */ | |
324 | if (no_sg_merge) | |
325 | goto new_segment; | |
326 | ||
38417468 | 327 | if (prev) { |
7988613b | 328 | if (seg_size + bv.bv_len |
ae03bf63 | 329 | > queue_max_segment_size(q)) |
1e428079 | 330 | goto new_segment; |
3dccdae5 | 331 | if (!biovec_phys_mergeable(q, &bvprv, &bv)) |
1e428079 | 332 | goto new_segment; |
d6d48196 | 333 | |
7988613b | 334 | seg_size += bv.bv_len; |
1e428079 JA |
335 | bvprv = bv; |
336 | continue; | |
337 | } | |
d6d48196 | 338 | new_segment: |
1e428079 JA |
339 | if (nr_phys_segs == 1 && seg_size > |
340 | fbio->bi_seg_front_size) | |
341 | fbio->bi_seg_front_size = seg_size; | |
86771427 | 342 | |
1e428079 JA |
343 | nr_phys_segs++; |
344 | bvprv = bv; | |
54efd50b | 345 | prev = 1; |
7988613b | 346 | seg_size = bv.bv_len; |
1e428079 | 347 | } |
59247eae | 348 | bbio = bio; |
d6d48196 JA |
349 | } |
350 | ||
59247eae JA |
351 | if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size) |
352 | fbio->bi_seg_front_size = seg_size; | |
353 | if (seg_size > bbio->bi_seg_back_size) | |
354 | bbio->bi_seg_back_size = seg_size; | |
1e428079 JA |
355 | |
356 | return nr_phys_segs; | |
357 | } | |
358 | ||
359 | void blk_recalc_rq_segments(struct request *rq) | |
360 | { | |
07388549 ML |
361 | bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE, |
362 | &rq->q->queue_flags); | |
363 | ||
364 | rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio, | |
365 | no_sg_merge); | |
d6d48196 JA |
366 | } |
367 | ||
368 | void blk_recount_segments(struct request_queue *q, struct bio *bio) | |
369 | { | |
7f60dcaa ML |
370 | unsigned short seg_cnt; |
371 | ||
372 | /* estimate segment number by bi_vcnt for non-cloned bio */ | |
373 | if (bio_flagged(bio, BIO_CLONED)) | |
374 | seg_cnt = bio_segments(bio); | |
375 | else | |
376 | seg_cnt = bio->bi_vcnt; | |
764f612c | 377 | |
7f60dcaa ML |
378 | if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) && |
379 | (seg_cnt < queue_max_segments(q))) | |
380 | bio->bi_phys_segments = seg_cnt; | |
05f1dd53 JA |
381 | else { |
382 | struct bio *nxt = bio->bi_next; | |
383 | ||
384 | bio->bi_next = NULL; | |
7f60dcaa | 385 | bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false); |
05f1dd53 JA |
386 | bio->bi_next = nxt; |
387 | } | |
1e428079 | 388 | |
b7c44ed9 | 389 | bio_set_flag(bio, BIO_SEG_VALID); |
d6d48196 | 390 | } |
d6d48196 JA |
391 | |
392 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, | |
393 | struct bio *nxt) | |
394 | { | |
2b8221e1 | 395 | struct bio_vec end_bv = { NULL }, nxt_bv; |
f619d254 | 396 | |
86771427 | 397 | if (bio->bi_seg_back_size + nxt->bi_seg_front_size > |
ae03bf63 | 398 | queue_max_segment_size(q)) |
d6d48196 JA |
399 | return 0; |
400 | ||
e17fc0a1 DW |
401 | if (!bio_has_data(bio)) |
402 | return 1; | |
403 | ||
e827091c ML |
404 | bio_get_last_bvec(bio, &end_bv); |
405 | bio_get_first_bvec(nxt, &nxt_bv); | |
f619d254 | 406 | |
3dccdae5 | 407 | return biovec_phys_mergeable(q, &end_bv, &nxt_bv); |
d6d48196 JA |
408 | } |
409 | ||
7988613b | 410 | static inline void |
963ab9e5 | 411 | __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec, |
7988613b | 412 | struct scatterlist *sglist, struct bio_vec *bvprv, |
38417468 | 413 | struct scatterlist **sg, int *nsegs) |
963ab9e5 AH |
414 | { |
415 | ||
416 | int nbytes = bvec->bv_len; | |
417 | ||
38417468 | 418 | if (*sg) { |
b4b6cb61 ML |
419 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
420 | goto new_segment; | |
3dccdae5 | 421 | if (!biovec_phys_mergeable(q, bvprv, bvec)) |
963ab9e5 AH |
422 | goto new_segment; |
423 | ||
424 | (*sg)->length += nbytes; | |
425 | } else { | |
426 | new_segment: | |
427 | if (!*sg) | |
428 | *sg = sglist; | |
429 | else { | |
430 | /* | |
431 | * If the driver previously mapped a shorter | |
432 | * list, we could see a termination bit | |
433 | * prematurely unless it fully inits the sg | |
434 | * table on each mapping. We KNOW that there | |
435 | * must be more entries here or the driver | |
436 | * would be buggy, so force clear the | |
437 | * termination bit to avoid doing a full | |
438 | * sg_init_table() in drivers for each command. | |
439 | */ | |
c8164d89 | 440 | sg_unmark_end(*sg); |
963ab9e5 AH |
441 | *sg = sg_next(*sg); |
442 | } | |
443 | ||
444 | sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset); | |
445 | (*nsegs)++; | |
446 | } | |
7988613b | 447 | *bvprv = *bvec; |
963ab9e5 AH |
448 | } |
449 | ||
f9d03f96 CH |
450 | static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv, |
451 | struct scatterlist *sglist, struct scatterlist **sg) | |
452 | { | |
453 | *sg = sglist; | |
454 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
455 | return 1; | |
456 | } | |
457 | ||
5cb8850c KO |
458 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
459 | struct scatterlist *sglist, | |
460 | struct scatterlist **sg) | |
d6d48196 | 461 | { |
2b8221e1 | 462 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 463 | struct bvec_iter iter; |
38417468 | 464 | int nsegs = 0; |
5cb8850c KO |
465 | |
466 | for_each_bio(bio) | |
467 | bio_for_each_segment(bvec, bio, iter) | |
468 | __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg, | |
38417468 | 469 | &nsegs); |
d6d48196 | 470 | |
5cb8850c KO |
471 | return nsegs; |
472 | } | |
473 | ||
474 | /* | |
475 | * map a request to scatterlist, return number of sg entries setup. Caller | |
476 | * must make sure sg can hold rq->nr_phys_segments entries | |
477 | */ | |
478 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, | |
479 | struct scatterlist *sglist) | |
480 | { | |
481 | struct scatterlist *sg = NULL; | |
482 | int nsegs = 0; | |
483 | ||
f9d03f96 CH |
484 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
485 | nsegs = __blk_bvec_map_sg(q, rq->special_vec, sglist, &sg); | |
486 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) | |
487 | nsegs = __blk_bvec_map_sg(q, bio_iovec(rq->bio), sglist, &sg); | |
488 | else if (rq->bio) | |
5cb8850c | 489 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg); |
f18573ab | 490 | |
e8064021 | 491 | if (unlikely(rq->rq_flags & RQF_COPY_USER) && |
2e46e8b2 TH |
492 | (blk_rq_bytes(rq) & q->dma_pad_mask)) { |
493 | unsigned int pad_len = | |
494 | (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1; | |
f18573ab FT |
495 | |
496 | sg->length += pad_len; | |
497 | rq->extra_len += pad_len; | |
498 | } | |
499 | ||
2fb98e84 | 500 | if (q->dma_drain_size && q->dma_drain_needed(rq)) { |
a8ebb056 | 501 | if (op_is_write(req_op(rq))) |
db0a2e00 TH |
502 | memset(q->dma_drain_buffer, 0, q->dma_drain_size); |
503 | ||
da81ed16 | 504 | sg_unmark_end(sg); |
d6d48196 JA |
505 | sg = sg_next(sg); |
506 | sg_set_page(sg, virt_to_page(q->dma_drain_buffer), | |
507 | q->dma_drain_size, | |
508 | ((unsigned long)q->dma_drain_buffer) & | |
509 | (PAGE_SIZE - 1)); | |
510 | nsegs++; | |
7a85f889 | 511 | rq->extra_len += q->dma_drain_size; |
d6d48196 JA |
512 | } |
513 | ||
514 | if (sg) | |
515 | sg_mark_end(sg); | |
516 | ||
12e57f59 ML |
517 | /* |
518 | * Something must have been wrong if the figured number of | |
519 | * segment is bigger than number of req's physical segments | |
520 | */ | |
f9d03f96 | 521 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 522 | |
d6d48196 JA |
523 | return nsegs; |
524 | } | |
d6d48196 JA |
525 | EXPORT_SYMBOL(blk_rq_map_sg); |
526 | ||
d6d48196 JA |
527 | static inline int ll_new_hw_segment(struct request_queue *q, |
528 | struct request *req, | |
529 | struct bio *bio) | |
530 | { | |
d6d48196 JA |
531 | int nr_phys_segs = bio_phys_segments(q, bio); |
532 | ||
13f05c8d MP |
533 | if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q)) |
534 | goto no_merge; | |
535 | ||
4eaf99be | 536 | if (blk_integrity_merge_bio(q, req, bio) == false) |
13f05c8d | 537 | goto no_merge; |
d6d48196 JA |
538 | |
539 | /* | |
540 | * This will form the start of a new hw segment. Bump both | |
541 | * counters. | |
542 | */ | |
d6d48196 JA |
543 | req->nr_phys_segments += nr_phys_segs; |
544 | return 1; | |
13f05c8d MP |
545 | |
546 | no_merge: | |
e0c72300 | 547 | req_set_nomerge(q, req); |
13f05c8d | 548 | return 0; |
d6d48196 JA |
549 | } |
550 | ||
551 | int ll_back_merge_fn(struct request_queue *q, struct request *req, | |
552 | struct bio *bio) | |
553 | { | |
5e7c4274 JA |
554 | if (req_gap_back_merge(req, bio)) |
555 | return 0; | |
7f39add3 SG |
556 | if (blk_integrity_rq(req) && |
557 | integrity_req_gap_back_merge(req, bio)) | |
558 | return 0; | |
f31dc1cd | 559 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 560 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
e0c72300 | 561 | req_set_nomerge(q, req); |
d6d48196 JA |
562 | return 0; |
563 | } | |
2cdf79ca | 564 | if (!bio_flagged(req->biotail, BIO_SEG_VALID)) |
d6d48196 | 565 | blk_recount_segments(q, req->biotail); |
2cdf79ca | 566 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 567 | blk_recount_segments(q, bio); |
d6d48196 JA |
568 | |
569 | return ll_new_hw_segment(q, req, bio); | |
570 | } | |
571 | ||
6728cb0e | 572 | int ll_front_merge_fn(struct request_queue *q, struct request *req, |
d6d48196 JA |
573 | struct bio *bio) |
574 | { | |
5e7c4274 JA |
575 | |
576 | if (req_gap_front_merge(req, bio)) | |
577 | return 0; | |
7f39add3 SG |
578 | if (blk_integrity_rq(req) && |
579 | integrity_req_gap_front_merge(req, bio)) | |
580 | return 0; | |
f31dc1cd | 581 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 582 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
e0c72300 | 583 | req_set_nomerge(q, req); |
d6d48196 JA |
584 | return 0; |
585 | } | |
2cdf79ca | 586 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 587 | blk_recount_segments(q, bio); |
2cdf79ca | 588 | if (!bio_flagged(req->bio, BIO_SEG_VALID)) |
d6d48196 | 589 | blk_recount_segments(q, req->bio); |
d6d48196 JA |
590 | |
591 | return ll_new_hw_segment(q, req, bio); | |
592 | } | |
593 | ||
445251d0 JA |
594 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
595 | struct request *next) | |
596 | { | |
597 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
598 | ||
599 | if (segments >= queue_max_discard_segments(q)) | |
600 | goto no_merge; | |
601 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
602 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
603 | goto no_merge; | |
604 | ||
605 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
606 | return true; | |
607 | no_merge: | |
608 | req_set_nomerge(q, req); | |
609 | return false; | |
610 | } | |
611 | ||
d6d48196 JA |
612 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
613 | struct request *next) | |
614 | { | |
615 | int total_phys_segments; | |
86771427 FT |
616 | unsigned int seg_size = |
617 | req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size; | |
d6d48196 | 618 | |
5e7c4274 | 619 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
620 | return 0; |
621 | ||
d6d48196 JA |
622 | /* |
623 | * Will it become too large? | |
624 | */ | |
f31dc1cd | 625 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 626 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
627 | return 0; |
628 | ||
629 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
86771427 FT |
630 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) { |
631 | if (req->nr_phys_segments == 1) | |
632 | req->bio->bi_seg_front_size = seg_size; | |
633 | if (next->nr_phys_segments == 1) | |
634 | next->biotail->bi_seg_back_size = seg_size; | |
d6d48196 | 635 | total_phys_segments--; |
86771427 | 636 | } |
d6d48196 | 637 | |
8a78362c | 638 | if (total_phys_segments > queue_max_segments(q)) |
d6d48196 JA |
639 | return 0; |
640 | ||
4eaf99be | 641 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
642 | return 0; |
643 | ||
d6d48196 JA |
644 | /* Merge is OK... */ |
645 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
646 | return 1; |
647 | } | |
648 | ||
80a761fd TH |
649 | /** |
650 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
651 | * @rq: request to mark as mixed merge | |
652 | * | |
653 | * Description: | |
654 | * @rq is about to be mixed merged. Make sure the attributes | |
655 | * which can be mixed are set in each bio and mark @rq as mixed | |
656 | * merged. | |
657 | */ | |
658 | void blk_rq_set_mixed_merge(struct request *rq) | |
659 | { | |
660 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
661 | struct bio *bio; | |
662 | ||
e8064021 | 663 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
664 | return; |
665 | ||
666 | /* | |
667 | * @rq will no longer represent mixable attributes for all the | |
668 | * contained bios. It will just track those of the first one. | |
669 | * Distributes the attributs to each bio. | |
670 | */ | |
671 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
672 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
673 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
674 | bio->bi_opf |= ff; | |
80a761fd | 675 | } |
e8064021 | 676 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
677 | } |
678 | ||
26308eab JM |
679 | static void blk_account_io_merge(struct request *req) |
680 | { | |
681 | if (blk_do_io_stat(req)) { | |
682 | struct hd_struct *part; | |
26308eab | 683 | |
112f158f | 684 | part_stat_lock(); |
09e099d4 | 685 | part = req->part; |
26308eab | 686 | |
d62e26b3 | 687 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
26308eab | 688 | |
6c23a968 | 689 | hd_struct_put(part); |
26308eab JM |
690 | part_stat_unlock(); |
691 | } | |
692 | } | |
69840466 JW |
693 | /* |
694 | * Two cases of handling DISCARD merge: | |
695 | * If max_discard_segments > 1, the driver takes every bio | |
696 | * as a range and send them to controller together. The ranges | |
697 | * needn't to be contiguous. | |
698 | * Otherwise, the bios/requests will be handled as same as | |
699 | * others which should be contiguous. | |
700 | */ | |
701 | static inline bool blk_discard_mergable(struct request *req) | |
702 | { | |
703 | if (req_op(req) == REQ_OP_DISCARD && | |
704 | queue_max_discard_segments(req->q) > 1) | |
705 | return true; | |
706 | return false; | |
707 | } | |
708 | ||
e96c0d83 EB |
709 | static enum elv_merge blk_try_req_merge(struct request *req, |
710 | struct request *next) | |
69840466 JW |
711 | { |
712 | if (blk_discard_mergable(req)) | |
713 | return ELEVATOR_DISCARD_MERGE; | |
714 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
715 | return ELEVATOR_BACK_MERGE; | |
716 | ||
717 | return ELEVATOR_NO_MERGE; | |
718 | } | |
26308eab | 719 | |
d6d48196 | 720 | /* |
b973cb7e JA |
721 | * For non-mq, this has to be called with the request spinlock acquired. |
722 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 723 | */ |
b973cb7e JA |
724 | static struct request *attempt_merge(struct request_queue *q, |
725 | struct request *req, struct request *next) | |
d6d48196 JA |
726 | { |
727 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 728 | return NULL; |
d6d48196 | 729 | |
288dab8a | 730 | if (req_op(req) != req_op(next)) |
b973cb7e | 731 | return NULL; |
f31dc1cd | 732 | |
d6d48196 | 733 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 734 | || req->rq_disk != next->rq_disk) |
b973cb7e | 735 | return NULL; |
d6d48196 | 736 | |
8fe0d473 | 737 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 738 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 739 | return NULL; |
4363ac7c | 740 | |
cb6934f8 JA |
741 | /* |
742 | * Don't allow merge of different write hints, or for a hint with | |
743 | * non-hint IO. | |
744 | */ | |
745 | if (req->write_hint != next->write_hint) | |
746 | return NULL; | |
747 | ||
668ffc03 DLM |
748 | if (req->ioprio != next->ioprio) |
749 | return NULL; | |
750 | ||
d6d48196 JA |
751 | /* |
752 | * If we are allowed to merge, then append bio list | |
753 | * from next to rq and release next. merge_requests_fn | |
754 | * will have updated segment counts, update sector | |
445251d0 JA |
755 | * counts here. Handle DISCARDs separately, as they |
756 | * have separate settings. | |
d6d48196 | 757 | */ |
69840466 JW |
758 | |
759 | switch (blk_try_req_merge(req, next)) { | |
760 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
761 | if (!req_attempt_discard_merge(q, req, next)) |
762 | return NULL; | |
69840466 JW |
763 | break; |
764 | case ELEVATOR_BACK_MERGE: | |
765 | if (!ll_merge_requests_fn(q, req, next)) | |
766 | return NULL; | |
767 | break; | |
768 | default: | |
b973cb7e | 769 | return NULL; |
69840466 | 770 | } |
d6d48196 | 771 | |
80a761fd TH |
772 | /* |
773 | * If failfast settings disagree or any of the two is already | |
774 | * a mixed merge, mark both as mixed before proceeding. This | |
775 | * makes sure that all involved bios have mixable attributes | |
776 | * set properly. | |
777 | */ | |
e8064021 | 778 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
779 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
780 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
781 | blk_rq_set_mixed_merge(req); | |
782 | blk_rq_set_mixed_merge(next); | |
783 | } | |
784 | ||
d6d48196 | 785 | /* |
522a7775 OS |
786 | * At this point we have either done a back merge or front merge. We |
787 | * need the smaller start_time_ns of the merged requests to be the | |
788 | * current request for accounting purposes. | |
d6d48196 | 789 | */ |
522a7775 OS |
790 | if (next->start_time_ns < req->start_time_ns) |
791 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
792 | |
793 | req->biotail->bi_next = next->bio; | |
794 | req->biotail = next->biotail; | |
795 | ||
a2dec7b3 | 796 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 797 | |
2a5cf35c | 798 | if (!blk_discard_mergable(req)) |
445251d0 | 799 | elv_merge_requests(q, req, next); |
d6d48196 | 800 | |
42dad764 JM |
801 | /* |
802 | * 'next' is going away, so update stats accordingly | |
803 | */ | |
804 | blk_account_io_merge(next); | |
d6d48196 | 805 | |
e4d750c9 JA |
806 | /* |
807 | * ownership of bio passed from next to req, return 'next' for | |
808 | * the caller to free | |
809 | */ | |
1cd96c24 | 810 | next->bio = NULL; |
b973cb7e | 811 | return next; |
d6d48196 JA |
812 | } |
813 | ||
b973cb7e | 814 | struct request *attempt_back_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
815 | { |
816 | struct request *next = elv_latter_request(q, rq); | |
817 | ||
818 | if (next) | |
819 | return attempt_merge(q, rq, next); | |
820 | ||
b973cb7e | 821 | return NULL; |
d6d48196 JA |
822 | } |
823 | ||
b973cb7e | 824 | struct request *attempt_front_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
825 | { |
826 | struct request *prev = elv_former_request(q, rq); | |
827 | ||
828 | if (prev) | |
829 | return attempt_merge(q, prev, rq); | |
830 | ||
b973cb7e | 831 | return NULL; |
d6d48196 | 832 | } |
5e84ea3a JA |
833 | |
834 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
835 | struct request *next) | |
836 | { | |
e4d750c9 | 837 | struct request *free; |
72ef799b | 838 | |
e4d750c9 JA |
839 | free = attempt_merge(q, rq, next); |
840 | if (free) { | |
92bc5a24 | 841 | blk_put_request(free); |
e4d750c9 JA |
842 | return 1; |
843 | } | |
844 | ||
845 | return 0; | |
5e84ea3a | 846 | } |
050c8ea8 TH |
847 | |
848 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
849 | { | |
e2a60da7 | 850 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
851 | return false; |
852 | ||
288dab8a | 853 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
854 | return false; |
855 | ||
050c8ea8 TH |
856 | /* different data direction or already started, don't merge */ |
857 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
858 | return false; | |
859 | ||
2081a56b JA |
860 | /* must be same device */ |
861 | if (rq->rq_disk != bio->bi_disk) | |
050c8ea8 TH |
862 | return false; |
863 | ||
864 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 865 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
866 | return false; |
867 | ||
4363ac7c | 868 | /* must be using the same buffer */ |
8fe0d473 | 869 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
870 | !blk_write_same_mergeable(rq->bio, bio)) |
871 | return false; | |
872 | ||
cb6934f8 JA |
873 | /* |
874 | * Don't allow merge of different write hints, or for a hint with | |
875 | * non-hint IO. | |
876 | */ | |
877 | if (rq->write_hint != bio->bi_write_hint) | |
878 | return false; | |
879 | ||
668ffc03 DLM |
880 | if (rq->ioprio != bio_prio(bio)) |
881 | return false; | |
882 | ||
050c8ea8 TH |
883 | return true; |
884 | } | |
885 | ||
34fe7c05 | 886 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 887 | { |
69840466 | 888 | if (blk_discard_mergable(rq)) |
1e739730 CH |
889 | return ELEVATOR_DISCARD_MERGE; |
890 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 891 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 892 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
893 | return ELEVATOR_FRONT_MERGE; |
894 | return ELEVATOR_NO_MERGE; | |
895 | } |