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 | ||
dcebd755 ML |
164 | static unsigned get_max_segment_size(struct request_queue *q, |
165 | unsigned offset) | |
166 | { | |
167 | unsigned long mask = queue_segment_boundary(q); | |
168 | ||
169 | /* default segment boundary mask means no boundary limit */ | |
170 | if (mask == BLK_SEG_BOUNDARY_MASK) | |
171 | return queue_max_segment_size(q); | |
172 | ||
173 | return min_t(unsigned long, mask - (mask & offset) + 1, | |
174 | queue_max_segment_size(q)); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Split the bvec @bv into segments, and update all kinds of | |
179 | * variables. | |
180 | */ | |
181 | static bool bvec_split_segs(struct request_queue *q, struct bio_vec *bv, | |
182 | unsigned *nsegs, unsigned *last_seg_size, | |
183 | unsigned *front_seg_size, unsigned *sectors) | |
184 | { | |
185 | unsigned len = bv->bv_len; | |
186 | unsigned total_len = 0; | |
187 | unsigned new_nsegs = 0, seg_size = 0; | |
188 | ||
189 | /* | |
190 | * Multi-page bvec may be too big to hold in one segment, so the | |
191 | * current bvec has to be splitted as multiple segments. | |
192 | */ | |
193 | while (len && new_nsegs + *nsegs < queue_max_segments(q)) { | |
194 | seg_size = get_max_segment_size(q, bv->bv_offset + total_len); | |
195 | seg_size = min(seg_size, len); | |
196 | ||
197 | new_nsegs++; | |
198 | total_len += seg_size; | |
199 | len -= seg_size; | |
200 | ||
201 | if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) | |
202 | break; | |
203 | } | |
204 | ||
205 | if (!new_nsegs) | |
206 | return !!len; | |
207 | ||
208 | /* update front segment size */ | |
209 | if (!*nsegs) { | |
210 | unsigned first_seg_size; | |
211 | ||
212 | if (new_nsegs == 1) | |
213 | first_seg_size = get_max_segment_size(q, bv->bv_offset); | |
214 | else | |
215 | first_seg_size = queue_max_segment_size(q); | |
216 | ||
217 | if (*front_seg_size < first_seg_size) | |
218 | *front_seg_size = first_seg_size; | |
219 | } | |
220 | ||
221 | /* update other varibles */ | |
222 | *last_seg_size = seg_size; | |
223 | *nsegs += new_nsegs; | |
224 | if (sectors) | |
225 | *sectors += total_len >> 9; | |
226 | ||
227 | /* split in the middle of the bvec if len != 0 */ | |
228 | return !!len; | |
229 | } | |
230 | ||
54efd50b KO |
231 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
232 | struct bio *bio, | |
bdced438 ML |
233 | struct bio_set *bs, |
234 | unsigned *segs) | |
54efd50b | 235 | { |
5014c311 | 236 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 237 | struct bvec_iter iter; |
8ae12666 | 238 | unsigned seg_size = 0, nsegs = 0, sectors = 0; |
02e70742 ML |
239 | unsigned front_seg_size = bio->bi_seg_front_size; |
240 | bool do_split = true; | |
241 | struct bio *new = NULL; | |
d0e5fbb0 | 242 | const unsigned max_sectors = get_max_io_size(q, bio); |
54efd50b | 243 | |
dcebd755 | 244 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
245 | /* |
246 | * If the queue doesn't support SG gaps and adding this | |
247 | * offset would create a gap, disallow it. | |
248 | */ | |
5014c311 | 249 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
250 | goto split; |
251 | ||
d0e5fbb0 | 252 | if (sectors + (bv.bv_len >> 9) > max_sectors) { |
e36f6204 KB |
253 | /* |
254 | * Consider this a new segment if we're splitting in | |
255 | * the middle of this vector. | |
256 | */ | |
257 | if (nsegs < queue_max_segments(q) && | |
d0e5fbb0 | 258 | sectors < max_sectors) { |
dcebd755 ML |
259 | /* split in the middle of bvec */ |
260 | bv.bv_len = (max_sectors - sectors) << 9; | |
261 | bvec_split_segs(q, &bv, &nsegs, | |
262 | &seg_size, | |
263 | &front_seg_size, | |
264 | §ors); | |
e36f6204 | 265 | } |
cf8c0c6a | 266 | goto split; |
e36f6204 KB |
267 | } |
268 | ||
38417468 | 269 | if (bvprvp) { |
b4b6cb61 ML |
270 | if (seg_size + bv.bv_len > queue_max_segment_size(q)) |
271 | goto new_segment; | |
3dccdae5 | 272 | if (!biovec_phys_mergeable(q, bvprvp, &bv)) |
54efd50b KO |
273 | goto new_segment; |
274 | ||
275 | seg_size += bv.bv_len; | |
276 | bvprv = bv; | |
578270bf | 277 | bvprvp = &bvprv; |
52cc6eea | 278 | sectors += bv.bv_len >> 9; |
a88d32af | 279 | |
aaeee62c ML |
280 | if (nsegs == 1 && seg_size > front_seg_size) |
281 | front_seg_size = seg_size; | |
282 | ||
54efd50b KO |
283 | continue; |
284 | } | |
285 | new_segment: | |
286 | if (nsegs == queue_max_segments(q)) | |
287 | goto split; | |
288 | ||
54efd50b | 289 | bvprv = bv; |
578270bf | 290 | bvprvp = &bvprv; |
dcebd755 | 291 | |
bbcbbd56 ML |
292 | if (bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
293 | nsegs++; | |
294 | seg_size = bv.bv_len; | |
295 | sectors += bv.bv_len >> 9; | |
296 | if (nsegs == 1 && seg_size > front_seg_size) | |
297 | front_seg_size = seg_size; | |
298 | } else if (bvec_split_segs(q, &bv, &nsegs, &seg_size, | |
299 | &front_seg_size, §ors)) { | |
dcebd755 | 300 | goto split; |
bbcbbd56 | 301 | } |
54efd50b KO |
302 | } |
303 | ||
02e70742 | 304 | do_split = false; |
54efd50b | 305 | split: |
bdced438 | 306 | *segs = nsegs; |
02e70742 ML |
307 | |
308 | if (do_split) { | |
309 | new = bio_split(bio, sectors, GFP_NOIO, bs); | |
310 | if (new) | |
311 | bio = new; | |
312 | } | |
313 | ||
314 | bio->bi_seg_front_size = front_seg_size; | |
315 | if (seg_size > bio->bi_seg_back_size) | |
316 | bio->bi_seg_back_size = seg_size; | |
317 | ||
318 | return do_split ? new : NULL; | |
54efd50b KO |
319 | } |
320 | ||
af67c31f | 321 | void blk_queue_split(struct request_queue *q, struct bio **bio) |
54efd50b | 322 | { |
bdced438 ML |
323 | struct bio *split, *res; |
324 | unsigned nsegs; | |
54efd50b | 325 | |
7afafc8a AH |
326 | switch (bio_op(*bio)) { |
327 | case REQ_OP_DISCARD: | |
328 | case REQ_OP_SECURE_ERASE: | |
338aa96d | 329 | split = blk_bio_discard_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a | 330 | break; |
a6f0788e | 331 | case REQ_OP_WRITE_ZEROES: |
338aa96d | 332 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, &nsegs); |
a6f0788e | 333 | break; |
7afafc8a | 334 | case REQ_OP_WRITE_SAME: |
338aa96d | 335 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
336 | break; |
337 | default: | |
338aa96d | 338 | split = blk_bio_segment_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
339 | break; |
340 | } | |
bdced438 ML |
341 | |
342 | /* physical segments can be figured out during splitting */ | |
343 | res = split ? split : *bio; | |
344 | res->bi_phys_segments = nsegs; | |
345 | bio_set_flag(res, BIO_SEG_VALID); | |
54efd50b KO |
346 | |
347 | if (split) { | |
6ac45aeb | 348 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 349 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 350 | |
947b7ac1 JA |
351 | /* |
352 | * Since we're recursing into make_request here, ensure | |
353 | * that we mark this bio as already having entered the queue. | |
354 | * If not, and the queue is going away, we can get stuck | |
355 | * forever on waiting for the queue reference to drop. But | |
356 | * that will never happen, as we're already holding a | |
357 | * reference to it. | |
358 | */ | |
359 | bio_set_flag(*bio, BIO_QUEUE_ENTERED); | |
360 | ||
54efd50b | 361 | bio_chain(split, *bio); |
cda22646 | 362 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
54efd50b KO |
363 | generic_make_request(*bio); |
364 | *bio = split; | |
365 | } | |
366 | } | |
367 | EXPORT_SYMBOL(blk_queue_split); | |
368 | ||
1e428079 | 369 | static unsigned int __blk_recalc_rq_segments(struct request_queue *q, |
2705c937 | 370 | struct bio *bio) |
d6d48196 | 371 | { |
7988613b | 372 | struct bio_vec bv, bvprv = { NULL }; |
38417468 | 373 | int prev = 0; |
1e428079 | 374 | unsigned int seg_size, nr_phys_segs; |
49b1f22b | 375 | unsigned front_seg_size; |
59247eae | 376 | struct bio *fbio, *bbio; |
7988613b | 377 | struct bvec_iter iter; |
d6d48196 | 378 | |
1e428079 JA |
379 | if (!bio) |
380 | return 0; | |
d6d48196 | 381 | |
49b1f22b ML |
382 | front_seg_size = bio->bi_seg_front_size; |
383 | ||
a6f0788e CK |
384 | switch (bio_op(bio)) { |
385 | case REQ_OP_DISCARD: | |
386 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 387 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
388 | return 0; |
389 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 390 | return 1; |
a6f0788e | 391 | } |
5cb8850c | 392 | |
1e428079 | 393 | fbio = bio; |
5df97b91 | 394 | seg_size = 0; |
2c8919de | 395 | nr_phys_segs = 0; |
1e428079 | 396 | for_each_bio(bio) { |
dcebd755 | 397 | bio_for_each_bvec(bv, bio, iter) { |
38417468 | 398 | if (prev) { |
7988613b | 399 | if (seg_size + bv.bv_len |
ae03bf63 | 400 | > queue_max_segment_size(q)) |
1e428079 | 401 | goto new_segment; |
3dccdae5 | 402 | if (!biovec_phys_mergeable(q, &bvprv, &bv)) |
1e428079 | 403 | goto new_segment; |
d6d48196 | 404 | |
7988613b | 405 | seg_size += bv.bv_len; |
1e428079 | 406 | bvprv = bv; |
aaeee62c ML |
407 | |
408 | if (nr_phys_segs == 1 && seg_size > | |
409 | front_seg_size) | |
410 | front_seg_size = seg_size; | |
411 | ||
1e428079 JA |
412 | continue; |
413 | } | |
d6d48196 | 414 | new_segment: |
1e428079 | 415 | bvprv = bv; |
54efd50b | 416 | prev = 1; |
dcebd755 ML |
417 | bvec_split_segs(q, &bv, &nr_phys_segs, &seg_size, |
418 | &front_seg_size, NULL); | |
1e428079 | 419 | } |
59247eae | 420 | bbio = bio; |
d6d48196 JA |
421 | } |
422 | ||
dcebd755 | 423 | fbio->bi_seg_front_size = front_seg_size; |
59247eae JA |
424 | if (seg_size > bbio->bi_seg_back_size) |
425 | bbio->bi_seg_back_size = seg_size; | |
1e428079 JA |
426 | |
427 | return nr_phys_segs; | |
428 | } | |
429 | ||
430 | void blk_recalc_rq_segments(struct request *rq) | |
431 | { | |
2705c937 | 432 | rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio); |
d6d48196 JA |
433 | } |
434 | ||
435 | void blk_recount_segments(struct request_queue *q, struct bio *bio) | |
436 | { | |
2705c937 | 437 | struct bio *nxt = bio->bi_next; |
05f1dd53 | 438 | |
2705c937 ML |
439 | bio->bi_next = NULL; |
440 | bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio); | |
441 | bio->bi_next = nxt; | |
1e428079 | 442 | |
b7c44ed9 | 443 | bio_set_flag(bio, BIO_SEG_VALID); |
d6d48196 | 444 | } |
d6d48196 JA |
445 | |
446 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, | |
447 | struct bio *nxt) | |
448 | { | |
2b8221e1 | 449 | struct bio_vec end_bv = { NULL }, nxt_bv; |
f619d254 | 450 | |
86771427 | 451 | if (bio->bi_seg_back_size + nxt->bi_seg_front_size > |
ae03bf63 | 452 | queue_max_segment_size(q)) |
d6d48196 JA |
453 | return 0; |
454 | ||
e17fc0a1 DW |
455 | if (!bio_has_data(bio)) |
456 | return 1; | |
457 | ||
e827091c ML |
458 | bio_get_last_bvec(bio, &end_bv); |
459 | bio_get_first_bvec(nxt, &nxt_bv); | |
f619d254 | 460 | |
3dccdae5 | 461 | return biovec_phys_mergeable(q, &end_bv, &nxt_bv); |
d6d48196 JA |
462 | } |
463 | ||
48d7727c | 464 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
465 | struct scatterlist *sglist) |
466 | { | |
467 | if (!*sg) | |
468 | return sglist; | |
469 | ||
470 | /* | |
471 | * If the driver previously mapped a shorter list, we could see a | |
472 | * termination bit prematurely unless it fully inits the sg table | |
473 | * on each mapping. We KNOW that there must be more entries here | |
474 | * or the driver would be buggy, so force clear the termination bit | |
475 | * to avoid doing a full sg_init_table() in drivers for each command. | |
476 | */ | |
477 | sg_unmark_end(*sg); | |
478 | return sg_next(*sg); | |
479 | } | |
480 | ||
481 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
482 | struct bio_vec *bvec, struct scatterlist *sglist, | |
483 | struct scatterlist **sg) | |
484 | { | |
485 | unsigned nbytes = bvec->bv_len; | |
486 | unsigned nsegs = 0, total = 0, offset = 0; | |
487 | ||
488 | while (nbytes > 0) { | |
489 | unsigned seg_size; | |
490 | struct page *pg; | |
491 | unsigned idx; | |
492 | ||
493 | *sg = blk_next_sg(sg, sglist); | |
494 | ||
495 | seg_size = get_max_segment_size(q, bvec->bv_offset + total); | |
496 | seg_size = min(nbytes, seg_size); | |
497 | ||
498 | offset = (total + bvec->bv_offset) % PAGE_SIZE; | |
499 | idx = (total + bvec->bv_offset) / PAGE_SIZE; | |
4d633062 | 500 | pg = bvec_nth_page(bvec->bv_page, idx); |
862e5a5e ML |
501 | |
502 | sg_set_page(*sg, pg, seg_size, offset); | |
503 | ||
504 | total += seg_size; | |
505 | nbytes -= seg_size; | |
506 | nsegs++; | |
507 | } | |
508 | ||
509 | return nsegs; | |
510 | } | |
511 | ||
7988613b | 512 | static inline void |
963ab9e5 | 513 | __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec, |
7988613b | 514 | struct scatterlist *sglist, struct bio_vec *bvprv, |
38417468 | 515 | struct scatterlist **sg, int *nsegs) |
963ab9e5 AH |
516 | { |
517 | ||
518 | int nbytes = bvec->bv_len; | |
519 | ||
38417468 | 520 | if (*sg) { |
b4b6cb61 ML |
521 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
522 | goto new_segment; | |
3dccdae5 | 523 | if (!biovec_phys_mergeable(q, bvprv, bvec)) |
963ab9e5 AH |
524 | goto new_segment; |
525 | ||
526 | (*sg)->length += nbytes; | |
527 | } else { | |
528 | new_segment: | |
48d7727c ML |
529 | if (bvec->bv_offset + bvec->bv_len <= PAGE_SIZE) { |
530 | *sg = blk_next_sg(sg, sglist); | |
531 | sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset); | |
532 | (*nsegs) += 1; | |
533 | } else | |
534 | (*nsegs) += blk_bvec_map_sg(q, bvec, sglist, sg); | |
963ab9e5 | 535 | } |
7988613b | 536 | *bvprv = *bvec; |
963ab9e5 AH |
537 | } |
538 | ||
f9d03f96 CH |
539 | static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv, |
540 | struct scatterlist *sglist, struct scatterlist **sg) | |
541 | { | |
542 | *sg = sglist; | |
543 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
544 | return 1; | |
545 | } | |
546 | ||
5cb8850c KO |
547 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
548 | struct scatterlist *sglist, | |
549 | struct scatterlist **sg) | |
d6d48196 | 550 | { |
2b8221e1 | 551 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 552 | struct bvec_iter iter; |
38417468 | 553 | int nsegs = 0; |
5cb8850c KO |
554 | |
555 | for_each_bio(bio) | |
862e5a5e | 556 | bio_for_each_bvec(bvec, bio, iter) |
5cb8850c | 557 | __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg, |
38417468 | 558 | &nsegs); |
d6d48196 | 559 | |
5cb8850c KO |
560 | return nsegs; |
561 | } | |
562 | ||
563 | /* | |
564 | * map a request to scatterlist, return number of sg entries setup. Caller | |
565 | * must make sure sg can hold rq->nr_phys_segments entries | |
566 | */ | |
567 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, | |
568 | struct scatterlist *sglist) | |
569 | { | |
570 | struct scatterlist *sg = NULL; | |
571 | int nsegs = 0; | |
572 | ||
f9d03f96 CH |
573 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
574 | nsegs = __blk_bvec_map_sg(q, rq->special_vec, sglist, &sg); | |
575 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) | |
576 | nsegs = __blk_bvec_map_sg(q, bio_iovec(rq->bio), sglist, &sg); | |
577 | else if (rq->bio) | |
5cb8850c | 578 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg); |
f18573ab | 579 | |
e8064021 | 580 | if (unlikely(rq->rq_flags & RQF_COPY_USER) && |
2e46e8b2 TH |
581 | (blk_rq_bytes(rq) & q->dma_pad_mask)) { |
582 | unsigned int pad_len = | |
583 | (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1; | |
f18573ab FT |
584 | |
585 | sg->length += pad_len; | |
586 | rq->extra_len += pad_len; | |
587 | } | |
588 | ||
2fb98e84 | 589 | if (q->dma_drain_size && q->dma_drain_needed(rq)) { |
a8ebb056 | 590 | if (op_is_write(req_op(rq))) |
db0a2e00 TH |
591 | memset(q->dma_drain_buffer, 0, q->dma_drain_size); |
592 | ||
da81ed16 | 593 | sg_unmark_end(sg); |
d6d48196 JA |
594 | sg = sg_next(sg); |
595 | sg_set_page(sg, virt_to_page(q->dma_drain_buffer), | |
596 | q->dma_drain_size, | |
597 | ((unsigned long)q->dma_drain_buffer) & | |
598 | (PAGE_SIZE - 1)); | |
599 | nsegs++; | |
7a85f889 | 600 | rq->extra_len += q->dma_drain_size; |
d6d48196 JA |
601 | } |
602 | ||
603 | if (sg) | |
604 | sg_mark_end(sg); | |
605 | ||
12e57f59 ML |
606 | /* |
607 | * Something must have been wrong if the figured number of | |
608 | * segment is bigger than number of req's physical segments | |
609 | */ | |
f9d03f96 | 610 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 611 | |
d6d48196 JA |
612 | return nsegs; |
613 | } | |
d6d48196 JA |
614 | EXPORT_SYMBOL(blk_rq_map_sg); |
615 | ||
d6d48196 JA |
616 | static inline int ll_new_hw_segment(struct request_queue *q, |
617 | struct request *req, | |
618 | struct bio *bio) | |
619 | { | |
d6d48196 JA |
620 | int nr_phys_segs = bio_phys_segments(q, bio); |
621 | ||
13f05c8d MP |
622 | if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q)) |
623 | goto no_merge; | |
624 | ||
4eaf99be | 625 | if (blk_integrity_merge_bio(q, req, bio) == false) |
13f05c8d | 626 | goto no_merge; |
d6d48196 JA |
627 | |
628 | /* | |
629 | * This will form the start of a new hw segment. Bump both | |
630 | * counters. | |
631 | */ | |
d6d48196 JA |
632 | req->nr_phys_segments += nr_phys_segs; |
633 | return 1; | |
13f05c8d MP |
634 | |
635 | no_merge: | |
e0c72300 | 636 | req_set_nomerge(q, req); |
13f05c8d | 637 | return 0; |
d6d48196 JA |
638 | } |
639 | ||
640 | int ll_back_merge_fn(struct request_queue *q, struct request *req, | |
641 | struct bio *bio) | |
642 | { | |
5e7c4274 JA |
643 | if (req_gap_back_merge(req, bio)) |
644 | return 0; | |
7f39add3 SG |
645 | if (blk_integrity_rq(req) && |
646 | integrity_req_gap_back_merge(req, bio)) | |
647 | return 0; | |
f31dc1cd | 648 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 649 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
e0c72300 | 650 | req_set_nomerge(q, req); |
d6d48196 JA |
651 | return 0; |
652 | } | |
2cdf79ca | 653 | if (!bio_flagged(req->biotail, BIO_SEG_VALID)) |
d6d48196 | 654 | blk_recount_segments(q, req->biotail); |
2cdf79ca | 655 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 656 | blk_recount_segments(q, bio); |
d6d48196 JA |
657 | |
658 | return ll_new_hw_segment(q, req, bio); | |
659 | } | |
660 | ||
6728cb0e | 661 | int ll_front_merge_fn(struct request_queue *q, struct request *req, |
d6d48196 JA |
662 | struct bio *bio) |
663 | { | |
5e7c4274 JA |
664 | |
665 | if (req_gap_front_merge(req, bio)) | |
666 | return 0; | |
7f39add3 SG |
667 | if (blk_integrity_rq(req) && |
668 | integrity_req_gap_front_merge(req, bio)) | |
669 | return 0; | |
f31dc1cd | 670 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 671 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
e0c72300 | 672 | req_set_nomerge(q, req); |
d6d48196 JA |
673 | return 0; |
674 | } | |
2cdf79ca | 675 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 676 | blk_recount_segments(q, bio); |
2cdf79ca | 677 | if (!bio_flagged(req->bio, BIO_SEG_VALID)) |
d6d48196 | 678 | blk_recount_segments(q, req->bio); |
d6d48196 JA |
679 | |
680 | return ll_new_hw_segment(q, req, bio); | |
681 | } | |
682 | ||
445251d0 JA |
683 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
684 | struct request *next) | |
685 | { | |
686 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
687 | ||
688 | if (segments >= queue_max_discard_segments(q)) | |
689 | goto no_merge; | |
690 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
691 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
692 | goto no_merge; | |
693 | ||
694 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
695 | return true; | |
696 | no_merge: | |
697 | req_set_nomerge(q, req); | |
698 | return false; | |
699 | } | |
700 | ||
d6d48196 JA |
701 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
702 | struct request *next) | |
703 | { | |
704 | int total_phys_segments; | |
86771427 FT |
705 | unsigned int seg_size = |
706 | req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size; | |
d6d48196 | 707 | |
5e7c4274 | 708 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
709 | return 0; |
710 | ||
d6d48196 JA |
711 | /* |
712 | * Will it become too large? | |
713 | */ | |
f31dc1cd | 714 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 715 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
716 | return 0; |
717 | ||
718 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
86771427 FT |
719 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) { |
720 | if (req->nr_phys_segments == 1) | |
721 | req->bio->bi_seg_front_size = seg_size; | |
722 | if (next->nr_phys_segments == 1) | |
723 | next->biotail->bi_seg_back_size = seg_size; | |
d6d48196 | 724 | total_phys_segments--; |
86771427 | 725 | } |
d6d48196 | 726 | |
8a78362c | 727 | if (total_phys_segments > queue_max_segments(q)) |
d6d48196 JA |
728 | return 0; |
729 | ||
4eaf99be | 730 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
731 | return 0; |
732 | ||
d6d48196 JA |
733 | /* Merge is OK... */ |
734 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
735 | return 1; |
736 | } | |
737 | ||
80a761fd TH |
738 | /** |
739 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
740 | * @rq: request to mark as mixed merge | |
741 | * | |
742 | * Description: | |
743 | * @rq is about to be mixed merged. Make sure the attributes | |
744 | * which can be mixed are set in each bio and mark @rq as mixed | |
745 | * merged. | |
746 | */ | |
747 | void blk_rq_set_mixed_merge(struct request *rq) | |
748 | { | |
749 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
750 | struct bio *bio; | |
751 | ||
e8064021 | 752 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
753 | return; |
754 | ||
755 | /* | |
756 | * @rq will no longer represent mixable attributes for all the | |
757 | * contained bios. It will just track those of the first one. | |
758 | * Distributes the attributs to each bio. | |
759 | */ | |
760 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
761 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
762 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
763 | bio->bi_opf |= ff; | |
80a761fd | 764 | } |
e8064021 | 765 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
766 | } |
767 | ||
26308eab JM |
768 | static void blk_account_io_merge(struct request *req) |
769 | { | |
770 | if (blk_do_io_stat(req)) { | |
771 | struct hd_struct *part; | |
26308eab | 772 | |
112f158f | 773 | part_stat_lock(); |
09e099d4 | 774 | part = req->part; |
26308eab | 775 | |
d62e26b3 | 776 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
26308eab | 777 | |
6c23a968 | 778 | hd_struct_put(part); |
26308eab JM |
779 | part_stat_unlock(); |
780 | } | |
781 | } | |
69840466 JW |
782 | /* |
783 | * Two cases of handling DISCARD merge: | |
784 | * If max_discard_segments > 1, the driver takes every bio | |
785 | * as a range and send them to controller together. The ranges | |
786 | * needn't to be contiguous. | |
787 | * Otherwise, the bios/requests will be handled as same as | |
788 | * others which should be contiguous. | |
789 | */ | |
790 | static inline bool blk_discard_mergable(struct request *req) | |
791 | { | |
792 | if (req_op(req) == REQ_OP_DISCARD && | |
793 | queue_max_discard_segments(req->q) > 1) | |
794 | return true; | |
795 | return false; | |
796 | } | |
797 | ||
e96c0d83 EB |
798 | static enum elv_merge blk_try_req_merge(struct request *req, |
799 | struct request *next) | |
69840466 JW |
800 | { |
801 | if (blk_discard_mergable(req)) | |
802 | return ELEVATOR_DISCARD_MERGE; | |
803 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
804 | return ELEVATOR_BACK_MERGE; | |
805 | ||
806 | return ELEVATOR_NO_MERGE; | |
807 | } | |
26308eab | 808 | |
d6d48196 | 809 | /* |
b973cb7e JA |
810 | * For non-mq, this has to be called with the request spinlock acquired. |
811 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 812 | */ |
b973cb7e JA |
813 | static struct request *attempt_merge(struct request_queue *q, |
814 | struct request *req, struct request *next) | |
d6d48196 JA |
815 | { |
816 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 817 | return NULL; |
d6d48196 | 818 | |
288dab8a | 819 | if (req_op(req) != req_op(next)) |
b973cb7e | 820 | return NULL; |
f31dc1cd | 821 | |
d6d48196 | 822 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 823 | || req->rq_disk != next->rq_disk) |
b973cb7e | 824 | return NULL; |
d6d48196 | 825 | |
8fe0d473 | 826 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 827 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 828 | return NULL; |
4363ac7c | 829 | |
cb6934f8 JA |
830 | /* |
831 | * Don't allow merge of different write hints, or for a hint with | |
832 | * non-hint IO. | |
833 | */ | |
834 | if (req->write_hint != next->write_hint) | |
835 | return NULL; | |
836 | ||
668ffc03 DLM |
837 | if (req->ioprio != next->ioprio) |
838 | return NULL; | |
839 | ||
d6d48196 JA |
840 | /* |
841 | * If we are allowed to merge, then append bio list | |
842 | * from next to rq and release next. merge_requests_fn | |
843 | * will have updated segment counts, update sector | |
445251d0 JA |
844 | * counts here. Handle DISCARDs separately, as they |
845 | * have separate settings. | |
d6d48196 | 846 | */ |
69840466 JW |
847 | |
848 | switch (blk_try_req_merge(req, next)) { | |
849 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
850 | if (!req_attempt_discard_merge(q, req, next)) |
851 | return NULL; | |
69840466 JW |
852 | break; |
853 | case ELEVATOR_BACK_MERGE: | |
854 | if (!ll_merge_requests_fn(q, req, next)) | |
855 | return NULL; | |
856 | break; | |
857 | default: | |
b973cb7e | 858 | return NULL; |
69840466 | 859 | } |
d6d48196 | 860 | |
80a761fd TH |
861 | /* |
862 | * If failfast settings disagree or any of the two is already | |
863 | * a mixed merge, mark both as mixed before proceeding. This | |
864 | * makes sure that all involved bios have mixable attributes | |
865 | * set properly. | |
866 | */ | |
e8064021 | 867 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
868 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
869 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
870 | blk_rq_set_mixed_merge(req); | |
871 | blk_rq_set_mixed_merge(next); | |
872 | } | |
873 | ||
d6d48196 | 874 | /* |
522a7775 OS |
875 | * At this point we have either done a back merge or front merge. We |
876 | * need the smaller start_time_ns of the merged requests to be the | |
877 | * current request for accounting purposes. | |
d6d48196 | 878 | */ |
522a7775 OS |
879 | if (next->start_time_ns < req->start_time_ns) |
880 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
881 | |
882 | req->biotail->bi_next = next->bio; | |
883 | req->biotail = next->biotail; | |
884 | ||
a2dec7b3 | 885 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 886 | |
2a5cf35c | 887 | if (!blk_discard_mergable(req)) |
445251d0 | 888 | elv_merge_requests(q, req, next); |
d6d48196 | 889 | |
42dad764 JM |
890 | /* |
891 | * 'next' is going away, so update stats accordingly | |
892 | */ | |
893 | blk_account_io_merge(next); | |
d6d48196 | 894 | |
e4d750c9 JA |
895 | /* |
896 | * ownership of bio passed from next to req, return 'next' for | |
897 | * the caller to free | |
898 | */ | |
1cd96c24 | 899 | next->bio = NULL; |
b973cb7e | 900 | return next; |
d6d48196 JA |
901 | } |
902 | ||
b973cb7e | 903 | struct request *attempt_back_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
904 | { |
905 | struct request *next = elv_latter_request(q, rq); | |
906 | ||
907 | if (next) | |
908 | return attempt_merge(q, rq, next); | |
909 | ||
b973cb7e | 910 | return NULL; |
d6d48196 JA |
911 | } |
912 | ||
b973cb7e | 913 | struct request *attempt_front_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
914 | { |
915 | struct request *prev = elv_former_request(q, rq); | |
916 | ||
917 | if (prev) | |
918 | return attempt_merge(q, prev, rq); | |
919 | ||
b973cb7e | 920 | return NULL; |
d6d48196 | 921 | } |
5e84ea3a JA |
922 | |
923 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
924 | struct request *next) | |
925 | { | |
e4d750c9 | 926 | struct request *free; |
72ef799b | 927 | |
e4d750c9 JA |
928 | free = attempt_merge(q, rq, next); |
929 | if (free) { | |
92bc5a24 | 930 | blk_put_request(free); |
e4d750c9 JA |
931 | return 1; |
932 | } | |
933 | ||
934 | return 0; | |
5e84ea3a | 935 | } |
050c8ea8 TH |
936 | |
937 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
938 | { | |
e2a60da7 | 939 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
940 | return false; |
941 | ||
288dab8a | 942 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
943 | return false; |
944 | ||
050c8ea8 TH |
945 | /* different data direction or already started, don't merge */ |
946 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
947 | return false; | |
948 | ||
2081a56b JA |
949 | /* must be same device */ |
950 | if (rq->rq_disk != bio->bi_disk) | |
050c8ea8 TH |
951 | return false; |
952 | ||
953 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 954 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
955 | return false; |
956 | ||
4363ac7c | 957 | /* must be using the same buffer */ |
8fe0d473 | 958 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
959 | !blk_write_same_mergeable(rq->bio, bio)) |
960 | return false; | |
961 | ||
cb6934f8 JA |
962 | /* |
963 | * Don't allow merge of different write hints, or for a hint with | |
964 | * non-hint IO. | |
965 | */ | |
966 | if (rq->write_hint != bio->bi_write_hint) | |
967 | return false; | |
968 | ||
668ffc03 DLM |
969 | if (rq->ioprio != bio_prio(bio)) |
970 | return false; | |
971 | ||
050c8ea8 TH |
972 | return true; |
973 | } | |
974 | ||
34fe7c05 | 975 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 976 | { |
69840466 | 977 | if (blk_discard_mergable(rq)) |
1e739730 CH |
978 | return ELEVATOR_DISCARD_MERGE; |
979 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 980 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 981 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
982 | return ELEVATOR_FRONT_MERGE; |
983 | return ELEVATOR_NO_MERGE; | |
984 | } |