]>
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 | static inline bool bio_will_gap(struct request_queue *q, |
16 | struct request *prev_rq, struct bio *prev, struct bio *next) | |
17 | { | |
18 | struct bio_vec pb, nb; | |
19 | ||
20 | if (!bio_has_data(prev) || !queue_virt_boundary(q)) | |
21 | return false; | |
22 | ||
23 | /* | |
24 | * Don't merge if the 1st bio starts with non-zero offset, otherwise it | |
25 | * is quite difficult to respect the sg gap limit. We work hard to | |
26 | * merge a huge number of small single bios in case of mkfs. | |
27 | */ | |
28 | if (prev_rq) | |
29 | bio_get_first_bvec(prev_rq->bio, &pb); | |
30 | else | |
31 | bio_get_first_bvec(prev, &pb); | |
df376b2e | 32 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
33 | return true; |
34 | ||
35 | /* | |
36 | * We don't need to worry about the situation that the merged segment | |
37 | * ends in unaligned virt boundary: | |
38 | * | |
39 | * - if 'pb' ends aligned, the merged segment ends aligned | |
40 | * - if 'pb' ends unaligned, the next bio must include | |
41 | * one single bvec of 'nb', otherwise the 'nb' can't | |
42 | * merge with 'pb' | |
43 | */ | |
44 | bio_get_last_bvec(prev, &pb); | |
45 | bio_get_first_bvec(next, &nb); | |
200a9aff | 46 | if (biovec_phys_mergeable(q, &pb, &nb)) |
e9907009 CH |
47 | return false; |
48 | return __bvec_gap_to_prev(q, &pb, nb.bv_offset); | |
49 | } | |
50 | ||
51 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) | |
52 | { | |
53 | return bio_will_gap(req->q, req, req->biotail, bio); | |
54 | } | |
55 | ||
56 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) | |
57 | { | |
58 | return bio_will_gap(req->q, NULL, bio, req->bio); | |
59 | } | |
60 | ||
54efd50b KO |
61 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
62 | struct bio *bio, | |
bdced438 ML |
63 | struct bio_set *bs, |
64 | unsigned *nsegs) | |
54efd50b KO |
65 | { |
66 | unsigned int max_discard_sectors, granularity; | |
67 | int alignment; | |
68 | sector_t tmp; | |
69 | unsigned split_sectors; | |
70 | ||
bdced438 ML |
71 | *nsegs = 1; |
72 | ||
54efd50b KO |
73 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
74 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
75 | ||
1adfc5e4 ML |
76 | max_discard_sectors = min(q->limits.max_discard_sectors, |
77 | bio_allowed_max_sectors(q)); | |
54efd50b KO |
78 | max_discard_sectors -= max_discard_sectors % granularity; |
79 | ||
80 | if (unlikely(!max_discard_sectors)) { | |
81 | /* XXX: warn */ | |
82 | return NULL; | |
83 | } | |
84 | ||
85 | if (bio_sectors(bio) <= max_discard_sectors) | |
86 | return NULL; | |
87 | ||
88 | split_sectors = max_discard_sectors; | |
89 | ||
90 | /* | |
91 | * If the next starting sector would be misaligned, stop the discard at | |
92 | * the previous aligned sector. | |
93 | */ | |
94 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
95 | ||
96 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
97 | tmp = sector_div(tmp, granularity); | |
98 | ||
99 | if (split_sectors > tmp) | |
100 | split_sectors -= tmp; | |
101 | ||
102 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
103 | } | |
104 | ||
885fa13f CH |
105 | static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, |
106 | struct bio *bio, struct bio_set *bs, unsigned *nsegs) | |
107 | { | |
108 | *nsegs = 1; | |
109 | ||
110 | if (!q->limits.max_write_zeroes_sectors) | |
111 | return NULL; | |
112 | ||
113 | if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) | |
114 | return NULL; | |
115 | ||
116 | return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); | |
117 | } | |
118 | ||
54efd50b KO |
119 | static struct bio *blk_bio_write_same_split(struct request_queue *q, |
120 | struct bio *bio, | |
bdced438 ML |
121 | struct bio_set *bs, |
122 | unsigned *nsegs) | |
54efd50b | 123 | { |
bdced438 ML |
124 | *nsegs = 1; |
125 | ||
54efd50b KO |
126 | if (!q->limits.max_write_same_sectors) |
127 | return NULL; | |
128 | ||
129 | if (bio_sectors(bio) <= q->limits.max_write_same_sectors) | |
130 | return NULL; | |
131 | ||
132 | return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs); | |
133 | } | |
134 | ||
d0e5fbb0 ML |
135 | static inline unsigned get_max_io_size(struct request_queue *q, |
136 | struct bio *bio) | |
137 | { | |
138 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); | |
139 | unsigned mask = queue_logical_block_size(q) - 1; | |
140 | ||
141 | /* aligned to logical block size */ | |
142 | sectors &= ~(mask >> 9); | |
143 | ||
144 | return sectors; | |
145 | } | |
146 | ||
dcebd755 ML |
147 | static unsigned get_max_segment_size(struct request_queue *q, |
148 | unsigned offset) | |
149 | { | |
150 | unsigned long mask = queue_segment_boundary(q); | |
151 | ||
152 | /* default segment boundary mask means no boundary limit */ | |
153 | if (mask == BLK_SEG_BOUNDARY_MASK) | |
154 | return queue_max_segment_size(q); | |
155 | ||
156 | return min_t(unsigned long, mask - (mask & offset) + 1, | |
157 | queue_max_segment_size(q)); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Split the bvec @bv into segments, and update all kinds of | |
162 | * variables. | |
163 | */ | |
164 | static bool bvec_split_segs(struct request_queue *q, struct bio_vec *bv, | |
165 | unsigned *nsegs, unsigned *last_seg_size, | |
05b700ba | 166 | unsigned *front_seg_size, unsigned *sectors, unsigned max_segs) |
dcebd755 ML |
167 | { |
168 | unsigned len = bv->bv_len; | |
169 | unsigned total_len = 0; | |
170 | unsigned new_nsegs = 0, seg_size = 0; | |
171 | ||
172 | /* | |
173 | * Multi-page bvec may be too big to hold in one segment, so the | |
174 | * current bvec has to be splitted as multiple segments. | |
175 | */ | |
05b700ba | 176 | while (len && new_nsegs + *nsegs < max_segs) { |
dcebd755 ML |
177 | seg_size = get_max_segment_size(q, bv->bv_offset + total_len); |
178 | seg_size = min(seg_size, len); | |
179 | ||
180 | new_nsegs++; | |
181 | total_len += seg_size; | |
182 | len -= seg_size; | |
183 | ||
184 | if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) | |
185 | break; | |
186 | } | |
187 | ||
188 | if (!new_nsegs) | |
189 | return !!len; | |
190 | ||
191 | /* update front segment size */ | |
192 | if (!*nsegs) { | |
193 | unsigned first_seg_size; | |
194 | ||
195 | if (new_nsegs == 1) | |
196 | first_seg_size = get_max_segment_size(q, bv->bv_offset); | |
197 | else | |
198 | first_seg_size = queue_max_segment_size(q); | |
199 | ||
200 | if (*front_seg_size < first_seg_size) | |
201 | *front_seg_size = first_seg_size; | |
202 | } | |
203 | ||
204 | /* update other varibles */ | |
205 | *last_seg_size = seg_size; | |
206 | *nsegs += new_nsegs; | |
207 | if (sectors) | |
208 | *sectors += total_len >> 9; | |
209 | ||
210 | /* split in the middle of the bvec if len != 0 */ | |
211 | return !!len; | |
212 | } | |
213 | ||
54efd50b KO |
214 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
215 | struct bio *bio, | |
bdced438 ML |
216 | struct bio_set *bs, |
217 | unsigned *segs) | |
54efd50b | 218 | { |
5014c311 | 219 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 220 | struct bvec_iter iter; |
8ae12666 | 221 | unsigned seg_size = 0, nsegs = 0, sectors = 0; |
02e70742 ML |
222 | unsigned front_seg_size = bio->bi_seg_front_size; |
223 | bool do_split = true; | |
224 | struct bio *new = NULL; | |
d0e5fbb0 | 225 | const unsigned max_sectors = get_max_io_size(q, bio); |
05b700ba | 226 | const unsigned max_segs = queue_max_segments(q); |
54efd50b | 227 | |
dcebd755 | 228 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
229 | /* |
230 | * If the queue doesn't support SG gaps and adding this | |
231 | * offset would create a gap, disallow it. | |
232 | */ | |
5014c311 | 233 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
234 | goto split; |
235 | ||
d0e5fbb0 | 236 | if (sectors + (bv.bv_len >> 9) > max_sectors) { |
e36f6204 KB |
237 | /* |
238 | * Consider this a new segment if we're splitting in | |
239 | * the middle of this vector. | |
240 | */ | |
05b700ba | 241 | if (nsegs < max_segs && |
d0e5fbb0 | 242 | sectors < max_sectors) { |
dcebd755 ML |
243 | /* split in the middle of bvec */ |
244 | bv.bv_len = (max_sectors - sectors) << 9; | |
245 | bvec_split_segs(q, &bv, &nsegs, | |
246 | &seg_size, | |
247 | &front_seg_size, | |
05b700ba | 248 | §ors, max_segs); |
e36f6204 | 249 | } |
cf8c0c6a | 250 | goto split; |
e36f6204 KB |
251 | } |
252 | ||
05b700ba | 253 | if (nsegs == max_segs) |
54efd50b KO |
254 | goto split; |
255 | ||
54efd50b | 256 | bvprv = bv; |
578270bf | 257 | bvprvp = &bvprv; |
dcebd755 | 258 | |
bbcbbd56 ML |
259 | if (bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
260 | nsegs++; | |
261 | seg_size = bv.bv_len; | |
262 | sectors += bv.bv_len >> 9; | |
263 | if (nsegs == 1 && seg_size > front_seg_size) | |
264 | front_seg_size = seg_size; | |
265 | } else if (bvec_split_segs(q, &bv, &nsegs, &seg_size, | |
05b700ba | 266 | &front_seg_size, §ors, max_segs)) { |
dcebd755 | 267 | goto split; |
bbcbbd56 | 268 | } |
54efd50b KO |
269 | } |
270 | ||
02e70742 | 271 | do_split = false; |
54efd50b | 272 | split: |
bdced438 | 273 | *segs = nsegs; |
02e70742 ML |
274 | |
275 | if (do_split) { | |
276 | new = bio_split(bio, sectors, GFP_NOIO, bs); | |
277 | if (new) | |
278 | bio = new; | |
279 | } | |
280 | ||
281 | bio->bi_seg_front_size = front_seg_size; | |
282 | if (seg_size > bio->bi_seg_back_size) | |
283 | bio->bi_seg_back_size = seg_size; | |
284 | ||
285 | return do_split ? new : NULL; | |
54efd50b KO |
286 | } |
287 | ||
af67c31f | 288 | void blk_queue_split(struct request_queue *q, struct bio **bio) |
54efd50b | 289 | { |
bdced438 ML |
290 | struct bio *split, *res; |
291 | unsigned nsegs; | |
54efd50b | 292 | |
7afafc8a AH |
293 | switch (bio_op(*bio)) { |
294 | case REQ_OP_DISCARD: | |
295 | case REQ_OP_SECURE_ERASE: | |
338aa96d | 296 | split = blk_bio_discard_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a | 297 | break; |
a6f0788e | 298 | case REQ_OP_WRITE_ZEROES: |
338aa96d | 299 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, &nsegs); |
a6f0788e | 300 | break; |
7afafc8a | 301 | case REQ_OP_WRITE_SAME: |
338aa96d | 302 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
303 | break; |
304 | default: | |
338aa96d | 305 | split = blk_bio_segment_split(q, *bio, &q->bio_split, &nsegs); |
7afafc8a AH |
306 | break; |
307 | } | |
bdced438 ML |
308 | |
309 | /* physical segments can be figured out during splitting */ | |
310 | res = split ? split : *bio; | |
311 | res->bi_phys_segments = nsegs; | |
312 | bio_set_flag(res, BIO_SEG_VALID); | |
54efd50b KO |
313 | |
314 | if (split) { | |
6ac45aeb | 315 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 316 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 317 | |
947b7ac1 JA |
318 | /* |
319 | * Since we're recursing into make_request here, ensure | |
320 | * that we mark this bio as already having entered the queue. | |
321 | * If not, and the queue is going away, we can get stuck | |
322 | * forever on waiting for the queue reference to drop. But | |
323 | * that will never happen, as we're already holding a | |
324 | * reference to it. | |
325 | */ | |
326 | bio_set_flag(*bio, BIO_QUEUE_ENTERED); | |
327 | ||
54efd50b | 328 | bio_chain(split, *bio); |
cda22646 | 329 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
54efd50b KO |
330 | generic_make_request(*bio); |
331 | *bio = split; | |
332 | } | |
333 | } | |
334 | EXPORT_SYMBOL(blk_queue_split); | |
335 | ||
1e428079 | 336 | static unsigned int __blk_recalc_rq_segments(struct request_queue *q, |
2705c937 | 337 | struct bio *bio) |
d6d48196 | 338 | { |
b21e11c5 | 339 | struct bio_vec uninitialized_var(bv), bvprv = { NULL }; |
1e428079 | 340 | unsigned int seg_size, nr_phys_segs; |
49b1f22b | 341 | unsigned front_seg_size; |
59247eae | 342 | struct bio *fbio, *bbio; |
7988613b | 343 | struct bvec_iter iter; |
d6d48196 | 344 | |
1e428079 JA |
345 | if (!bio) |
346 | return 0; | |
d6d48196 | 347 | |
49b1f22b ML |
348 | front_seg_size = bio->bi_seg_front_size; |
349 | ||
a6f0788e CK |
350 | switch (bio_op(bio)) { |
351 | case REQ_OP_DISCARD: | |
352 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 353 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
354 | return 0; |
355 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 356 | return 1; |
a6f0788e | 357 | } |
5cb8850c | 358 | |
1e428079 | 359 | fbio = bio; |
5df97b91 | 360 | seg_size = 0; |
2c8919de | 361 | nr_phys_segs = 0; |
1e428079 | 362 | for_each_bio(bio) { |
dcebd755 | 363 | bio_for_each_bvec(bv, bio, iter) { |
dcebd755 | 364 | bvec_split_segs(q, &bv, &nr_phys_segs, &seg_size, |
05b700ba | 365 | &front_seg_size, NULL, UINT_MAX); |
1e428079 | 366 | } |
59247eae | 367 | bbio = bio; |
eded341c | 368 | if (likely(bio->bi_iter.bi_size)) |
b21e11c5 | 369 | bvprv = bv; |
d6d48196 JA |
370 | } |
371 | ||
dcebd755 | 372 | fbio->bi_seg_front_size = front_seg_size; |
59247eae JA |
373 | if (seg_size > bbio->bi_seg_back_size) |
374 | bbio->bi_seg_back_size = seg_size; | |
1e428079 JA |
375 | |
376 | return nr_phys_segs; | |
377 | } | |
378 | ||
379 | void blk_recalc_rq_segments(struct request *rq) | |
380 | { | |
2705c937 | 381 | rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio); |
d6d48196 JA |
382 | } |
383 | ||
384 | void blk_recount_segments(struct request_queue *q, struct bio *bio) | |
385 | { | |
2705c937 | 386 | struct bio *nxt = bio->bi_next; |
05f1dd53 | 387 | |
2705c937 ML |
388 | bio->bi_next = NULL; |
389 | bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio); | |
390 | bio->bi_next = nxt; | |
1e428079 | 391 | |
b7c44ed9 | 392 | bio_set_flag(bio, BIO_SEG_VALID); |
d6d48196 | 393 | } |
d6d48196 JA |
394 | |
395 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, | |
396 | struct bio *nxt) | |
397 | { | |
2b8221e1 | 398 | struct bio_vec end_bv = { NULL }, nxt_bv; |
f619d254 | 399 | |
86771427 | 400 | if (bio->bi_seg_back_size + nxt->bi_seg_front_size > |
ae03bf63 | 401 | queue_max_segment_size(q)) |
d6d48196 JA |
402 | return 0; |
403 | ||
e17fc0a1 DW |
404 | if (!bio_has_data(bio)) |
405 | return 1; | |
406 | ||
e827091c ML |
407 | bio_get_last_bvec(bio, &end_bv); |
408 | bio_get_first_bvec(nxt, &nxt_bv); | |
f619d254 | 409 | |
3dccdae5 | 410 | return biovec_phys_mergeable(q, &end_bv, &nxt_bv); |
d6d48196 JA |
411 | } |
412 | ||
48d7727c | 413 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
414 | struct scatterlist *sglist) |
415 | { | |
416 | if (!*sg) | |
417 | return sglist; | |
418 | ||
419 | /* | |
420 | * If the driver previously mapped a shorter list, we could see a | |
421 | * termination bit prematurely unless it fully inits the sg table | |
422 | * on each mapping. We KNOW that there must be more entries here | |
423 | * or the driver would be buggy, so force clear the termination bit | |
424 | * to avoid doing a full sg_init_table() in drivers for each command. | |
425 | */ | |
426 | sg_unmark_end(*sg); | |
427 | return sg_next(*sg); | |
428 | } | |
429 | ||
430 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
431 | struct bio_vec *bvec, struct scatterlist *sglist, | |
432 | struct scatterlist **sg) | |
433 | { | |
434 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 435 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
436 | |
437 | while (nbytes > 0) { | |
8a96a0e4 CH |
438 | unsigned offset = bvec->bv_offset + total; |
439 | unsigned len = min(get_max_segment_size(q, offset), nbytes); | |
f9f76879 CH |
440 | struct page *page = bvec->bv_page; |
441 | ||
442 | /* | |
443 | * Unfortunately a fair number of drivers barf on scatterlists | |
444 | * that have an offset larger than PAGE_SIZE, despite other | |
445 | * subsystems dealing with that invariant just fine. For now | |
446 | * stick to the legacy format where we never present those from | |
447 | * the block layer, but the code below should be removed once | |
448 | * these offenders (mostly MMC/SD drivers) are fixed. | |
449 | */ | |
450 | page += (offset >> PAGE_SHIFT); | |
451 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
452 | |
453 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 454 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 455 | |
8a96a0e4 CH |
456 | total += len; |
457 | nbytes -= len; | |
862e5a5e ML |
458 | nsegs++; |
459 | } | |
460 | ||
461 | return nsegs; | |
462 | } | |
463 | ||
16e3e418 ML |
464 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
465 | struct scatterlist *sglist, struct scatterlist **sg) | |
466 | { | |
467 | *sg = blk_next_sg(sg, sglist); | |
468 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
469 | return 1; | |
470 | } | |
471 | ||
f6970f83 ML |
472 | /* only try to merge bvecs into one sg if they are from two bios */ |
473 | static inline bool | |
474 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
475 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
476 | { |
477 | ||
478 | int nbytes = bvec->bv_len; | |
479 | ||
f6970f83 ML |
480 | if (!*sg) |
481 | return false; | |
963ab9e5 | 482 | |
f6970f83 ML |
483 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
484 | return false; | |
485 | ||
486 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
487 | return false; | |
488 | ||
489 | (*sg)->length += nbytes; | |
490 | ||
491 | return true; | |
963ab9e5 AH |
492 | } |
493 | ||
5cb8850c KO |
494 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
495 | struct scatterlist *sglist, | |
496 | struct scatterlist **sg) | |
d6d48196 | 497 | { |
b21e11c5 | 498 | struct bio_vec uninitialized_var(bvec), bvprv = { NULL }; |
5cb8850c | 499 | struct bvec_iter iter; |
38417468 | 500 | int nsegs = 0; |
f6970f83 | 501 | bool new_bio = false; |
5cb8850c | 502 | |
f6970f83 ML |
503 | for_each_bio(bio) { |
504 | bio_for_each_bvec(bvec, bio, iter) { | |
505 | /* | |
506 | * Only try to merge bvecs from two bios given we | |
507 | * have done bio internal merge when adding pages | |
508 | * to bio | |
509 | */ | |
510 | if (new_bio && | |
511 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
512 | goto next_bvec; | |
513 | ||
514 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
515 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
516 | else | |
517 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
518 | next_bvec: | |
519 | new_bio = false; | |
520 | } | |
b21e11c5 ML |
521 | if (likely(bio->bi_iter.bi_size)) { |
522 | bvprv = bvec; | |
523 | new_bio = true; | |
524 | } | |
f6970f83 | 525 | } |
d6d48196 | 526 | |
5cb8850c KO |
527 | return nsegs; |
528 | } | |
529 | ||
530 | /* | |
531 | * map a request to scatterlist, return number of sg entries setup. Caller | |
532 | * must make sure sg can hold rq->nr_phys_segments entries | |
533 | */ | |
534 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, | |
535 | struct scatterlist *sglist) | |
536 | { | |
537 | struct scatterlist *sg = NULL; | |
538 | int nsegs = 0; | |
539 | ||
f9d03f96 | 540 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
cae6c2e5 | 541 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, &sg); |
f9d03f96 | 542 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) |
cae6c2e5 | 543 | nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, &sg); |
f9d03f96 | 544 | else if (rq->bio) |
5cb8850c | 545 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg); |
f18573ab | 546 | |
e8064021 | 547 | if (unlikely(rq->rq_flags & RQF_COPY_USER) && |
2e46e8b2 TH |
548 | (blk_rq_bytes(rq) & q->dma_pad_mask)) { |
549 | unsigned int pad_len = | |
550 | (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1; | |
f18573ab FT |
551 | |
552 | sg->length += pad_len; | |
553 | rq->extra_len += pad_len; | |
554 | } | |
555 | ||
2fb98e84 | 556 | if (q->dma_drain_size && q->dma_drain_needed(rq)) { |
a8ebb056 | 557 | if (op_is_write(req_op(rq))) |
db0a2e00 TH |
558 | memset(q->dma_drain_buffer, 0, q->dma_drain_size); |
559 | ||
da81ed16 | 560 | sg_unmark_end(sg); |
d6d48196 JA |
561 | sg = sg_next(sg); |
562 | sg_set_page(sg, virt_to_page(q->dma_drain_buffer), | |
563 | q->dma_drain_size, | |
564 | ((unsigned long)q->dma_drain_buffer) & | |
565 | (PAGE_SIZE - 1)); | |
566 | nsegs++; | |
7a85f889 | 567 | rq->extra_len += q->dma_drain_size; |
d6d48196 JA |
568 | } |
569 | ||
570 | if (sg) | |
571 | sg_mark_end(sg); | |
572 | ||
12e57f59 ML |
573 | /* |
574 | * Something must have been wrong if the figured number of | |
575 | * segment is bigger than number of req's physical segments | |
576 | */ | |
f9d03f96 | 577 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 578 | |
d6d48196 JA |
579 | return nsegs; |
580 | } | |
d6d48196 JA |
581 | EXPORT_SYMBOL(blk_rq_map_sg); |
582 | ||
d6d48196 JA |
583 | static inline int ll_new_hw_segment(struct request_queue *q, |
584 | struct request *req, | |
585 | struct bio *bio) | |
586 | { | |
d6d48196 JA |
587 | int nr_phys_segs = bio_phys_segments(q, bio); |
588 | ||
13f05c8d MP |
589 | if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q)) |
590 | goto no_merge; | |
591 | ||
4eaf99be | 592 | if (blk_integrity_merge_bio(q, req, bio) == false) |
13f05c8d | 593 | goto no_merge; |
d6d48196 JA |
594 | |
595 | /* | |
596 | * This will form the start of a new hw segment. Bump both | |
597 | * counters. | |
598 | */ | |
d6d48196 JA |
599 | req->nr_phys_segments += nr_phys_segs; |
600 | return 1; | |
13f05c8d MP |
601 | |
602 | no_merge: | |
e0c72300 | 603 | req_set_nomerge(q, req); |
13f05c8d | 604 | return 0; |
d6d48196 JA |
605 | } |
606 | ||
607 | int ll_back_merge_fn(struct request_queue *q, struct request *req, | |
608 | struct bio *bio) | |
609 | { | |
5e7c4274 JA |
610 | if (req_gap_back_merge(req, bio)) |
611 | return 0; | |
7f39add3 SG |
612 | if (blk_integrity_rq(req) && |
613 | integrity_req_gap_back_merge(req, bio)) | |
614 | return 0; | |
f31dc1cd | 615 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 616 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
e0c72300 | 617 | req_set_nomerge(q, req); |
d6d48196 JA |
618 | return 0; |
619 | } | |
2cdf79ca | 620 | if (!bio_flagged(req->biotail, BIO_SEG_VALID)) |
d6d48196 | 621 | blk_recount_segments(q, req->biotail); |
2cdf79ca | 622 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 623 | blk_recount_segments(q, bio); |
d6d48196 JA |
624 | |
625 | return ll_new_hw_segment(q, req, bio); | |
626 | } | |
627 | ||
6728cb0e | 628 | int ll_front_merge_fn(struct request_queue *q, struct request *req, |
d6d48196 JA |
629 | struct bio *bio) |
630 | { | |
5e7c4274 JA |
631 | |
632 | if (req_gap_front_merge(req, bio)) | |
633 | return 0; | |
7f39add3 SG |
634 | if (blk_integrity_rq(req) && |
635 | integrity_req_gap_front_merge(req, bio)) | |
636 | return 0; | |
f31dc1cd | 637 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 638 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
e0c72300 | 639 | req_set_nomerge(q, req); |
d6d48196 JA |
640 | return 0; |
641 | } | |
2cdf79ca | 642 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 643 | blk_recount_segments(q, bio); |
2cdf79ca | 644 | if (!bio_flagged(req->bio, BIO_SEG_VALID)) |
d6d48196 | 645 | blk_recount_segments(q, req->bio); |
d6d48196 JA |
646 | |
647 | return ll_new_hw_segment(q, req, bio); | |
648 | } | |
649 | ||
445251d0 JA |
650 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
651 | struct request *next) | |
652 | { | |
653 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
654 | ||
655 | if (segments >= queue_max_discard_segments(q)) | |
656 | goto no_merge; | |
657 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
658 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
659 | goto no_merge; | |
660 | ||
661 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
662 | return true; | |
663 | no_merge: | |
664 | req_set_nomerge(q, req); | |
665 | return false; | |
666 | } | |
667 | ||
d6d48196 JA |
668 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
669 | struct request *next) | |
670 | { | |
671 | int total_phys_segments; | |
86771427 FT |
672 | unsigned int seg_size = |
673 | req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size; | |
d6d48196 | 674 | |
5e7c4274 | 675 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
676 | return 0; |
677 | ||
d6d48196 JA |
678 | /* |
679 | * Will it become too large? | |
680 | */ | |
f31dc1cd | 681 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 682 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
683 | return 0; |
684 | ||
685 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
86771427 FT |
686 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) { |
687 | if (req->nr_phys_segments == 1) | |
688 | req->bio->bi_seg_front_size = seg_size; | |
689 | if (next->nr_phys_segments == 1) | |
690 | next->biotail->bi_seg_back_size = seg_size; | |
86771427 | 691 | } |
d6d48196 | 692 | |
8a78362c | 693 | if (total_phys_segments > queue_max_segments(q)) |
d6d48196 JA |
694 | return 0; |
695 | ||
4eaf99be | 696 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
697 | return 0; |
698 | ||
d6d48196 JA |
699 | /* Merge is OK... */ |
700 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
701 | return 1; |
702 | } | |
703 | ||
80a761fd TH |
704 | /** |
705 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
706 | * @rq: request to mark as mixed merge | |
707 | * | |
708 | * Description: | |
709 | * @rq is about to be mixed merged. Make sure the attributes | |
710 | * which can be mixed are set in each bio and mark @rq as mixed | |
711 | * merged. | |
712 | */ | |
713 | void blk_rq_set_mixed_merge(struct request *rq) | |
714 | { | |
715 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
716 | struct bio *bio; | |
717 | ||
e8064021 | 718 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
719 | return; |
720 | ||
721 | /* | |
722 | * @rq will no longer represent mixable attributes for all the | |
723 | * contained bios. It will just track those of the first one. | |
724 | * Distributes the attributs to each bio. | |
725 | */ | |
726 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
727 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
728 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
729 | bio->bi_opf |= ff; | |
80a761fd | 730 | } |
e8064021 | 731 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
732 | } |
733 | ||
26308eab JM |
734 | static void blk_account_io_merge(struct request *req) |
735 | { | |
736 | if (blk_do_io_stat(req)) { | |
737 | struct hd_struct *part; | |
26308eab | 738 | |
112f158f | 739 | part_stat_lock(); |
09e099d4 | 740 | part = req->part; |
26308eab | 741 | |
d62e26b3 | 742 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
26308eab | 743 | |
6c23a968 | 744 | hd_struct_put(part); |
26308eab JM |
745 | part_stat_unlock(); |
746 | } | |
747 | } | |
69840466 JW |
748 | /* |
749 | * Two cases of handling DISCARD merge: | |
750 | * If max_discard_segments > 1, the driver takes every bio | |
751 | * as a range and send them to controller together. The ranges | |
752 | * needn't to be contiguous. | |
753 | * Otherwise, the bios/requests will be handled as same as | |
754 | * others which should be contiguous. | |
755 | */ | |
756 | static inline bool blk_discard_mergable(struct request *req) | |
757 | { | |
758 | if (req_op(req) == REQ_OP_DISCARD && | |
759 | queue_max_discard_segments(req->q) > 1) | |
760 | return true; | |
761 | return false; | |
762 | } | |
763 | ||
e96c0d83 EB |
764 | static enum elv_merge blk_try_req_merge(struct request *req, |
765 | struct request *next) | |
69840466 JW |
766 | { |
767 | if (blk_discard_mergable(req)) | |
768 | return ELEVATOR_DISCARD_MERGE; | |
769 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
770 | return ELEVATOR_BACK_MERGE; | |
771 | ||
772 | return ELEVATOR_NO_MERGE; | |
773 | } | |
26308eab | 774 | |
d6d48196 | 775 | /* |
b973cb7e JA |
776 | * For non-mq, this has to be called with the request spinlock acquired. |
777 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 778 | */ |
b973cb7e JA |
779 | static struct request *attempt_merge(struct request_queue *q, |
780 | struct request *req, struct request *next) | |
d6d48196 JA |
781 | { |
782 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 783 | return NULL; |
d6d48196 | 784 | |
288dab8a | 785 | if (req_op(req) != req_op(next)) |
b973cb7e | 786 | return NULL; |
f31dc1cd | 787 | |
d6d48196 | 788 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 789 | || req->rq_disk != next->rq_disk) |
b973cb7e | 790 | return NULL; |
d6d48196 | 791 | |
8fe0d473 | 792 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 793 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 794 | return NULL; |
4363ac7c | 795 | |
cb6934f8 JA |
796 | /* |
797 | * Don't allow merge of different write hints, or for a hint with | |
798 | * non-hint IO. | |
799 | */ | |
800 | if (req->write_hint != next->write_hint) | |
801 | return NULL; | |
802 | ||
668ffc03 DLM |
803 | if (req->ioprio != next->ioprio) |
804 | return NULL; | |
805 | ||
d6d48196 JA |
806 | /* |
807 | * If we are allowed to merge, then append bio list | |
808 | * from next to rq and release next. merge_requests_fn | |
809 | * will have updated segment counts, update sector | |
445251d0 JA |
810 | * counts here. Handle DISCARDs separately, as they |
811 | * have separate settings. | |
d6d48196 | 812 | */ |
69840466 JW |
813 | |
814 | switch (blk_try_req_merge(req, next)) { | |
815 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
816 | if (!req_attempt_discard_merge(q, req, next)) |
817 | return NULL; | |
69840466 JW |
818 | break; |
819 | case ELEVATOR_BACK_MERGE: | |
820 | if (!ll_merge_requests_fn(q, req, next)) | |
821 | return NULL; | |
822 | break; | |
823 | default: | |
b973cb7e | 824 | return NULL; |
69840466 | 825 | } |
d6d48196 | 826 | |
80a761fd TH |
827 | /* |
828 | * If failfast settings disagree or any of the two is already | |
829 | * a mixed merge, mark both as mixed before proceeding. This | |
830 | * makes sure that all involved bios have mixable attributes | |
831 | * set properly. | |
832 | */ | |
e8064021 | 833 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
834 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
835 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
836 | blk_rq_set_mixed_merge(req); | |
837 | blk_rq_set_mixed_merge(next); | |
838 | } | |
839 | ||
d6d48196 | 840 | /* |
522a7775 OS |
841 | * At this point we have either done a back merge or front merge. We |
842 | * need the smaller start_time_ns of the merged requests to be the | |
843 | * current request for accounting purposes. | |
d6d48196 | 844 | */ |
522a7775 OS |
845 | if (next->start_time_ns < req->start_time_ns) |
846 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
847 | |
848 | req->biotail->bi_next = next->bio; | |
849 | req->biotail = next->biotail; | |
850 | ||
a2dec7b3 | 851 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 852 | |
2a5cf35c | 853 | if (!blk_discard_mergable(req)) |
445251d0 | 854 | elv_merge_requests(q, req, next); |
d6d48196 | 855 | |
42dad764 JM |
856 | /* |
857 | * 'next' is going away, so update stats accordingly | |
858 | */ | |
859 | blk_account_io_merge(next); | |
d6d48196 | 860 | |
e4d750c9 JA |
861 | /* |
862 | * ownership of bio passed from next to req, return 'next' for | |
863 | * the caller to free | |
864 | */ | |
1cd96c24 | 865 | next->bio = NULL; |
b973cb7e | 866 | return next; |
d6d48196 JA |
867 | } |
868 | ||
b973cb7e | 869 | struct request *attempt_back_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
870 | { |
871 | struct request *next = elv_latter_request(q, rq); | |
872 | ||
873 | if (next) | |
874 | return attempt_merge(q, rq, next); | |
875 | ||
b973cb7e | 876 | return NULL; |
d6d48196 JA |
877 | } |
878 | ||
b973cb7e | 879 | struct request *attempt_front_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
880 | { |
881 | struct request *prev = elv_former_request(q, rq); | |
882 | ||
883 | if (prev) | |
884 | return attempt_merge(q, prev, rq); | |
885 | ||
b973cb7e | 886 | return NULL; |
d6d48196 | 887 | } |
5e84ea3a JA |
888 | |
889 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
890 | struct request *next) | |
891 | { | |
e4d750c9 | 892 | struct request *free; |
72ef799b | 893 | |
e4d750c9 JA |
894 | free = attempt_merge(q, rq, next); |
895 | if (free) { | |
92bc5a24 | 896 | blk_put_request(free); |
e4d750c9 JA |
897 | return 1; |
898 | } | |
899 | ||
900 | return 0; | |
5e84ea3a | 901 | } |
050c8ea8 TH |
902 | |
903 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
904 | { | |
e2a60da7 | 905 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
906 | return false; |
907 | ||
288dab8a | 908 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
909 | return false; |
910 | ||
050c8ea8 TH |
911 | /* different data direction or already started, don't merge */ |
912 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
913 | return false; | |
914 | ||
2081a56b JA |
915 | /* must be same device */ |
916 | if (rq->rq_disk != bio->bi_disk) | |
050c8ea8 TH |
917 | return false; |
918 | ||
919 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 920 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
921 | return false; |
922 | ||
4363ac7c | 923 | /* must be using the same buffer */ |
8fe0d473 | 924 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
925 | !blk_write_same_mergeable(rq->bio, bio)) |
926 | return false; | |
927 | ||
cb6934f8 JA |
928 | /* |
929 | * Don't allow merge of different write hints, or for a hint with | |
930 | * non-hint IO. | |
931 | */ | |
932 | if (rq->write_hint != bio->bi_write_hint) | |
933 | return false; | |
934 | ||
668ffc03 DLM |
935 | if (rq->ioprio != bio_prio(bio)) |
936 | return false; | |
937 | ||
050c8ea8 TH |
938 | return true; |
939 | } | |
940 | ||
34fe7c05 | 941 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 942 | { |
69840466 | 943 | if (blk_discard_mergable(rq)) |
1e739730 CH |
944 | return ELEVATOR_DISCARD_MERGE; |
945 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 946 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 947 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
948 | return ELEVATOR_FRONT_MERGE; |
949 | return ELEVATOR_NO_MERGE; | |
950 | } |