]>
Commit | Line | Data |
---|---|---|
d6d48196 JA |
1 | /* |
2 | * Functions related to segment and merge handling | |
3 | */ | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/bio.h> | |
7 | #include <linux/blkdev.h> | |
8 | #include <linux/scatterlist.h> | |
9 | ||
cda22646 MK |
10 | #include <trace/events/block.h> |
11 | ||
d6d48196 JA |
12 | #include "blk.h" |
13 | ||
54efd50b KO |
14 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
15 | struct bio *bio, | |
bdced438 ML |
16 | struct bio_set *bs, |
17 | unsigned *nsegs) | |
54efd50b KO |
18 | { |
19 | unsigned int max_discard_sectors, granularity; | |
20 | int alignment; | |
21 | sector_t tmp; | |
22 | unsigned split_sectors; | |
23 | ||
bdced438 ML |
24 | *nsegs = 1; |
25 | ||
54efd50b KO |
26 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
27 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
28 | ||
29 | max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9); | |
30 | max_discard_sectors -= max_discard_sectors % granularity; | |
31 | ||
32 | if (unlikely(!max_discard_sectors)) { | |
33 | /* XXX: warn */ | |
34 | return NULL; | |
35 | } | |
36 | ||
37 | if (bio_sectors(bio) <= max_discard_sectors) | |
38 | return NULL; | |
39 | ||
40 | split_sectors = max_discard_sectors; | |
41 | ||
42 | /* | |
43 | * If the next starting sector would be misaligned, stop the discard at | |
44 | * the previous aligned sector. | |
45 | */ | |
46 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
47 | ||
48 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
49 | tmp = sector_div(tmp, granularity); | |
50 | ||
51 | if (split_sectors > tmp) | |
52 | split_sectors -= tmp; | |
53 | ||
54 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
55 | } | |
56 | ||
57 | static struct bio *blk_bio_write_same_split(struct request_queue *q, | |
58 | struct bio *bio, | |
bdced438 ML |
59 | struct bio_set *bs, |
60 | unsigned *nsegs) | |
54efd50b | 61 | { |
bdced438 ML |
62 | *nsegs = 1; |
63 | ||
54efd50b KO |
64 | if (!q->limits.max_write_same_sectors) |
65 | return NULL; | |
66 | ||
67 | if (bio_sectors(bio) <= q->limits.max_write_same_sectors) | |
68 | return NULL; | |
69 | ||
70 | return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs); | |
71 | } | |
72 | ||
d0e5fbb0 ML |
73 | static inline unsigned get_max_io_size(struct request_queue *q, |
74 | struct bio *bio) | |
75 | { | |
76 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); | |
77 | unsigned mask = queue_logical_block_size(q) - 1; | |
78 | ||
79 | /* aligned to logical block size */ | |
80 | sectors &= ~(mask >> 9); | |
81 | ||
82 | return sectors; | |
83 | } | |
84 | ||
54efd50b KO |
85 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
86 | struct bio *bio, | |
bdced438 ML |
87 | struct bio_set *bs, |
88 | unsigned *segs) | |
54efd50b | 89 | { |
5014c311 | 90 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 91 | struct bvec_iter iter; |
8ae12666 | 92 | unsigned seg_size = 0, nsegs = 0, sectors = 0; |
02e70742 ML |
93 | unsigned front_seg_size = bio->bi_seg_front_size; |
94 | bool do_split = true; | |
95 | struct bio *new = NULL; | |
d0e5fbb0 | 96 | const unsigned max_sectors = get_max_io_size(q, bio); |
4d70dca4 | 97 | unsigned bvecs = 0; |
54efd50b | 98 | |
54efd50b | 99 | bio_for_each_segment(bv, bio, iter) { |
4d70dca4 ML |
100 | /* |
101 | * With arbitrary bio size, the incoming bio may be very | |
102 | * big. We have to split the bio into small bios so that | |
103 | * each holds at most BIO_MAX_PAGES bvecs because | |
104 | * bio_clone() can fail to allocate big bvecs. | |
105 | * | |
106 | * It should have been better to apply the limit per | |
107 | * request queue in which bio_clone() is involved, | |
108 | * instead of globally. The biggest blocker is the | |
109 | * bio_clone() in bio bounce. | |
110 | * | |
111 | * If bio is splitted by this reason, we should have | |
112 | * allowed to continue bios merging, but don't do | |
113 | * that now for making the change simple. | |
114 | * | |
115 | * TODO: deal with bio bounce's bio_clone() gracefully | |
116 | * and convert the global limit into per-queue limit. | |
117 | */ | |
118 | if (bvecs++ >= BIO_MAX_PAGES) | |
119 | goto split; | |
120 | ||
54efd50b KO |
121 | /* |
122 | * If the queue doesn't support SG gaps and adding this | |
123 | * offset would create a gap, disallow it. | |
124 | */ | |
5014c311 | 125 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
126 | goto split; |
127 | ||
d0e5fbb0 | 128 | if (sectors + (bv.bv_len >> 9) > max_sectors) { |
e36f6204 KB |
129 | /* |
130 | * Consider this a new segment if we're splitting in | |
131 | * the middle of this vector. | |
132 | */ | |
133 | if (nsegs < queue_max_segments(q) && | |
d0e5fbb0 | 134 | sectors < max_sectors) { |
e36f6204 | 135 | nsegs++; |
d0e5fbb0 | 136 | sectors = max_sectors; |
e36f6204 | 137 | } |
d0e5fbb0 ML |
138 | if (sectors) |
139 | goto split; | |
140 | /* Make this single bvec as the 1st segment */ | |
e36f6204 KB |
141 | } |
142 | ||
5014c311 | 143 | if (bvprvp && blk_queue_cluster(q)) { |
54efd50b KO |
144 | if (seg_size + bv.bv_len > queue_max_segment_size(q)) |
145 | goto new_segment; | |
5014c311 | 146 | if (!BIOVEC_PHYS_MERGEABLE(bvprvp, &bv)) |
54efd50b | 147 | goto new_segment; |
5014c311 | 148 | if (!BIOVEC_SEG_BOUNDARY(q, bvprvp, &bv)) |
54efd50b KO |
149 | goto new_segment; |
150 | ||
151 | seg_size += bv.bv_len; | |
152 | bvprv = bv; | |
578270bf | 153 | bvprvp = &bvprv; |
52cc6eea | 154 | sectors += bv.bv_len >> 9; |
a88d32af ML |
155 | |
156 | if (nsegs == 1 && seg_size > front_seg_size) | |
157 | front_seg_size = seg_size; | |
54efd50b KO |
158 | continue; |
159 | } | |
160 | new_segment: | |
161 | if (nsegs == queue_max_segments(q)) | |
162 | goto split; | |
163 | ||
164 | nsegs++; | |
165 | bvprv = bv; | |
578270bf | 166 | bvprvp = &bvprv; |
54efd50b | 167 | seg_size = bv.bv_len; |
52cc6eea | 168 | sectors += bv.bv_len >> 9; |
02e70742 ML |
169 | |
170 | if (nsegs == 1 && seg_size > front_seg_size) | |
171 | front_seg_size = seg_size; | |
54efd50b KO |
172 | } |
173 | ||
02e70742 | 174 | do_split = false; |
54efd50b | 175 | split: |
bdced438 | 176 | *segs = nsegs; |
02e70742 ML |
177 | |
178 | if (do_split) { | |
179 | new = bio_split(bio, sectors, GFP_NOIO, bs); | |
180 | if (new) | |
181 | bio = new; | |
182 | } | |
183 | ||
184 | bio->bi_seg_front_size = front_seg_size; | |
185 | if (seg_size > bio->bi_seg_back_size) | |
186 | bio->bi_seg_back_size = seg_size; | |
187 | ||
188 | return do_split ? new : NULL; | |
54efd50b KO |
189 | } |
190 | ||
191 | void blk_queue_split(struct request_queue *q, struct bio **bio, | |
192 | struct bio_set *bs) | |
193 | { | |
bdced438 ML |
194 | struct bio *split, *res; |
195 | unsigned nsegs; | |
54efd50b | 196 | |
7afafc8a AH |
197 | switch (bio_op(*bio)) { |
198 | case REQ_OP_DISCARD: | |
199 | case REQ_OP_SECURE_ERASE: | |
bdced438 | 200 | split = blk_bio_discard_split(q, *bio, bs, &nsegs); |
7afafc8a | 201 | break; |
a6f0788e CK |
202 | case REQ_OP_WRITE_ZEROES: |
203 | split = NULL; | |
204 | nsegs = (*bio)->bi_phys_segments; | |
205 | break; | |
7afafc8a | 206 | case REQ_OP_WRITE_SAME: |
bdced438 | 207 | split = blk_bio_write_same_split(q, *bio, bs, &nsegs); |
7afafc8a AH |
208 | break; |
209 | default: | |
bdced438 | 210 | split = blk_bio_segment_split(q, *bio, q->bio_split, &nsegs); |
7afafc8a AH |
211 | break; |
212 | } | |
bdced438 ML |
213 | |
214 | /* physical segments can be figured out during splitting */ | |
215 | res = split ? split : *bio; | |
216 | res->bi_phys_segments = nsegs; | |
217 | bio_set_flag(res, BIO_SEG_VALID); | |
54efd50b KO |
218 | |
219 | if (split) { | |
6ac45aeb | 220 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 221 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 222 | |
54efd50b | 223 | bio_chain(split, *bio); |
cda22646 | 224 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
54efd50b KO |
225 | generic_make_request(*bio); |
226 | *bio = split; | |
227 | } | |
228 | } | |
229 | EXPORT_SYMBOL(blk_queue_split); | |
230 | ||
1e428079 | 231 | static unsigned int __blk_recalc_rq_segments(struct request_queue *q, |
07388549 ML |
232 | struct bio *bio, |
233 | bool no_sg_merge) | |
d6d48196 | 234 | { |
7988613b | 235 | struct bio_vec bv, bvprv = { NULL }; |
54efd50b | 236 | int cluster, prev = 0; |
1e428079 | 237 | unsigned int seg_size, nr_phys_segs; |
59247eae | 238 | struct bio *fbio, *bbio; |
7988613b | 239 | struct bvec_iter iter; |
d6d48196 | 240 | |
1e428079 JA |
241 | if (!bio) |
242 | return 0; | |
d6d48196 | 243 | |
5cb8850c KO |
244 | /* |
245 | * This should probably be returning 0, but blk_add_request_payload() | |
246 | * (Christoph!!!!) | |
247 | */ | |
a6f0788e CK |
248 | switch (bio_op(bio)) { |
249 | case REQ_OP_DISCARD: | |
250 | case REQ_OP_SECURE_ERASE: | |
251 | case REQ_OP_WRITE_SAME: | |
252 | case REQ_OP_WRITE_ZEROES: | |
5cb8850c | 253 | return 1; |
a6f0788e CK |
254 | default: |
255 | break; | |
256 | } | |
5cb8850c | 257 | |
1e428079 | 258 | fbio = bio; |
e692cb66 | 259 | cluster = blk_queue_cluster(q); |
5df97b91 | 260 | seg_size = 0; |
2c8919de | 261 | nr_phys_segs = 0; |
1e428079 | 262 | for_each_bio(bio) { |
7988613b | 263 | bio_for_each_segment(bv, bio, iter) { |
05f1dd53 JA |
264 | /* |
265 | * If SG merging is disabled, each bio vector is | |
266 | * a segment | |
267 | */ | |
268 | if (no_sg_merge) | |
269 | goto new_segment; | |
270 | ||
54efd50b | 271 | if (prev && cluster) { |
7988613b | 272 | if (seg_size + bv.bv_len |
ae03bf63 | 273 | > queue_max_segment_size(q)) |
1e428079 | 274 | goto new_segment; |
7988613b | 275 | if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv)) |
1e428079 | 276 | goto new_segment; |
7988613b | 277 | if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv)) |
1e428079 | 278 | goto new_segment; |
d6d48196 | 279 | |
7988613b | 280 | seg_size += bv.bv_len; |
1e428079 JA |
281 | bvprv = bv; |
282 | continue; | |
283 | } | |
d6d48196 | 284 | new_segment: |
1e428079 JA |
285 | if (nr_phys_segs == 1 && seg_size > |
286 | fbio->bi_seg_front_size) | |
287 | fbio->bi_seg_front_size = seg_size; | |
86771427 | 288 | |
1e428079 JA |
289 | nr_phys_segs++; |
290 | bvprv = bv; | |
54efd50b | 291 | prev = 1; |
7988613b | 292 | seg_size = bv.bv_len; |
1e428079 | 293 | } |
59247eae | 294 | bbio = bio; |
d6d48196 JA |
295 | } |
296 | ||
59247eae JA |
297 | if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size) |
298 | fbio->bi_seg_front_size = seg_size; | |
299 | if (seg_size > bbio->bi_seg_back_size) | |
300 | bbio->bi_seg_back_size = seg_size; | |
1e428079 JA |
301 | |
302 | return nr_phys_segs; | |
303 | } | |
304 | ||
305 | void blk_recalc_rq_segments(struct request *rq) | |
306 | { | |
07388549 ML |
307 | bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE, |
308 | &rq->q->queue_flags); | |
309 | ||
310 | rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio, | |
311 | no_sg_merge); | |
d6d48196 JA |
312 | } |
313 | ||
314 | void blk_recount_segments(struct request_queue *q, struct bio *bio) | |
315 | { | |
7f60dcaa ML |
316 | unsigned short seg_cnt; |
317 | ||
318 | /* estimate segment number by bi_vcnt for non-cloned bio */ | |
319 | if (bio_flagged(bio, BIO_CLONED)) | |
320 | seg_cnt = bio_segments(bio); | |
321 | else | |
322 | seg_cnt = bio->bi_vcnt; | |
764f612c | 323 | |
7f60dcaa ML |
324 | if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) && |
325 | (seg_cnt < queue_max_segments(q))) | |
326 | bio->bi_phys_segments = seg_cnt; | |
05f1dd53 JA |
327 | else { |
328 | struct bio *nxt = bio->bi_next; | |
329 | ||
330 | bio->bi_next = NULL; | |
7f60dcaa | 331 | bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false); |
05f1dd53 JA |
332 | bio->bi_next = nxt; |
333 | } | |
1e428079 | 334 | |
b7c44ed9 | 335 | bio_set_flag(bio, BIO_SEG_VALID); |
d6d48196 JA |
336 | } |
337 | EXPORT_SYMBOL(blk_recount_segments); | |
338 | ||
339 | static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, | |
340 | struct bio *nxt) | |
341 | { | |
2b8221e1 | 342 | struct bio_vec end_bv = { NULL }, nxt_bv; |
f619d254 | 343 | |
e692cb66 | 344 | if (!blk_queue_cluster(q)) |
d6d48196 JA |
345 | return 0; |
346 | ||
86771427 | 347 | if (bio->bi_seg_back_size + nxt->bi_seg_front_size > |
ae03bf63 | 348 | queue_max_segment_size(q)) |
d6d48196 JA |
349 | return 0; |
350 | ||
e17fc0a1 DW |
351 | if (!bio_has_data(bio)) |
352 | return 1; | |
353 | ||
e827091c ML |
354 | bio_get_last_bvec(bio, &end_bv); |
355 | bio_get_first_bvec(nxt, &nxt_bv); | |
f619d254 KO |
356 | |
357 | if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv)) | |
e17fc0a1 DW |
358 | return 0; |
359 | ||
d6d48196 | 360 | /* |
e17fc0a1 | 361 | * bio and nxt are contiguous in memory; check if the queue allows |
d6d48196 JA |
362 | * these two to be merged into one |
363 | */ | |
f619d254 | 364 | if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv)) |
d6d48196 JA |
365 | return 1; |
366 | ||
367 | return 0; | |
368 | } | |
369 | ||
7988613b | 370 | static inline void |
963ab9e5 | 371 | __blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec, |
7988613b | 372 | struct scatterlist *sglist, struct bio_vec *bvprv, |
963ab9e5 AH |
373 | struct scatterlist **sg, int *nsegs, int *cluster) |
374 | { | |
375 | ||
376 | int nbytes = bvec->bv_len; | |
377 | ||
7988613b | 378 | if (*sg && *cluster) { |
963ab9e5 AH |
379 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
380 | goto new_segment; | |
381 | ||
7988613b | 382 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) |
963ab9e5 | 383 | goto new_segment; |
7988613b | 384 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) |
963ab9e5 AH |
385 | goto new_segment; |
386 | ||
387 | (*sg)->length += nbytes; | |
388 | } else { | |
389 | new_segment: | |
390 | if (!*sg) | |
391 | *sg = sglist; | |
392 | else { | |
393 | /* | |
394 | * If the driver previously mapped a shorter | |
395 | * list, we could see a termination bit | |
396 | * prematurely unless it fully inits the sg | |
397 | * table on each mapping. We KNOW that there | |
398 | * must be more entries here or the driver | |
399 | * would be buggy, so force clear the | |
400 | * termination bit to avoid doing a full | |
401 | * sg_init_table() in drivers for each command. | |
402 | */ | |
c8164d89 | 403 | sg_unmark_end(*sg); |
963ab9e5 AH |
404 | *sg = sg_next(*sg); |
405 | } | |
406 | ||
407 | sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset); | |
408 | (*nsegs)++; | |
409 | } | |
7988613b | 410 | *bvprv = *bvec; |
963ab9e5 AH |
411 | } |
412 | ||
5cb8850c KO |
413 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
414 | struct scatterlist *sglist, | |
415 | struct scatterlist **sg) | |
d6d48196 | 416 | { |
2b8221e1 | 417 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 418 | struct bvec_iter iter; |
d6d48196 JA |
419 | int nsegs, cluster; |
420 | ||
421 | nsegs = 0; | |
e692cb66 | 422 | cluster = blk_queue_cluster(q); |
d6d48196 | 423 | |
7afafc8a AH |
424 | switch (bio_op(bio)) { |
425 | case REQ_OP_DISCARD: | |
426 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 427 | case REQ_OP_WRITE_ZEROES: |
5cb8850c KO |
428 | /* |
429 | * This is a hack - drivers should be neither modifying the | |
430 | * biovec, nor relying on bi_vcnt - but because of | |
431 | * blk_add_request_payload(), a discard bio may or may not have | |
432 | * a payload we need to set up here (thank you Christoph) and | |
433 | * bi_vcnt is really the only way of telling if we need to. | |
434 | */ | |
7afafc8a AH |
435 | if (!bio->bi_vcnt) |
436 | return 0; | |
437 | /* Fall through */ | |
438 | case REQ_OP_WRITE_SAME: | |
5cb8850c KO |
439 | *sg = sglist; |
440 | bvec = bio_iovec(bio); | |
441 | sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset); | |
442 | return 1; | |
7afafc8a AH |
443 | default: |
444 | break; | |
5cb8850c KO |
445 | } |
446 | ||
447 | for_each_bio(bio) | |
448 | bio_for_each_segment(bvec, bio, iter) | |
449 | __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg, | |
450 | &nsegs, &cluster); | |
d6d48196 | 451 | |
5cb8850c KO |
452 | return nsegs; |
453 | } | |
454 | ||
455 | /* | |
456 | * map a request to scatterlist, return number of sg entries setup. Caller | |
457 | * must make sure sg can hold rq->nr_phys_segments entries | |
458 | */ | |
459 | int blk_rq_map_sg(struct request_queue *q, struct request *rq, | |
460 | struct scatterlist *sglist) | |
461 | { | |
462 | struct scatterlist *sg = NULL; | |
463 | int nsegs = 0; | |
464 | ||
465 | if (rq->bio) | |
466 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg); | |
f18573ab | 467 | |
e8064021 | 468 | if (unlikely(rq->rq_flags & RQF_COPY_USER) && |
2e46e8b2 TH |
469 | (blk_rq_bytes(rq) & q->dma_pad_mask)) { |
470 | unsigned int pad_len = | |
471 | (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1; | |
f18573ab FT |
472 | |
473 | sg->length += pad_len; | |
474 | rq->extra_len += pad_len; | |
475 | } | |
476 | ||
2fb98e84 | 477 | if (q->dma_drain_size && q->dma_drain_needed(rq)) { |
a8ebb056 | 478 | if (op_is_write(req_op(rq))) |
db0a2e00 TH |
479 | memset(q->dma_drain_buffer, 0, q->dma_drain_size); |
480 | ||
da81ed16 | 481 | sg_unmark_end(sg); |
d6d48196 JA |
482 | sg = sg_next(sg); |
483 | sg_set_page(sg, virt_to_page(q->dma_drain_buffer), | |
484 | q->dma_drain_size, | |
485 | ((unsigned long)q->dma_drain_buffer) & | |
486 | (PAGE_SIZE - 1)); | |
487 | nsegs++; | |
7a85f889 | 488 | rq->extra_len += q->dma_drain_size; |
d6d48196 JA |
489 | } |
490 | ||
491 | if (sg) | |
492 | sg_mark_end(sg); | |
493 | ||
12e57f59 ML |
494 | /* |
495 | * Something must have been wrong if the figured number of | |
496 | * segment is bigger than number of req's physical segments | |
497 | */ | |
498 | WARN_ON(nsegs > rq->nr_phys_segments); | |
499 | ||
d6d48196 JA |
500 | return nsegs; |
501 | } | |
d6d48196 JA |
502 | EXPORT_SYMBOL(blk_rq_map_sg); |
503 | ||
e0c72300 RH |
504 | static void req_set_nomerge(struct request_queue *q, struct request *req) |
505 | { | |
506 | req->cmd_flags |= REQ_NOMERGE; | |
507 | if (req == q->last_merge) | |
508 | q->last_merge = NULL; | |
509 | } | |
510 | ||
d6d48196 JA |
511 | static inline int ll_new_hw_segment(struct request_queue *q, |
512 | struct request *req, | |
513 | struct bio *bio) | |
514 | { | |
d6d48196 JA |
515 | int nr_phys_segs = bio_phys_segments(q, bio); |
516 | ||
13f05c8d MP |
517 | if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q)) |
518 | goto no_merge; | |
519 | ||
4eaf99be | 520 | if (blk_integrity_merge_bio(q, req, bio) == false) |
13f05c8d | 521 | goto no_merge; |
d6d48196 JA |
522 | |
523 | /* | |
524 | * This will form the start of a new hw segment. Bump both | |
525 | * counters. | |
526 | */ | |
d6d48196 JA |
527 | req->nr_phys_segments += nr_phys_segs; |
528 | return 1; | |
13f05c8d MP |
529 | |
530 | no_merge: | |
e0c72300 | 531 | req_set_nomerge(q, req); |
13f05c8d | 532 | return 0; |
d6d48196 JA |
533 | } |
534 | ||
535 | int ll_back_merge_fn(struct request_queue *q, struct request *req, | |
536 | struct bio *bio) | |
537 | { | |
5e7c4274 JA |
538 | if (req_gap_back_merge(req, bio)) |
539 | return 0; | |
7f39add3 SG |
540 | if (blk_integrity_rq(req) && |
541 | integrity_req_gap_back_merge(req, bio)) | |
542 | return 0; | |
f31dc1cd | 543 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 544 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
e0c72300 | 545 | req_set_nomerge(q, req); |
d6d48196 JA |
546 | return 0; |
547 | } | |
2cdf79ca | 548 | if (!bio_flagged(req->biotail, BIO_SEG_VALID)) |
d6d48196 | 549 | blk_recount_segments(q, req->biotail); |
2cdf79ca | 550 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 551 | blk_recount_segments(q, bio); |
d6d48196 JA |
552 | |
553 | return ll_new_hw_segment(q, req, bio); | |
554 | } | |
555 | ||
6728cb0e | 556 | int ll_front_merge_fn(struct request_queue *q, struct request *req, |
d6d48196 JA |
557 | struct bio *bio) |
558 | { | |
5e7c4274 JA |
559 | |
560 | if (req_gap_front_merge(req, bio)) | |
561 | return 0; | |
7f39add3 SG |
562 | if (blk_integrity_rq(req) && |
563 | integrity_req_gap_front_merge(req, bio)) | |
564 | return 0; | |
f31dc1cd | 565 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 566 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
e0c72300 | 567 | req_set_nomerge(q, req); |
d6d48196 JA |
568 | return 0; |
569 | } | |
2cdf79ca | 570 | if (!bio_flagged(bio, BIO_SEG_VALID)) |
d6d48196 | 571 | blk_recount_segments(q, bio); |
2cdf79ca | 572 | if (!bio_flagged(req->bio, BIO_SEG_VALID)) |
d6d48196 | 573 | blk_recount_segments(q, req->bio); |
d6d48196 JA |
574 | |
575 | return ll_new_hw_segment(q, req, bio); | |
576 | } | |
577 | ||
e7e24500 JA |
578 | /* |
579 | * blk-mq uses req->special to carry normal driver per-request payload, it | |
580 | * does not indicate a prepared command that we cannot merge with. | |
581 | */ | |
582 | static bool req_no_special_merge(struct request *req) | |
583 | { | |
584 | struct request_queue *q = req->q; | |
585 | ||
586 | return !q->mq_ops && req->special; | |
587 | } | |
588 | ||
d6d48196 JA |
589 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
590 | struct request *next) | |
591 | { | |
592 | int total_phys_segments; | |
86771427 FT |
593 | unsigned int seg_size = |
594 | req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size; | |
d6d48196 JA |
595 | |
596 | /* | |
597 | * First check if the either of the requests are re-queued | |
598 | * requests. Can't merge them if they are. | |
599 | */ | |
e7e24500 | 600 | if (req_no_special_merge(req) || req_no_special_merge(next)) |
d6d48196 JA |
601 | return 0; |
602 | ||
5e7c4274 | 603 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
604 | return 0; |
605 | ||
d6d48196 JA |
606 | /* |
607 | * Will it become too large? | |
608 | */ | |
f31dc1cd | 609 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 610 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
611 | return 0; |
612 | ||
613 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
86771427 FT |
614 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) { |
615 | if (req->nr_phys_segments == 1) | |
616 | req->bio->bi_seg_front_size = seg_size; | |
617 | if (next->nr_phys_segments == 1) | |
618 | next->biotail->bi_seg_back_size = seg_size; | |
d6d48196 | 619 | total_phys_segments--; |
86771427 | 620 | } |
d6d48196 | 621 | |
8a78362c | 622 | if (total_phys_segments > queue_max_segments(q)) |
d6d48196 JA |
623 | return 0; |
624 | ||
4eaf99be | 625 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
626 | return 0; |
627 | ||
d6d48196 JA |
628 | /* Merge is OK... */ |
629 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
630 | return 1; |
631 | } | |
632 | ||
80a761fd TH |
633 | /** |
634 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
635 | * @rq: request to mark as mixed merge | |
636 | * | |
637 | * Description: | |
638 | * @rq is about to be mixed merged. Make sure the attributes | |
639 | * which can be mixed are set in each bio and mark @rq as mixed | |
640 | * merged. | |
641 | */ | |
642 | void blk_rq_set_mixed_merge(struct request *rq) | |
643 | { | |
644 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
645 | struct bio *bio; | |
646 | ||
e8064021 | 647 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
648 | return; |
649 | ||
650 | /* | |
651 | * @rq will no longer represent mixable attributes for all the | |
652 | * contained bios. It will just track those of the first one. | |
653 | * Distributes the attributs to each bio. | |
654 | */ | |
655 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
656 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
657 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
658 | bio->bi_opf |= ff; | |
80a761fd | 659 | } |
e8064021 | 660 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
661 | } |
662 | ||
26308eab JM |
663 | static void blk_account_io_merge(struct request *req) |
664 | { | |
665 | if (blk_do_io_stat(req)) { | |
666 | struct hd_struct *part; | |
667 | int cpu; | |
668 | ||
669 | cpu = part_stat_lock(); | |
09e099d4 | 670 | part = req->part; |
26308eab JM |
671 | |
672 | part_round_stats(cpu, part); | |
316d315b | 673 | part_dec_in_flight(part, rq_data_dir(req)); |
26308eab | 674 | |
6c23a968 | 675 | hd_struct_put(part); |
26308eab JM |
676 | part_stat_unlock(); |
677 | } | |
678 | } | |
679 | ||
d6d48196 JA |
680 | /* |
681 | * Has to be called with the request spinlock acquired | |
682 | */ | |
683 | static int attempt_merge(struct request_queue *q, struct request *req, | |
684 | struct request *next) | |
685 | { | |
686 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
687 | return 0; | |
688 | ||
288dab8a | 689 | if (req_op(req) != req_op(next)) |
f31dc1cd MP |
690 | return 0; |
691 | ||
d6d48196 JA |
692 | /* |
693 | * not contiguous | |
694 | */ | |
83096ebf | 695 | if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next)) |
d6d48196 JA |
696 | return 0; |
697 | ||
698 | if (rq_data_dir(req) != rq_data_dir(next) | |
699 | || req->rq_disk != next->rq_disk | |
e7e24500 | 700 | || req_no_special_merge(next)) |
d6d48196 JA |
701 | return 0; |
702 | ||
8fe0d473 | 703 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
704 | !blk_write_same_mergeable(req->bio, next->bio)) |
705 | return 0; | |
706 | ||
d6d48196 JA |
707 | /* |
708 | * If we are allowed to merge, then append bio list | |
709 | * from next to rq and release next. merge_requests_fn | |
710 | * will have updated segment counts, update sector | |
711 | * counts here. | |
712 | */ | |
713 | if (!ll_merge_requests_fn(q, req, next)) | |
714 | return 0; | |
715 | ||
80a761fd TH |
716 | /* |
717 | * If failfast settings disagree or any of the two is already | |
718 | * a mixed merge, mark both as mixed before proceeding. This | |
719 | * makes sure that all involved bios have mixable attributes | |
720 | * set properly. | |
721 | */ | |
e8064021 | 722 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
723 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
724 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
725 | blk_rq_set_mixed_merge(req); | |
726 | blk_rq_set_mixed_merge(next); | |
727 | } | |
728 | ||
d6d48196 JA |
729 | /* |
730 | * At this point we have either done a back merge | |
731 | * or front merge. We need the smaller start_time of | |
732 | * the merged requests to be the current request | |
733 | * for accounting purposes. | |
734 | */ | |
735 | if (time_after(req->start_time, next->start_time)) | |
736 | req->start_time = next->start_time; | |
737 | ||
738 | req->biotail->bi_next = next->bio; | |
739 | req->biotail = next->biotail; | |
740 | ||
a2dec7b3 | 741 | req->__data_len += blk_rq_bytes(next); |
d6d48196 JA |
742 | |
743 | elv_merge_requests(q, req, next); | |
744 | ||
42dad764 JM |
745 | /* |
746 | * 'next' is going away, so update stats accordingly | |
747 | */ | |
748 | blk_account_io_merge(next); | |
d6d48196 JA |
749 | |
750 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); | |
ab780f1e JA |
751 | if (blk_rq_cpu_valid(next)) |
752 | req->cpu = next->cpu; | |
d6d48196 | 753 | |
1cd96c24 BH |
754 | /* owner-ship of bio passed from next to req */ |
755 | next->bio = NULL; | |
d6d48196 JA |
756 | __blk_put_request(q, next); |
757 | return 1; | |
758 | } | |
759 | ||
760 | int attempt_back_merge(struct request_queue *q, struct request *rq) | |
761 | { | |
762 | struct request *next = elv_latter_request(q, rq); | |
763 | ||
764 | if (next) | |
765 | return attempt_merge(q, rq, next); | |
766 | ||
767 | return 0; | |
768 | } | |
769 | ||
770 | int attempt_front_merge(struct request_queue *q, struct request *rq) | |
771 | { | |
772 | struct request *prev = elv_former_request(q, rq); | |
773 | ||
774 | if (prev) | |
775 | return attempt_merge(q, prev, rq); | |
776 | ||
777 | return 0; | |
778 | } | |
5e84ea3a JA |
779 | |
780 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
781 | struct request *next) | |
782 | { | |
72ef799b TE |
783 | struct elevator_queue *e = q->elevator; |
784 | ||
785 | if (e->type->ops.elevator_allow_rq_merge_fn) | |
786 | if (!e->type->ops.elevator_allow_rq_merge_fn(q, rq, next)) | |
787 | return 0; | |
788 | ||
5e84ea3a JA |
789 | return attempt_merge(q, rq, next); |
790 | } | |
050c8ea8 TH |
791 | |
792 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
793 | { | |
e2a60da7 | 794 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
795 | return false; |
796 | ||
288dab8a | 797 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
798 | return false; |
799 | ||
050c8ea8 TH |
800 | /* different data direction or already started, don't merge */ |
801 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
802 | return false; | |
803 | ||
804 | /* must be same device and not a special request */ | |
e7e24500 | 805 | if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq)) |
050c8ea8 TH |
806 | return false; |
807 | ||
808 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 809 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
810 | return false; |
811 | ||
4363ac7c | 812 | /* must be using the same buffer */ |
8fe0d473 | 813 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
814 | !blk_write_same_mergeable(rq->bio, bio)) |
815 | return false; | |
816 | ||
050c8ea8 TH |
817 | return true; |
818 | } | |
819 | ||
820 | int blk_try_merge(struct request *rq, struct bio *bio) | |
821 | { | |
4f024f37 | 822 | if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) |
050c8ea8 | 823 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 824 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
825 | return ELEVATOR_FRONT_MERGE; |
826 | return ELEVATOR_NO_MERGE; | |
827 | } |