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 | 13 | #include "blk.h" |
8e756373 | 14 | #include "blk-rq-qos.h" |
d6d48196 | 15 | |
e9907009 CH |
16 | static inline bool bio_will_gap(struct request_queue *q, |
17 | struct request *prev_rq, struct bio *prev, struct bio *next) | |
18 | { | |
19 | struct bio_vec pb, nb; | |
20 | ||
21 | if (!bio_has_data(prev) || !queue_virt_boundary(q)) | |
22 | return false; | |
23 | ||
24 | /* | |
25 | * Don't merge if the 1st bio starts with non-zero offset, otherwise it | |
26 | * is quite difficult to respect the sg gap limit. We work hard to | |
27 | * merge a huge number of small single bios in case of mkfs. | |
28 | */ | |
29 | if (prev_rq) | |
30 | bio_get_first_bvec(prev_rq->bio, &pb); | |
31 | else | |
32 | bio_get_first_bvec(prev, &pb); | |
df376b2e | 33 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
34 | return true; |
35 | ||
36 | /* | |
37 | * We don't need to worry about the situation that the merged segment | |
38 | * ends in unaligned virt boundary: | |
39 | * | |
40 | * - if 'pb' ends aligned, the merged segment ends aligned | |
41 | * - if 'pb' ends unaligned, the next bio must include | |
42 | * one single bvec of 'nb', otherwise the 'nb' can't | |
43 | * merge with 'pb' | |
44 | */ | |
45 | bio_get_last_bvec(prev, &pb); | |
46 | bio_get_first_bvec(next, &nb); | |
200a9aff | 47 | if (biovec_phys_mergeable(q, &pb, &nb)) |
e9907009 CH |
48 | return false; |
49 | return __bvec_gap_to_prev(q, &pb, nb.bv_offset); | |
50 | } | |
51 | ||
52 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) | |
53 | { | |
54 | return bio_will_gap(req->q, req, req->biotail, bio); | |
55 | } | |
56 | ||
57 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) | |
58 | { | |
59 | return bio_will_gap(req->q, NULL, bio, req->bio); | |
60 | } | |
61 | ||
54efd50b KO |
62 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
63 | struct bio *bio, | |
bdced438 ML |
64 | struct bio_set *bs, |
65 | unsigned *nsegs) | |
54efd50b KO |
66 | { |
67 | unsigned int max_discard_sectors, granularity; | |
68 | int alignment; | |
69 | sector_t tmp; | |
70 | unsigned split_sectors; | |
71 | ||
bdced438 ML |
72 | *nsegs = 1; |
73 | ||
54efd50b KO |
74 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
75 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
76 | ||
1adfc5e4 ML |
77 | max_discard_sectors = min(q->limits.max_discard_sectors, |
78 | bio_allowed_max_sectors(q)); | |
54efd50b KO |
79 | max_discard_sectors -= max_discard_sectors % granularity; |
80 | ||
81 | if (unlikely(!max_discard_sectors)) { | |
82 | /* XXX: warn */ | |
83 | return NULL; | |
84 | } | |
85 | ||
86 | if (bio_sectors(bio) <= max_discard_sectors) | |
87 | return NULL; | |
88 | ||
89 | split_sectors = max_discard_sectors; | |
90 | ||
91 | /* | |
92 | * If the next starting sector would be misaligned, stop the discard at | |
93 | * the previous aligned sector. | |
94 | */ | |
95 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
96 | ||
97 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
98 | tmp = sector_div(tmp, granularity); | |
99 | ||
100 | if (split_sectors > tmp) | |
101 | split_sectors -= tmp; | |
102 | ||
103 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
104 | } | |
105 | ||
885fa13f CH |
106 | static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, |
107 | struct bio *bio, struct bio_set *bs, unsigned *nsegs) | |
108 | { | |
d665e12a | 109 | *nsegs = 0; |
885fa13f CH |
110 | |
111 | if (!q->limits.max_write_zeroes_sectors) | |
112 | return NULL; | |
113 | ||
114 | if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) | |
115 | return NULL; | |
116 | ||
117 | return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); | |
118 | } | |
119 | ||
54efd50b KO |
120 | static struct bio *blk_bio_write_same_split(struct request_queue *q, |
121 | struct bio *bio, | |
bdced438 ML |
122 | struct bio_set *bs, |
123 | unsigned *nsegs) | |
54efd50b | 124 | { |
bdced438 ML |
125 | *nsegs = 1; |
126 | ||
54efd50b KO |
127 | if (!q->limits.max_write_same_sectors) |
128 | return NULL; | |
129 | ||
130 | if (bio_sectors(bio) <= q->limits.max_write_same_sectors) | |
131 | return NULL; | |
132 | ||
133 | return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs); | |
134 | } | |
135 | ||
9cc5169c BVA |
136 | /* |
137 | * Return the maximum number of sectors from the start of a bio that may be | |
138 | * submitted as a single request to a block device. If enough sectors remain, | |
139 | * align the end to the physical block size. Otherwise align the end to the | |
140 | * logical block size. This approach minimizes the number of non-aligned | |
141 | * requests that are submitted to a block device if the start of a bio is not | |
142 | * aligned to a physical block boundary. | |
143 | */ | |
d0e5fbb0 ML |
144 | static inline unsigned get_max_io_size(struct request_queue *q, |
145 | struct bio *bio) | |
146 | { | |
3ee16db3 | 147 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector, 0); |
9cc5169c BVA |
148 | unsigned max_sectors = sectors; |
149 | unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT; | |
150 | unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT; | |
151 | unsigned start_offset = bio->bi_iter.bi_sector & (pbs - 1); | |
d0e5fbb0 | 152 | |
9cc5169c BVA |
153 | max_sectors += start_offset; |
154 | max_sectors &= ~(pbs - 1); | |
155 | if (max_sectors > start_offset) | |
156 | return max_sectors - start_offset; | |
d0e5fbb0 | 157 | |
e4b469c6 | 158 | return sectors & ~(lbs - 1); |
d0e5fbb0 ML |
159 | } |
160 | ||
429120f3 ML |
161 | static inline unsigned get_max_segment_size(const struct request_queue *q, |
162 | struct page *start_page, | |
163 | unsigned long offset) | |
dcebd755 ML |
164 | { |
165 | unsigned long mask = queue_segment_boundary(q); | |
166 | ||
429120f3 | 167 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
168 | |
169 | /* | |
170 | * overflow may be triggered in case of zero page physical address | |
171 | * on 32bit arch, use queue's max segment size when that happens. | |
172 | */ | |
173 | return min_not_zero(mask - offset + 1, | |
174 | (unsigned long)queue_max_segment_size(q)); | |
dcebd755 ML |
175 | } |
176 | ||
708b25b3 BVA |
177 | /** |
178 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
179 | * @q: [in] request queue associated with the bio associated with @bv | |
180 | * @bv: [in] bvec to examine | |
181 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
182 | * by the number of segments from @bv that may be appended to that | |
183 | * bio without exceeding @max_segs | |
184 | * @sectors: [in,out] Number of sectors in the bio being built. Incremented | |
185 | * by the number of sectors from @bv that may be appended to that | |
186 | * bio without exceeding @max_sectors | |
187 | * @max_segs: [in] upper bound for *@nsegs | |
188 | * @max_sectors: [in] upper bound for *@sectors | |
189 | * | |
190 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
191 | * big to fit in a single segment and hence that it has to be split in the | |
192 | * middle. This function verifies whether or not that should happen. The value | |
193 | * %true is returned if and only if appending the entire @bv to a bio with | |
194 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
195 | * the block driver. | |
dcebd755 | 196 | */ |
af2c68fe BVA |
197 | static bool bvec_split_segs(const struct request_queue *q, |
198 | const struct bio_vec *bv, unsigned *nsegs, | |
708b25b3 BVA |
199 | unsigned *sectors, unsigned max_segs, |
200 | unsigned max_sectors) | |
dcebd755 | 201 | { |
708b25b3 BVA |
202 | unsigned max_len = (min(max_sectors, UINT_MAX >> 9) - *sectors) << 9; |
203 | unsigned len = min(bv->bv_len, max_len); | |
dcebd755 | 204 | unsigned total_len = 0; |
ff9811b3 | 205 | unsigned seg_size = 0; |
dcebd755 | 206 | |
ff9811b3 | 207 | while (len && *nsegs < max_segs) { |
429120f3 ML |
208 | seg_size = get_max_segment_size(q, bv->bv_page, |
209 | bv->bv_offset + total_len); | |
dcebd755 ML |
210 | seg_size = min(seg_size, len); |
211 | ||
ff9811b3 | 212 | (*nsegs)++; |
dcebd755 ML |
213 | total_len += seg_size; |
214 | len -= seg_size; | |
215 | ||
216 | if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) | |
217 | break; | |
218 | } | |
219 | ||
ff9811b3 | 220 | *sectors += total_len >> 9; |
dcebd755 | 221 | |
708b25b3 BVA |
222 | /* tell the caller to split the bvec if it is too big to fit */ |
223 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
224 | } |
225 | ||
dad77584 BVA |
226 | /** |
227 | * blk_bio_segment_split - split a bio in two bios | |
228 | * @q: [in] request queue pointer | |
229 | * @bio: [in] bio to be split | |
230 | * @bs: [in] bio set to allocate the clone from | |
231 | * @segs: [out] number of segments in the bio with the first half of the sectors | |
232 | * | |
233 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
234 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
235 | * following is guaranteed for the cloned bio: | |
236 | * - That it has at most get_max_io_size(@q, @bio) sectors. | |
237 | * - That it has at most queue_max_segments(@q) segments. | |
238 | * | |
239 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
240 | * the original bio. It is the responsibility of the caller to ensure that the | |
241 | * original bio is not freed before the cloned bio. The caller is also | |
242 | * responsible for ensuring that @bs is only destroyed after processing of the | |
243 | * split bio has finished. | |
244 | */ | |
54efd50b KO |
245 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
246 | struct bio *bio, | |
bdced438 ML |
247 | struct bio_set *bs, |
248 | unsigned *segs) | |
54efd50b | 249 | { |
5014c311 | 250 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 251 | struct bvec_iter iter; |
6869875f | 252 | unsigned nsegs = 0, sectors = 0; |
d0e5fbb0 | 253 | const unsigned max_sectors = get_max_io_size(q, bio); |
05b700ba | 254 | const unsigned max_segs = queue_max_segments(q); |
54efd50b | 255 | |
dcebd755 | 256 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
257 | /* |
258 | * If the queue doesn't support SG gaps and adding this | |
259 | * offset would create a gap, disallow it. | |
260 | */ | |
5014c311 | 261 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
262 | goto split; |
263 | ||
708b25b3 BVA |
264 | if (nsegs < max_segs && |
265 | sectors + (bv.bv_len >> 9) <= max_sectors && | |
266 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { | |
267 | nsegs++; | |
268 | sectors += bv.bv_len >> 9; | |
269 | } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs, | |
270 | max_sectors)) { | |
cf8c0c6a | 271 | goto split; |
e36f6204 KB |
272 | } |
273 | ||
54efd50b | 274 | bvprv = bv; |
578270bf | 275 | bvprvp = &bvprv; |
54efd50b KO |
276 | } |
277 | ||
d627065d CH |
278 | *segs = nsegs; |
279 | return NULL; | |
54efd50b | 280 | split: |
bdced438 | 281 | *segs = nsegs; |
cc29e1bf JX |
282 | |
283 | /* | |
284 | * Bio splitting may cause subtle trouble such as hang when doing sync | |
285 | * iopoll in direct IO routine. Given performance gain of iopoll for | |
286 | * big IO can be trival, disable iopoll when split needed. | |
287 | */ | |
288 | bio->bi_opf &= ~REQ_HIPRI; | |
289 | ||
d627065d | 290 | return bio_split(bio, sectors, GFP_NOIO, bs); |
54efd50b KO |
291 | } |
292 | ||
dad77584 BVA |
293 | /** |
294 | * __blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
295 | * @bio: [in, out] bio to be split |
296 | * @nr_segs: [out] number of segments in the first bio | |
297 | * | |
298 | * Split a bio into two bios, chain the two bios, submit the second half and | |
299 | * store a pointer to the first half in *@bio. If the second bio is still too | |
300 | * big it will be split by a recursive call to this function. Since this | |
309dca30 CH |
301 | * function may allocate a new bio from q->bio_split, it is the responsibility |
302 | * of the caller to ensure that q->bio_split is only released after processing | |
303 | * of the split bio has finished. | |
dad77584 | 304 | */ |
f695ca38 | 305 | void __blk_queue_split(struct bio **bio, unsigned int *nr_segs) |
54efd50b | 306 | { |
309dca30 | 307 | struct request_queue *q = (*bio)->bi_bdev->bd_disk->queue; |
fa532287 | 308 | struct bio *split = NULL; |
54efd50b | 309 | |
7afafc8a AH |
310 | switch (bio_op(*bio)) { |
311 | case REQ_OP_DISCARD: | |
312 | case REQ_OP_SECURE_ERASE: | |
14ccb66b | 313 | split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a | 314 | break; |
a6f0788e | 315 | case REQ_OP_WRITE_ZEROES: |
14ccb66b CH |
316 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, |
317 | nr_segs); | |
a6f0788e | 318 | break; |
7afafc8a | 319 | case REQ_OP_WRITE_SAME: |
14ccb66b CH |
320 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, |
321 | nr_segs); | |
7afafc8a AH |
322 | break; |
323 | default: | |
fa532287 CH |
324 | /* |
325 | * All drivers must accept single-segments bios that are <= | |
326 | * PAGE_SIZE. This is a quick and dirty check that relies on | |
327 | * the fact that bi_io_vec[0] is always valid if a bio has data. | |
328 | * The check might lead to occasional false negatives when bios | |
329 | * are cloned, but compared to the performance impact of cloned | |
330 | * bios themselves the loop below doesn't matter anyway. | |
331 | */ | |
332 | if (!q->limits.chunk_sectors && | |
333 | (*bio)->bi_vcnt == 1 && | |
59db8ba2 | 334 | ((*bio)->bi_io_vec[0].bv_len + |
1e279153 | 335 | (*bio)->bi_io_vec[0].bv_offset) <= PAGE_SIZE) { |
fa532287 CH |
336 | *nr_segs = 1; |
337 | break; | |
338 | } | |
14ccb66b | 339 | split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a AH |
340 | break; |
341 | } | |
bdced438 | 342 | |
54efd50b | 343 | if (split) { |
6ac45aeb | 344 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 345 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 346 | |
54efd50b | 347 | bio_chain(split, *bio); |
eb6f7f7c | 348 | trace_block_split(split, (*bio)->bi_iter.bi_sector); |
ed00aabd | 349 | submit_bio_noacct(*bio); |
54efd50b KO |
350 | *bio = split; |
351 | } | |
352 | } | |
14ccb66b | 353 | |
dad77584 BVA |
354 | /** |
355 | * blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
356 | * @bio: [in, out] bio to be split |
357 | * | |
358 | * Split a bio into two bios, chains the two bios, submit the second half and | |
359 | * store a pointer to the first half in *@bio. Since this function may allocate | |
309dca30 CH |
360 | * a new bio from q->bio_split, it is the responsibility of the caller to ensure |
361 | * that q->bio_split is only released after processing of the split bio has | |
362 | * finished. | |
dad77584 | 363 | */ |
f695ca38 | 364 | void blk_queue_split(struct bio **bio) |
14ccb66b CH |
365 | { |
366 | unsigned int nr_segs; | |
367 | ||
f695ca38 | 368 | __blk_queue_split(bio, &nr_segs); |
14ccb66b | 369 | } |
54efd50b KO |
370 | EXPORT_SYMBOL(blk_queue_split); |
371 | ||
e9cd19c0 | 372 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 373 | { |
6869875f | 374 | unsigned int nr_phys_segs = 0; |
ff9811b3 | 375 | unsigned int nr_sectors = 0; |
e9cd19c0 | 376 | struct req_iterator iter; |
6869875f | 377 | struct bio_vec bv; |
d6d48196 | 378 | |
e9cd19c0 | 379 | if (!rq->bio) |
1e428079 | 380 | return 0; |
d6d48196 | 381 | |
e9cd19c0 | 382 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
383 | case REQ_OP_DISCARD: |
384 | case REQ_OP_SECURE_ERASE: | |
a958937f DJ |
385 | if (queue_max_discard_segments(rq->q) > 1) { |
386 | struct bio *bio = rq->bio; | |
387 | ||
388 | for_each_bio(bio) | |
389 | nr_phys_segs++; | |
390 | return nr_phys_segs; | |
391 | } | |
392 | return 1; | |
a6f0788e | 393 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
394 | return 0; |
395 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 396 | return 1; |
a6f0788e | 397 | } |
5cb8850c | 398 | |
e9cd19c0 | 399 | rq_for_each_bvec(bv, rq, iter) |
ff9811b3 | 400 | bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors, |
708b25b3 | 401 | UINT_MAX, UINT_MAX); |
1e428079 JA |
402 | return nr_phys_segs; |
403 | } | |
404 | ||
48d7727c | 405 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
406 | struct scatterlist *sglist) |
407 | { | |
408 | if (!*sg) | |
409 | return sglist; | |
410 | ||
411 | /* | |
412 | * If the driver previously mapped a shorter list, we could see a | |
413 | * termination bit prematurely unless it fully inits the sg table | |
414 | * on each mapping. We KNOW that there must be more entries here | |
415 | * or the driver would be buggy, so force clear the termination bit | |
416 | * to avoid doing a full sg_init_table() in drivers for each command. | |
417 | */ | |
418 | sg_unmark_end(*sg); | |
419 | return sg_next(*sg); | |
420 | } | |
421 | ||
422 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
423 | struct bio_vec *bvec, struct scatterlist *sglist, | |
424 | struct scatterlist **sg) | |
425 | { | |
426 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 427 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
428 | |
429 | while (nbytes > 0) { | |
8a96a0e4 | 430 | unsigned offset = bvec->bv_offset + total; |
429120f3 ML |
431 | unsigned len = min(get_max_segment_size(q, bvec->bv_page, |
432 | offset), nbytes); | |
f9f76879 CH |
433 | struct page *page = bvec->bv_page; |
434 | ||
435 | /* | |
436 | * Unfortunately a fair number of drivers barf on scatterlists | |
437 | * that have an offset larger than PAGE_SIZE, despite other | |
438 | * subsystems dealing with that invariant just fine. For now | |
439 | * stick to the legacy format where we never present those from | |
440 | * the block layer, but the code below should be removed once | |
441 | * these offenders (mostly MMC/SD drivers) are fixed. | |
442 | */ | |
443 | page += (offset >> PAGE_SHIFT); | |
444 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
445 | |
446 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 447 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 448 | |
8a96a0e4 CH |
449 | total += len; |
450 | nbytes -= len; | |
862e5a5e ML |
451 | nsegs++; |
452 | } | |
453 | ||
454 | return nsegs; | |
455 | } | |
456 | ||
16e3e418 ML |
457 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
458 | struct scatterlist *sglist, struct scatterlist **sg) | |
459 | { | |
460 | *sg = blk_next_sg(sg, sglist); | |
461 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
462 | return 1; | |
463 | } | |
464 | ||
f6970f83 ML |
465 | /* only try to merge bvecs into one sg if they are from two bios */ |
466 | static inline bool | |
467 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
468 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
469 | { |
470 | ||
471 | int nbytes = bvec->bv_len; | |
472 | ||
f6970f83 ML |
473 | if (!*sg) |
474 | return false; | |
963ab9e5 | 475 | |
f6970f83 ML |
476 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
477 | return false; | |
478 | ||
479 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
480 | return false; | |
481 | ||
482 | (*sg)->length += nbytes; | |
483 | ||
484 | return true; | |
963ab9e5 AH |
485 | } |
486 | ||
5cb8850c KO |
487 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
488 | struct scatterlist *sglist, | |
489 | struct scatterlist **sg) | |
d6d48196 | 490 | { |
3f649ab7 | 491 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 492 | struct bvec_iter iter; |
38417468 | 493 | int nsegs = 0; |
f6970f83 | 494 | bool new_bio = false; |
5cb8850c | 495 | |
f6970f83 ML |
496 | for_each_bio(bio) { |
497 | bio_for_each_bvec(bvec, bio, iter) { | |
498 | /* | |
499 | * Only try to merge bvecs from two bios given we | |
500 | * have done bio internal merge when adding pages | |
501 | * to bio | |
502 | */ | |
503 | if (new_bio && | |
504 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
505 | goto next_bvec; | |
506 | ||
507 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
508 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
509 | else | |
510 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
511 | next_bvec: | |
512 | new_bio = false; | |
513 | } | |
b21e11c5 ML |
514 | if (likely(bio->bi_iter.bi_size)) { |
515 | bvprv = bvec; | |
516 | new_bio = true; | |
517 | } | |
f6970f83 | 518 | } |
d6d48196 | 519 | |
5cb8850c KO |
520 | return nsegs; |
521 | } | |
522 | ||
523 | /* | |
524 | * map a request to scatterlist, return number of sg entries setup. Caller | |
525 | * must make sure sg can hold rq->nr_phys_segments entries | |
526 | */ | |
89de1504 CH |
527 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
528 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 529 | { |
5cb8850c KO |
530 | int nsegs = 0; |
531 | ||
f9d03f96 | 532 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 533 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 534 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) |
89de1504 | 535 | nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg); |
f9d03f96 | 536 | else if (rq->bio) |
89de1504 | 537 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 538 | |
89de1504 CH |
539 | if (*last_sg) |
540 | sg_mark_end(*last_sg); | |
d6d48196 | 541 | |
12e57f59 ML |
542 | /* |
543 | * Something must have been wrong if the figured number of | |
544 | * segment is bigger than number of req's physical segments | |
545 | */ | |
f9d03f96 | 546 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 547 | |
d6d48196 JA |
548 | return nsegs; |
549 | } | |
89de1504 | 550 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 551 | |
943b40c8 ML |
552 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
553 | { | |
554 | if (req_op(rq) == REQ_OP_DISCARD) | |
555 | return queue_max_discard_segments(rq->q); | |
556 | return queue_max_segments(rq->q); | |
557 | } | |
558 | ||
14ccb66b CH |
559 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
560 | unsigned int nr_phys_segs) | |
d6d48196 | 561 | { |
2705dfb2 | 562 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d MP |
563 | goto no_merge; |
564 | ||
2705dfb2 ML |
565 | /* discard request merge won't add new segment */ |
566 | if (req_op(req) == REQ_OP_DISCARD) | |
567 | return 1; | |
568 | ||
569 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) | |
13f05c8d | 570 | goto no_merge; |
d6d48196 JA |
571 | |
572 | /* | |
573 | * This will form the start of a new hw segment. Bump both | |
574 | * counters. | |
575 | */ | |
d6d48196 JA |
576 | req->nr_phys_segments += nr_phys_segs; |
577 | return 1; | |
13f05c8d MP |
578 | |
579 | no_merge: | |
14ccb66b | 580 | req_set_nomerge(req->q, req); |
13f05c8d | 581 | return 0; |
d6d48196 JA |
582 | } |
583 | ||
14ccb66b | 584 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 585 | { |
5e7c4274 JA |
586 | if (req_gap_back_merge(req, bio)) |
587 | return 0; | |
7f39add3 SG |
588 | if (blk_integrity_rq(req) && |
589 | integrity_req_gap_back_merge(req, bio)) | |
590 | return 0; | |
a892c8d5 ST |
591 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
592 | return 0; | |
f31dc1cd | 593 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 594 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 595 | req_set_nomerge(req->q, req); |
d6d48196 JA |
596 | return 0; |
597 | } | |
d6d48196 | 598 | |
14ccb66b | 599 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
600 | } |
601 | ||
eda5cc99 CH |
602 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
603 | unsigned int nr_segs) | |
d6d48196 | 604 | { |
5e7c4274 JA |
605 | if (req_gap_front_merge(req, bio)) |
606 | return 0; | |
7f39add3 SG |
607 | if (blk_integrity_rq(req) && |
608 | integrity_req_gap_front_merge(req, bio)) | |
609 | return 0; | |
a892c8d5 ST |
610 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
611 | return 0; | |
f31dc1cd | 612 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 613 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 614 | req_set_nomerge(req->q, req); |
d6d48196 JA |
615 | return 0; |
616 | } | |
d6d48196 | 617 | |
14ccb66b | 618 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
619 | } |
620 | ||
445251d0 JA |
621 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
622 | struct request *next) | |
623 | { | |
624 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
625 | ||
626 | if (segments >= queue_max_discard_segments(q)) | |
627 | goto no_merge; | |
628 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
629 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
630 | goto no_merge; | |
631 | ||
632 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
633 | return true; | |
634 | no_merge: | |
635 | req_set_nomerge(q, req); | |
636 | return false; | |
637 | } | |
638 | ||
d6d48196 JA |
639 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
640 | struct request *next) | |
641 | { | |
642 | int total_phys_segments; | |
d6d48196 | 643 | |
5e7c4274 | 644 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
645 | return 0; |
646 | ||
d6d48196 JA |
647 | /* |
648 | * Will it become too large? | |
649 | */ | |
f31dc1cd | 650 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 651 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
652 | return 0; |
653 | ||
654 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 655 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
656 | return 0; |
657 | ||
4eaf99be | 658 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
659 | return 0; |
660 | ||
a892c8d5 ST |
661 | if (!bio_crypt_ctx_merge_rq(req, next)) |
662 | return 0; | |
663 | ||
d6d48196 JA |
664 | /* Merge is OK... */ |
665 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
666 | return 1; |
667 | } | |
668 | ||
80a761fd TH |
669 | /** |
670 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
671 | * @rq: request to mark as mixed merge | |
672 | * | |
673 | * Description: | |
674 | * @rq is about to be mixed merged. Make sure the attributes | |
675 | * which can be mixed are set in each bio and mark @rq as mixed | |
676 | * merged. | |
677 | */ | |
678 | void blk_rq_set_mixed_merge(struct request *rq) | |
679 | { | |
680 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
681 | struct bio *bio; | |
682 | ||
e8064021 | 683 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
684 | return; |
685 | ||
686 | /* | |
687 | * @rq will no longer represent mixable attributes for all the | |
688 | * contained bios. It will just track those of the first one. | |
689 | * Distributes the attributs to each bio. | |
690 | */ | |
691 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
692 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
693 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
694 | bio->bi_opf |= ff; | |
80a761fd | 695 | } |
e8064021 | 696 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
697 | } |
698 | ||
b9c54f56 | 699 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
700 | { |
701 | if (blk_do_io_stat(req)) { | |
112f158f | 702 | part_stat_lock(); |
b9c54f56 | 703 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab JM |
704 | part_stat_unlock(); |
705 | } | |
706 | } | |
b9c54f56 | 707 | |
69840466 JW |
708 | /* |
709 | * Two cases of handling DISCARD merge: | |
710 | * If max_discard_segments > 1, the driver takes every bio | |
711 | * as a range and send them to controller together. The ranges | |
712 | * needn't to be contiguous. | |
713 | * Otherwise, the bios/requests will be handled as same as | |
714 | * others which should be contiguous. | |
715 | */ | |
716 | static inline bool blk_discard_mergable(struct request *req) | |
717 | { | |
718 | if (req_op(req) == REQ_OP_DISCARD && | |
719 | queue_max_discard_segments(req->q) > 1) | |
720 | return true; | |
721 | return false; | |
722 | } | |
723 | ||
e96c0d83 EB |
724 | static enum elv_merge blk_try_req_merge(struct request *req, |
725 | struct request *next) | |
69840466 JW |
726 | { |
727 | if (blk_discard_mergable(req)) | |
728 | return ELEVATOR_DISCARD_MERGE; | |
729 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
730 | return ELEVATOR_BACK_MERGE; | |
731 | ||
732 | return ELEVATOR_NO_MERGE; | |
733 | } | |
26308eab | 734 | |
d6d48196 | 735 | /* |
b973cb7e JA |
736 | * For non-mq, this has to be called with the request spinlock acquired. |
737 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 738 | */ |
b973cb7e JA |
739 | static struct request *attempt_merge(struct request_queue *q, |
740 | struct request *req, struct request *next) | |
d6d48196 JA |
741 | { |
742 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 743 | return NULL; |
d6d48196 | 744 | |
288dab8a | 745 | if (req_op(req) != req_op(next)) |
b973cb7e | 746 | return NULL; |
f31dc1cd | 747 | |
d6d48196 | 748 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 749 | || req->rq_disk != next->rq_disk) |
b973cb7e | 750 | return NULL; |
d6d48196 | 751 | |
8fe0d473 | 752 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 753 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 754 | return NULL; |
4363ac7c | 755 | |
cb6934f8 JA |
756 | /* |
757 | * Don't allow merge of different write hints, or for a hint with | |
758 | * non-hint IO. | |
759 | */ | |
760 | if (req->write_hint != next->write_hint) | |
761 | return NULL; | |
762 | ||
668ffc03 DLM |
763 | if (req->ioprio != next->ioprio) |
764 | return NULL; | |
765 | ||
d6d48196 JA |
766 | /* |
767 | * If we are allowed to merge, then append bio list | |
768 | * from next to rq and release next. merge_requests_fn | |
769 | * will have updated segment counts, update sector | |
445251d0 JA |
770 | * counts here. Handle DISCARDs separately, as they |
771 | * have separate settings. | |
d6d48196 | 772 | */ |
69840466 JW |
773 | |
774 | switch (blk_try_req_merge(req, next)) { | |
775 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
776 | if (!req_attempt_discard_merge(q, req, next)) |
777 | return NULL; | |
69840466 JW |
778 | break; |
779 | case ELEVATOR_BACK_MERGE: | |
780 | if (!ll_merge_requests_fn(q, req, next)) | |
781 | return NULL; | |
782 | break; | |
783 | default: | |
b973cb7e | 784 | return NULL; |
69840466 | 785 | } |
d6d48196 | 786 | |
80a761fd TH |
787 | /* |
788 | * If failfast settings disagree or any of the two is already | |
789 | * a mixed merge, mark both as mixed before proceeding. This | |
790 | * makes sure that all involved bios have mixable attributes | |
791 | * set properly. | |
792 | */ | |
e8064021 | 793 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
794 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
795 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
796 | blk_rq_set_mixed_merge(req); | |
797 | blk_rq_set_mixed_merge(next); | |
798 | } | |
799 | ||
d6d48196 | 800 | /* |
522a7775 OS |
801 | * At this point we have either done a back merge or front merge. We |
802 | * need the smaller start_time_ns of the merged requests to be the | |
803 | * current request for accounting purposes. | |
d6d48196 | 804 | */ |
522a7775 OS |
805 | if (next->start_time_ns < req->start_time_ns) |
806 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
807 | |
808 | req->biotail->bi_next = next->bio; | |
809 | req->biotail = next->biotail; | |
810 | ||
a2dec7b3 | 811 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 812 | |
2a5cf35c | 813 | if (!blk_discard_mergable(req)) |
445251d0 | 814 | elv_merge_requests(q, req, next); |
d6d48196 | 815 | |
42dad764 JM |
816 | /* |
817 | * 'next' is going away, so update stats accordingly | |
818 | */ | |
b9c54f56 | 819 | blk_account_io_merge_request(next); |
d6d48196 | 820 | |
a54895fa | 821 | trace_block_rq_merge(next); |
f3bdc62f | 822 | |
e4d750c9 JA |
823 | /* |
824 | * ownership of bio passed from next to req, return 'next' for | |
825 | * the caller to free | |
826 | */ | |
1cd96c24 | 827 | next->bio = NULL; |
b973cb7e | 828 | return next; |
d6d48196 JA |
829 | } |
830 | ||
eda5cc99 CH |
831 | static struct request *attempt_back_merge(struct request_queue *q, |
832 | struct request *rq) | |
d6d48196 JA |
833 | { |
834 | struct request *next = elv_latter_request(q, rq); | |
835 | ||
836 | if (next) | |
837 | return attempt_merge(q, rq, next); | |
838 | ||
b973cb7e | 839 | return NULL; |
d6d48196 JA |
840 | } |
841 | ||
eda5cc99 CH |
842 | static struct request *attempt_front_merge(struct request_queue *q, |
843 | struct request *rq) | |
d6d48196 JA |
844 | { |
845 | struct request *prev = elv_former_request(q, rq); | |
846 | ||
847 | if (prev) | |
848 | return attempt_merge(q, prev, rq); | |
849 | ||
b973cb7e | 850 | return NULL; |
d6d48196 | 851 | } |
5e84ea3a | 852 | |
fd2ef39c JK |
853 | /* |
854 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false | |
855 | * otherwise. The caller is responsible for freeing 'next' if the merge | |
856 | * happened. | |
857 | */ | |
858 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
859 | struct request *next) | |
5e84ea3a | 860 | { |
fd2ef39c | 861 | return attempt_merge(q, rq, next); |
5e84ea3a | 862 | } |
050c8ea8 TH |
863 | |
864 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
865 | { | |
e2a60da7 | 866 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
867 | return false; |
868 | ||
288dab8a | 869 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
870 | return false; |
871 | ||
050c8ea8 TH |
872 | /* different data direction or already started, don't merge */ |
873 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
874 | return false; | |
875 | ||
2081a56b | 876 | /* must be same device */ |
309dca30 | 877 | if (rq->rq_disk != bio->bi_bdev->bd_disk) |
050c8ea8 TH |
878 | return false; |
879 | ||
880 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 881 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
882 | return false; |
883 | ||
a892c8d5 ST |
884 | /* Only merge if the crypt contexts are compatible */ |
885 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
886 | return false; | |
887 | ||
4363ac7c | 888 | /* must be using the same buffer */ |
8fe0d473 | 889 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
890 | !blk_write_same_mergeable(rq->bio, bio)) |
891 | return false; | |
892 | ||
cb6934f8 JA |
893 | /* |
894 | * Don't allow merge of different write hints, or for a hint with | |
895 | * non-hint IO. | |
896 | */ | |
897 | if (rq->write_hint != bio->bi_write_hint) | |
898 | return false; | |
899 | ||
668ffc03 DLM |
900 | if (rq->ioprio != bio_prio(bio)) |
901 | return false; | |
902 | ||
050c8ea8 TH |
903 | return true; |
904 | } | |
905 | ||
34fe7c05 | 906 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 907 | { |
69840466 | 908 | if (blk_discard_mergable(rq)) |
1e739730 CH |
909 | return ELEVATOR_DISCARD_MERGE; |
910 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 911 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 912 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
913 | return ELEVATOR_FRONT_MERGE; |
914 | return ELEVATOR_NO_MERGE; | |
915 | } | |
8e756373 BW |
916 | |
917 | static void blk_account_io_merge_bio(struct request *req) | |
918 | { | |
919 | if (!blk_do_io_stat(req)) | |
920 | return; | |
921 | ||
922 | part_stat_lock(); | |
923 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
924 | part_stat_unlock(); | |
925 | } | |
926 | ||
eda5cc99 CH |
927 | enum bio_merge_status { |
928 | BIO_MERGE_OK, | |
929 | BIO_MERGE_NONE, | |
930 | BIO_MERGE_FAILED, | |
931 | }; | |
932 | ||
933 | static enum bio_merge_status bio_attempt_back_merge(struct request *req, | |
934 | struct bio *bio, unsigned int nr_segs) | |
8e756373 BW |
935 | { |
936 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
937 | ||
938 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 939 | return BIO_MERGE_FAILED; |
8e756373 | 940 | |
e8a676d6 | 941 | trace_block_bio_backmerge(bio); |
8e756373 BW |
942 | rq_qos_merge(req->q, req, bio); |
943 | ||
944 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
945 | blk_rq_set_mixed_merge(req); | |
946 | ||
947 | req->biotail->bi_next = bio; | |
948 | req->biotail = bio; | |
949 | req->__data_len += bio->bi_iter.bi_size; | |
950 | ||
951 | bio_crypt_free_ctx(bio); | |
952 | ||
953 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 954 | return BIO_MERGE_OK; |
8e756373 BW |
955 | } |
956 | ||
eda5cc99 CH |
957 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
958 | struct bio *bio, unsigned int nr_segs) | |
8e756373 BW |
959 | { |
960 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
961 | ||
962 | if (!ll_front_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 963 | return BIO_MERGE_FAILED; |
8e756373 | 964 | |
e8a676d6 | 965 | trace_block_bio_frontmerge(bio); |
8e756373 BW |
966 | rq_qos_merge(req->q, req, bio); |
967 | ||
968 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
969 | blk_rq_set_mixed_merge(req); | |
970 | ||
971 | bio->bi_next = req->bio; | |
972 | req->bio = bio; | |
973 | ||
974 | req->__sector = bio->bi_iter.bi_sector; | |
975 | req->__data_len += bio->bi_iter.bi_size; | |
976 | ||
977 | bio_crypt_do_front_merge(req, bio); | |
978 | ||
979 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 980 | return BIO_MERGE_OK; |
8e756373 BW |
981 | } |
982 | ||
eda5cc99 CH |
983 | static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
984 | struct request *req, struct bio *bio) | |
8e756373 BW |
985 | { |
986 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
987 | ||
988 | if (segments >= queue_max_discard_segments(q)) | |
989 | goto no_merge; | |
990 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
991 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
992 | goto no_merge; | |
993 | ||
994 | rq_qos_merge(q, req, bio); | |
995 | ||
996 | req->biotail->bi_next = bio; | |
997 | req->biotail = bio; | |
998 | req->__data_len += bio->bi_iter.bi_size; | |
999 | req->nr_phys_segments = segments + 1; | |
1000 | ||
1001 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1002 | return BIO_MERGE_OK; |
8e756373 BW |
1003 | no_merge: |
1004 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
1005 | return BIO_MERGE_FAILED; |
1006 | } | |
1007 | ||
1008 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
1009 | struct request *rq, | |
1010 | struct bio *bio, | |
1011 | unsigned int nr_segs, | |
1012 | bool sched_allow_merge) | |
1013 | { | |
1014 | if (!blk_rq_merge_ok(rq, bio)) | |
1015 | return BIO_MERGE_NONE; | |
1016 | ||
1017 | switch (blk_try_merge(rq, bio)) { | |
1018 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 1019 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1020 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1021 | break; | |
1022 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1023 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1024 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1025 | break; | |
1026 | case ELEVATOR_DISCARD_MERGE: | |
1027 | return bio_attempt_discard_merge(q, rq, bio); | |
1028 | default: | |
1029 | return BIO_MERGE_NONE; | |
1030 | } | |
1031 | ||
1032 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1033 | } |
1034 | ||
1035 | /** | |
1036 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1037 | * @q: request_queue new bio is being queued at | |
1038 | * @bio: new bio being queued | |
1039 | * @nr_segs: number of segments in @bio | |
1040 | * @same_queue_rq: pointer to &struct request that gets filled in when | |
1041 | * another request associated with @q is found on the plug list | |
1042 | * (optional, may be %NULL) | |
1043 | * | |
1044 | * Determine whether @bio being queued on @q can be merged with a request | |
1045 | * on %current's plugged list. Returns %true if merge was successful, | |
1046 | * otherwise %false. | |
1047 | * | |
1048 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1049 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1050 | * than scheduling, and the request, while may have elvpriv data, is not | |
1051 | * added on the elevator at this point. In addition, we don't have | |
1052 | * reliable access to the elevator outside queue lock. Only check basic | |
1053 | * merging parameters without querying the elevator. | |
1054 | * | |
1055 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1056 | */ | |
1057 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
1058 | unsigned int nr_segs, struct request **same_queue_rq) | |
1059 | { | |
1060 | struct blk_plug *plug; | |
1061 | struct request *rq; | |
1062 | struct list_head *plug_list; | |
1063 | ||
1064 | plug = blk_mq_plug(q, bio); | |
1065 | if (!plug) | |
1066 | return false; | |
1067 | ||
1068 | plug_list = &plug->mq_list; | |
1069 | ||
1070 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
8e756373 BW |
1071 | if (rq->q == q && same_queue_rq) { |
1072 | /* | |
1073 | * Only blk-mq multiple hardware queues case checks the | |
1074 | * rq in the same queue, there should be only one such | |
1075 | * rq in a queue | |
1076 | **/ | |
1077 | *same_queue_rq = rq; | |
1078 | } | |
1079 | ||
7d7ca7c5 | 1080 | if (rq->q != q) |
8e756373 BW |
1081 | continue; |
1082 | ||
7d7ca7c5 BW |
1083 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == |
1084 | BIO_MERGE_OK) | |
8e756373 BW |
1085 | return true; |
1086 | } | |
1087 | ||
1088 | return false; | |
1089 | } | |
bdc6a287 BW |
1090 | |
1091 | /* | |
1092 | * Iterate list of requests and see if we can merge this bio with any | |
1093 | * of them. | |
1094 | */ | |
1095 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1096 | struct bio *bio, unsigned int nr_segs) | |
1097 | { | |
1098 | struct request *rq; | |
1099 | int checked = 8; | |
1100 | ||
1101 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1102 | if (!checked--) |
1103 | break; | |
1104 | ||
7d7ca7c5 BW |
1105 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1106 | case BIO_MERGE_NONE: | |
bdc6a287 | 1107 | continue; |
7d7ca7c5 BW |
1108 | case BIO_MERGE_OK: |
1109 | return true; | |
1110 | case BIO_MERGE_FAILED: | |
1111 | return false; | |
bdc6a287 BW |
1112 | } |
1113 | ||
bdc6a287 BW |
1114 | } |
1115 | ||
1116 | return false; | |
1117 | } | |
1118 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); | |
eda5cc99 CH |
1119 | |
1120 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, | |
1121 | unsigned int nr_segs, struct request **merged_request) | |
1122 | { | |
1123 | struct request *rq; | |
1124 | ||
1125 | switch (elv_merge(q, &rq, bio)) { | |
1126 | case ELEVATOR_BACK_MERGE: | |
1127 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1128 | return false; | |
1129 | if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1130 | return false; | |
1131 | *merged_request = attempt_back_merge(q, rq); | |
1132 | if (!*merged_request) | |
1133 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
1134 | return true; | |
1135 | case ELEVATOR_FRONT_MERGE: | |
1136 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1137 | return false; | |
1138 | if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1139 | return false; | |
1140 | *merged_request = attempt_front_merge(q, rq); | |
1141 | if (!*merged_request) | |
1142 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
1143 | return true; | |
1144 | case ELEVATOR_DISCARD_MERGE: | |
1145 | return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; | |
1146 | default: | |
1147 | return false; | |
1148 | } | |
1149 | } | |
1150 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |