]>
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 | { | |
147 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); | |
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; |
d627065d | 282 | return bio_split(bio, sectors, GFP_NOIO, bs); |
54efd50b KO |
283 | } |
284 | ||
dad77584 BVA |
285 | /** |
286 | * __blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
287 | * @bio: [in, out] bio to be split |
288 | * @nr_segs: [out] number of segments in the first bio | |
289 | * | |
290 | * Split a bio into two bios, chain the two bios, submit the second half and | |
291 | * store a pointer to the first half in *@bio. If the second bio is still too | |
292 | * big it will be split by a recursive call to this function. Since this | |
f695ca38 CH |
293 | * function may allocate a new bio from @bio->bi_disk->queue->bio_split, it is |
294 | * the responsibility of the caller to ensure that | |
295 | * @bio->bi_disk->queue->bio_split is only released after processing of the | |
dad77584 BVA |
296 | * split bio has finished. |
297 | */ | |
f695ca38 | 298 | void __blk_queue_split(struct bio **bio, unsigned int *nr_segs) |
54efd50b | 299 | { |
f695ca38 | 300 | struct request_queue *q = (*bio)->bi_disk->queue; |
fa532287 | 301 | struct bio *split = NULL; |
54efd50b | 302 | |
7afafc8a AH |
303 | switch (bio_op(*bio)) { |
304 | case REQ_OP_DISCARD: | |
305 | case REQ_OP_SECURE_ERASE: | |
14ccb66b | 306 | split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a | 307 | break; |
a6f0788e | 308 | case REQ_OP_WRITE_ZEROES: |
14ccb66b CH |
309 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, |
310 | nr_segs); | |
a6f0788e | 311 | break; |
7afafc8a | 312 | case REQ_OP_WRITE_SAME: |
14ccb66b CH |
313 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, |
314 | nr_segs); | |
7afafc8a AH |
315 | break; |
316 | default: | |
fa532287 CH |
317 | /* |
318 | * All drivers must accept single-segments bios that are <= | |
319 | * PAGE_SIZE. This is a quick and dirty check that relies on | |
320 | * the fact that bi_io_vec[0] is always valid if a bio has data. | |
321 | * The check might lead to occasional false negatives when bios | |
322 | * are cloned, but compared to the performance impact of cloned | |
323 | * bios themselves the loop below doesn't matter anyway. | |
324 | */ | |
325 | if (!q->limits.chunk_sectors && | |
326 | (*bio)->bi_vcnt == 1 && | |
59db8ba2 | 327 | ((*bio)->bi_io_vec[0].bv_len + |
1e279153 | 328 | (*bio)->bi_io_vec[0].bv_offset) <= PAGE_SIZE) { |
fa532287 CH |
329 | *nr_segs = 1; |
330 | break; | |
331 | } | |
14ccb66b | 332 | split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a AH |
333 | break; |
334 | } | |
bdced438 | 335 | |
54efd50b | 336 | if (split) { |
6ac45aeb | 337 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 338 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 339 | |
54efd50b | 340 | bio_chain(split, *bio); |
cda22646 | 341 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
ed00aabd | 342 | submit_bio_noacct(*bio); |
54efd50b KO |
343 | *bio = split; |
344 | } | |
345 | } | |
14ccb66b | 346 | |
dad77584 BVA |
347 | /** |
348 | * blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
349 | * @bio: [in, out] bio to be split |
350 | * | |
351 | * Split a bio into two bios, chains the two bios, submit the second half and | |
352 | * store a pointer to the first half in *@bio. Since this function may allocate | |
f695ca38 CH |
353 | * a new bio from @bio->bi_disk->queue->bio_split, it is the responsibility of |
354 | * the caller to ensure that @bio->bi_disk->queue->bio_split is only released | |
355 | * after processing of the split bio has finished. | |
dad77584 | 356 | */ |
f695ca38 | 357 | void blk_queue_split(struct bio **bio) |
14ccb66b CH |
358 | { |
359 | unsigned int nr_segs; | |
360 | ||
f695ca38 | 361 | __blk_queue_split(bio, &nr_segs); |
14ccb66b | 362 | } |
54efd50b KO |
363 | EXPORT_SYMBOL(blk_queue_split); |
364 | ||
e9cd19c0 | 365 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 366 | { |
6869875f | 367 | unsigned int nr_phys_segs = 0; |
ff9811b3 | 368 | unsigned int nr_sectors = 0; |
e9cd19c0 | 369 | struct req_iterator iter; |
6869875f | 370 | struct bio_vec bv; |
d6d48196 | 371 | |
e9cd19c0 | 372 | if (!rq->bio) |
1e428079 | 373 | return 0; |
d6d48196 | 374 | |
e9cd19c0 | 375 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
376 | case REQ_OP_DISCARD: |
377 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 378 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
379 | return 0; |
380 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 381 | return 1; |
a6f0788e | 382 | } |
5cb8850c | 383 | |
e9cd19c0 | 384 | rq_for_each_bvec(bv, rq, iter) |
ff9811b3 | 385 | bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors, |
708b25b3 | 386 | UINT_MAX, UINT_MAX); |
1e428079 JA |
387 | return nr_phys_segs; |
388 | } | |
389 | ||
48d7727c | 390 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
391 | struct scatterlist *sglist) |
392 | { | |
393 | if (!*sg) | |
394 | return sglist; | |
395 | ||
396 | /* | |
397 | * If the driver previously mapped a shorter list, we could see a | |
398 | * termination bit prematurely unless it fully inits the sg table | |
399 | * on each mapping. We KNOW that there must be more entries here | |
400 | * or the driver would be buggy, so force clear the termination bit | |
401 | * to avoid doing a full sg_init_table() in drivers for each command. | |
402 | */ | |
403 | sg_unmark_end(*sg); | |
404 | return sg_next(*sg); | |
405 | } | |
406 | ||
407 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
408 | struct bio_vec *bvec, struct scatterlist *sglist, | |
409 | struct scatterlist **sg) | |
410 | { | |
411 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 412 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
413 | |
414 | while (nbytes > 0) { | |
8a96a0e4 | 415 | unsigned offset = bvec->bv_offset + total; |
429120f3 ML |
416 | unsigned len = min(get_max_segment_size(q, bvec->bv_page, |
417 | offset), nbytes); | |
f9f76879 CH |
418 | struct page *page = bvec->bv_page; |
419 | ||
420 | /* | |
421 | * Unfortunately a fair number of drivers barf on scatterlists | |
422 | * that have an offset larger than PAGE_SIZE, despite other | |
423 | * subsystems dealing with that invariant just fine. For now | |
424 | * stick to the legacy format where we never present those from | |
425 | * the block layer, but the code below should be removed once | |
426 | * these offenders (mostly MMC/SD drivers) are fixed. | |
427 | */ | |
428 | page += (offset >> PAGE_SHIFT); | |
429 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
430 | |
431 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 432 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 433 | |
8a96a0e4 CH |
434 | total += len; |
435 | nbytes -= len; | |
862e5a5e ML |
436 | nsegs++; |
437 | } | |
438 | ||
439 | return nsegs; | |
440 | } | |
441 | ||
16e3e418 ML |
442 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
443 | struct scatterlist *sglist, struct scatterlist **sg) | |
444 | { | |
445 | *sg = blk_next_sg(sg, sglist); | |
446 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
447 | return 1; | |
448 | } | |
449 | ||
f6970f83 ML |
450 | /* only try to merge bvecs into one sg if they are from two bios */ |
451 | static inline bool | |
452 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
453 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
454 | { |
455 | ||
456 | int nbytes = bvec->bv_len; | |
457 | ||
f6970f83 ML |
458 | if (!*sg) |
459 | return false; | |
963ab9e5 | 460 | |
f6970f83 ML |
461 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
462 | return false; | |
463 | ||
464 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
465 | return false; | |
466 | ||
467 | (*sg)->length += nbytes; | |
468 | ||
469 | return true; | |
963ab9e5 AH |
470 | } |
471 | ||
5cb8850c KO |
472 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
473 | struct scatterlist *sglist, | |
474 | struct scatterlist **sg) | |
d6d48196 | 475 | { |
3f649ab7 | 476 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 477 | struct bvec_iter iter; |
38417468 | 478 | int nsegs = 0; |
f6970f83 | 479 | bool new_bio = false; |
5cb8850c | 480 | |
f6970f83 ML |
481 | for_each_bio(bio) { |
482 | bio_for_each_bvec(bvec, bio, iter) { | |
483 | /* | |
484 | * Only try to merge bvecs from two bios given we | |
485 | * have done bio internal merge when adding pages | |
486 | * to bio | |
487 | */ | |
488 | if (new_bio && | |
489 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
490 | goto next_bvec; | |
491 | ||
492 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
493 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
494 | else | |
495 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
496 | next_bvec: | |
497 | new_bio = false; | |
498 | } | |
b21e11c5 ML |
499 | if (likely(bio->bi_iter.bi_size)) { |
500 | bvprv = bvec; | |
501 | new_bio = true; | |
502 | } | |
f6970f83 | 503 | } |
d6d48196 | 504 | |
5cb8850c KO |
505 | return nsegs; |
506 | } | |
507 | ||
508 | /* | |
509 | * map a request to scatterlist, return number of sg entries setup. Caller | |
510 | * must make sure sg can hold rq->nr_phys_segments entries | |
511 | */ | |
89de1504 CH |
512 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
513 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 514 | { |
5cb8850c KO |
515 | int nsegs = 0; |
516 | ||
f9d03f96 | 517 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 518 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 519 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) |
89de1504 | 520 | nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg); |
f9d03f96 | 521 | else if (rq->bio) |
89de1504 | 522 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 523 | |
89de1504 CH |
524 | if (*last_sg) |
525 | sg_mark_end(*last_sg); | |
d6d48196 | 526 | |
12e57f59 ML |
527 | /* |
528 | * Something must have been wrong if the figured number of | |
529 | * segment is bigger than number of req's physical segments | |
530 | */ | |
f9d03f96 | 531 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 532 | |
d6d48196 JA |
533 | return nsegs; |
534 | } | |
89de1504 | 535 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 536 | |
943b40c8 ML |
537 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
538 | { | |
539 | if (req_op(rq) == REQ_OP_DISCARD) | |
540 | return queue_max_discard_segments(rq->q); | |
541 | return queue_max_segments(rq->q); | |
542 | } | |
543 | ||
14ccb66b CH |
544 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
545 | unsigned int nr_phys_segs) | |
d6d48196 | 546 | { |
943b40c8 | 547 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) |
13f05c8d MP |
548 | goto no_merge; |
549 | ||
14ccb66b | 550 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d | 551 | goto no_merge; |
d6d48196 JA |
552 | |
553 | /* | |
554 | * This will form the start of a new hw segment. Bump both | |
555 | * counters. | |
556 | */ | |
d6d48196 JA |
557 | req->nr_phys_segments += nr_phys_segs; |
558 | return 1; | |
13f05c8d MP |
559 | |
560 | no_merge: | |
14ccb66b | 561 | req_set_nomerge(req->q, req); |
13f05c8d | 562 | return 0; |
d6d48196 JA |
563 | } |
564 | ||
14ccb66b | 565 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 566 | { |
5e7c4274 JA |
567 | if (req_gap_back_merge(req, bio)) |
568 | return 0; | |
7f39add3 SG |
569 | if (blk_integrity_rq(req) && |
570 | integrity_req_gap_back_merge(req, bio)) | |
571 | return 0; | |
a892c8d5 ST |
572 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
573 | return 0; | |
f31dc1cd | 574 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 575 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 576 | req_set_nomerge(req->q, req); |
d6d48196 JA |
577 | return 0; |
578 | } | |
d6d48196 | 579 | |
14ccb66b | 580 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
581 | } |
582 | ||
14ccb66b | 583 | int ll_front_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 584 | { |
5e7c4274 JA |
585 | if (req_gap_front_merge(req, bio)) |
586 | return 0; | |
7f39add3 SG |
587 | if (blk_integrity_rq(req) && |
588 | integrity_req_gap_front_merge(req, bio)) | |
589 | return 0; | |
a892c8d5 ST |
590 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
591 | return 0; | |
f31dc1cd | 592 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 593 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 594 | req_set_nomerge(req->q, req); |
d6d48196 JA |
595 | return 0; |
596 | } | |
d6d48196 | 597 | |
14ccb66b | 598 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
599 | } |
600 | ||
445251d0 JA |
601 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
602 | struct request *next) | |
603 | { | |
604 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
605 | ||
606 | if (segments >= queue_max_discard_segments(q)) | |
607 | goto no_merge; | |
608 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
609 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
610 | goto no_merge; | |
611 | ||
612 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
613 | return true; | |
614 | no_merge: | |
615 | req_set_nomerge(q, req); | |
616 | return false; | |
617 | } | |
618 | ||
d6d48196 JA |
619 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
620 | struct request *next) | |
621 | { | |
622 | int total_phys_segments; | |
d6d48196 | 623 | |
5e7c4274 | 624 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
625 | return 0; |
626 | ||
d6d48196 JA |
627 | /* |
628 | * Will it become too large? | |
629 | */ | |
f31dc1cd | 630 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 631 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
632 | return 0; |
633 | ||
634 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 635 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
636 | return 0; |
637 | ||
4eaf99be | 638 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
639 | return 0; |
640 | ||
a892c8d5 ST |
641 | if (!bio_crypt_ctx_merge_rq(req, next)) |
642 | return 0; | |
643 | ||
d6d48196 JA |
644 | /* Merge is OK... */ |
645 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
646 | return 1; |
647 | } | |
648 | ||
80a761fd TH |
649 | /** |
650 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
651 | * @rq: request to mark as mixed merge | |
652 | * | |
653 | * Description: | |
654 | * @rq is about to be mixed merged. Make sure the attributes | |
655 | * which can be mixed are set in each bio and mark @rq as mixed | |
656 | * merged. | |
657 | */ | |
658 | void blk_rq_set_mixed_merge(struct request *rq) | |
659 | { | |
660 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
661 | struct bio *bio; | |
662 | ||
e8064021 | 663 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
664 | return; |
665 | ||
666 | /* | |
667 | * @rq will no longer represent mixable attributes for all the | |
668 | * contained bios. It will just track those of the first one. | |
669 | * Distributes the attributs to each bio. | |
670 | */ | |
671 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
672 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
673 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
674 | bio->bi_opf |= ff; | |
80a761fd | 675 | } |
e8064021 | 676 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
677 | } |
678 | ||
b9c54f56 | 679 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
680 | { |
681 | if (blk_do_io_stat(req)) { | |
112f158f | 682 | part_stat_lock(); |
b9c54f56 | 683 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab | 684 | part_stat_unlock(); |
524f9ffd CH |
685 | |
686 | hd_struct_put(req->part); | |
26308eab JM |
687 | } |
688 | } | |
b9c54f56 | 689 | |
69840466 JW |
690 | /* |
691 | * Two cases of handling DISCARD merge: | |
692 | * If max_discard_segments > 1, the driver takes every bio | |
693 | * as a range and send them to controller together. The ranges | |
694 | * needn't to be contiguous. | |
695 | * Otherwise, the bios/requests will be handled as same as | |
696 | * others which should be contiguous. | |
697 | */ | |
698 | static inline bool blk_discard_mergable(struct request *req) | |
699 | { | |
700 | if (req_op(req) == REQ_OP_DISCARD && | |
701 | queue_max_discard_segments(req->q) > 1) | |
702 | return true; | |
703 | return false; | |
704 | } | |
705 | ||
e96c0d83 EB |
706 | static enum elv_merge blk_try_req_merge(struct request *req, |
707 | struct request *next) | |
69840466 JW |
708 | { |
709 | if (blk_discard_mergable(req)) | |
710 | return ELEVATOR_DISCARD_MERGE; | |
711 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
712 | return ELEVATOR_BACK_MERGE; | |
713 | ||
714 | return ELEVATOR_NO_MERGE; | |
715 | } | |
26308eab | 716 | |
d6d48196 | 717 | /* |
b973cb7e JA |
718 | * For non-mq, this has to be called with the request spinlock acquired. |
719 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 720 | */ |
b973cb7e JA |
721 | static struct request *attempt_merge(struct request_queue *q, |
722 | struct request *req, struct request *next) | |
d6d48196 JA |
723 | { |
724 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 725 | return NULL; |
d6d48196 | 726 | |
288dab8a | 727 | if (req_op(req) != req_op(next)) |
b973cb7e | 728 | return NULL; |
f31dc1cd | 729 | |
d6d48196 | 730 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 731 | || req->rq_disk != next->rq_disk) |
b973cb7e | 732 | return NULL; |
d6d48196 | 733 | |
8fe0d473 | 734 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 735 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 736 | return NULL; |
4363ac7c | 737 | |
cb6934f8 JA |
738 | /* |
739 | * Don't allow merge of different write hints, or for a hint with | |
740 | * non-hint IO. | |
741 | */ | |
742 | if (req->write_hint != next->write_hint) | |
743 | return NULL; | |
744 | ||
668ffc03 DLM |
745 | if (req->ioprio != next->ioprio) |
746 | return NULL; | |
747 | ||
d6d48196 JA |
748 | /* |
749 | * If we are allowed to merge, then append bio list | |
750 | * from next to rq and release next. merge_requests_fn | |
751 | * will have updated segment counts, update sector | |
445251d0 JA |
752 | * counts here. Handle DISCARDs separately, as they |
753 | * have separate settings. | |
d6d48196 | 754 | */ |
69840466 JW |
755 | |
756 | switch (blk_try_req_merge(req, next)) { | |
757 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
758 | if (!req_attempt_discard_merge(q, req, next)) |
759 | return NULL; | |
69840466 JW |
760 | break; |
761 | case ELEVATOR_BACK_MERGE: | |
762 | if (!ll_merge_requests_fn(q, req, next)) | |
763 | return NULL; | |
764 | break; | |
765 | default: | |
b973cb7e | 766 | return NULL; |
69840466 | 767 | } |
d6d48196 | 768 | |
80a761fd TH |
769 | /* |
770 | * If failfast settings disagree or any of the two is already | |
771 | * a mixed merge, mark both as mixed before proceeding. This | |
772 | * makes sure that all involved bios have mixable attributes | |
773 | * set properly. | |
774 | */ | |
e8064021 | 775 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
776 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
777 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
778 | blk_rq_set_mixed_merge(req); | |
779 | blk_rq_set_mixed_merge(next); | |
780 | } | |
781 | ||
d6d48196 | 782 | /* |
522a7775 OS |
783 | * At this point we have either done a back merge or front merge. We |
784 | * need the smaller start_time_ns of the merged requests to be the | |
785 | * current request for accounting purposes. | |
d6d48196 | 786 | */ |
522a7775 OS |
787 | if (next->start_time_ns < req->start_time_ns) |
788 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
789 | |
790 | req->biotail->bi_next = next->bio; | |
791 | req->biotail = next->biotail; | |
792 | ||
a2dec7b3 | 793 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 794 | |
2a5cf35c | 795 | if (!blk_discard_mergable(req)) |
445251d0 | 796 | elv_merge_requests(q, req, next); |
d6d48196 | 797 | |
42dad764 JM |
798 | /* |
799 | * 'next' is going away, so update stats accordingly | |
800 | */ | |
b9c54f56 | 801 | blk_account_io_merge_request(next); |
d6d48196 | 802 | |
f3bdc62f JK |
803 | trace_block_rq_merge(q, next); |
804 | ||
e4d750c9 JA |
805 | /* |
806 | * ownership of bio passed from next to req, return 'next' for | |
807 | * the caller to free | |
808 | */ | |
1cd96c24 | 809 | next->bio = NULL; |
b973cb7e | 810 | return next; |
d6d48196 JA |
811 | } |
812 | ||
b973cb7e | 813 | struct request *attempt_back_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
814 | { |
815 | struct request *next = elv_latter_request(q, rq); | |
816 | ||
817 | if (next) | |
818 | return attempt_merge(q, rq, next); | |
819 | ||
b973cb7e | 820 | return NULL; |
d6d48196 JA |
821 | } |
822 | ||
b973cb7e | 823 | struct request *attempt_front_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
824 | { |
825 | struct request *prev = elv_former_request(q, rq); | |
826 | ||
827 | if (prev) | |
828 | return attempt_merge(q, prev, rq); | |
829 | ||
b973cb7e | 830 | return NULL; |
d6d48196 | 831 | } |
5e84ea3a JA |
832 | |
833 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
834 | struct request *next) | |
835 | { | |
e4d750c9 | 836 | struct request *free; |
72ef799b | 837 | |
e4d750c9 JA |
838 | free = attempt_merge(q, rq, next); |
839 | if (free) { | |
92bc5a24 | 840 | blk_put_request(free); |
e4d750c9 JA |
841 | return 1; |
842 | } | |
843 | ||
844 | return 0; | |
5e84ea3a | 845 | } |
050c8ea8 TH |
846 | |
847 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
848 | { | |
e2a60da7 | 849 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
850 | return false; |
851 | ||
288dab8a | 852 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
853 | return false; |
854 | ||
050c8ea8 TH |
855 | /* different data direction or already started, don't merge */ |
856 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
857 | return false; | |
858 | ||
2081a56b JA |
859 | /* must be same device */ |
860 | if (rq->rq_disk != bio->bi_disk) | |
050c8ea8 TH |
861 | return false; |
862 | ||
863 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 864 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
865 | return false; |
866 | ||
a892c8d5 ST |
867 | /* Only merge if the crypt contexts are compatible */ |
868 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
869 | return false; | |
870 | ||
4363ac7c | 871 | /* must be using the same buffer */ |
8fe0d473 | 872 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
873 | !blk_write_same_mergeable(rq->bio, bio)) |
874 | return false; | |
875 | ||
cb6934f8 JA |
876 | /* |
877 | * Don't allow merge of different write hints, or for a hint with | |
878 | * non-hint IO. | |
879 | */ | |
880 | if (rq->write_hint != bio->bi_write_hint) | |
881 | return false; | |
882 | ||
668ffc03 DLM |
883 | if (rq->ioprio != bio_prio(bio)) |
884 | return false; | |
885 | ||
050c8ea8 TH |
886 | return true; |
887 | } | |
888 | ||
34fe7c05 | 889 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 890 | { |
69840466 | 891 | if (blk_discard_mergable(rq)) |
1e739730 CH |
892 | return ELEVATOR_DISCARD_MERGE; |
893 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 894 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 895 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
896 | return ELEVATOR_FRONT_MERGE; |
897 | return ELEVATOR_NO_MERGE; | |
898 | } | |
8e756373 BW |
899 | |
900 | static void blk_account_io_merge_bio(struct request *req) | |
901 | { | |
902 | if (!blk_do_io_stat(req)) | |
903 | return; | |
904 | ||
905 | part_stat_lock(); | |
906 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
907 | part_stat_unlock(); | |
908 | } | |
909 | ||
7d7ca7c5 BW |
910 | enum bio_merge_status bio_attempt_back_merge(struct request *req, |
911 | struct bio *bio, | |
912 | unsigned int nr_segs) | |
8e756373 BW |
913 | { |
914 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
915 | ||
916 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 917 | return BIO_MERGE_FAILED; |
8e756373 BW |
918 | |
919 | trace_block_bio_backmerge(req->q, req, bio); | |
920 | rq_qos_merge(req->q, req, bio); | |
921 | ||
922 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
923 | blk_rq_set_mixed_merge(req); | |
924 | ||
925 | req->biotail->bi_next = bio; | |
926 | req->biotail = bio; | |
927 | req->__data_len += bio->bi_iter.bi_size; | |
928 | ||
929 | bio_crypt_free_ctx(bio); | |
930 | ||
931 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 932 | return BIO_MERGE_OK; |
8e756373 BW |
933 | } |
934 | ||
7d7ca7c5 BW |
935 | enum bio_merge_status bio_attempt_front_merge(struct request *req, |
936 | struct bio *bio, | |
937 | unsigned int nr_segs) | |
8e756373 BW |
938 | { |
939 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; | |
940 | ||
941 | if (!ll_front_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 942 | return BIO_MERGE_FAILED; |
8e756373 BW |
943 | |
944 | trace_block_bio_frontmerge(req->q, req, bio); | |
945 | rq_qos_merge(req->q, req, bio); | |
946 | ||
947 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
948 | blk_rq_set_mixed_merge(req); | |
949 | ||
950 | bio->bi_next = req->bio; | |
951 | req->bio = bio; | |
952 | ||
953 | req->__sector = bio->bi_iter.bi_sector; | |
954 | req->__data_len += bio->bi_iter.bi_size; | |
955 | ||
956 | bio_crypt_do_front_merge(req, bio); | |
957 | ||
958 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 959 | return BIO_MERGE_OK; |
8e756373 BW |
960 | } |
961 | ||
7d7ca7c5 BW |
962 | enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
963 | struct request *req, | |
964 | struct bio *bio) | |
8e756373 BW |
965 | { |
966 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
967 | ||
968 | if (segments >= queue_max_discard_segments(q)) | |
969 | goto no_merge; | |
970 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
971 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
972 | goto no_merge; | |
973 | ||
974 | rq_qos_merge(q, req, bio); | |
975 | ||
976 | req->biotail->bi_next = bio; | |
977 | req->biotail = bio; | |
978 | req->__data_len += bio->bi_iter.bi_size; | |
979 | req->nr_phys_segments = segments + 1; | |
980 | ||
981 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 982 | return BIO_MERGE_OK; |
8e756373 BW |
983 | no_merge: |
984 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
985 | return BIO_MERGE_FAILED; |
986 | } | |
987 | ||
988 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
989 | struct request *rq, | |
990 | struct bio *bio, | |
991 | unsigned int nr_segs, | |
992 | bool sched_allow_merge) | |
993 | { | |
994 | if (!blk_rq_merge_ok(rq, bio)) | |
995 | return BIO_MERGE_NONE; | |
996 | ||
997 | switch (blk_try_merge(rq, bio)) { | |
998 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 999 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1000 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1001 | break; | |
1002 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1003 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1004 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1005 | break; | |
1006 | case ELEVATOR_DISCARD_MERGE: | |
1007 | return bio_attempt_discard_merge(q, rq, bio); | |
1008 | default: | |
1009 | return BIO_MERGE_NONE; | |
1010 | } | |
1011 | ||
1012 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1013 | } |
1014 | ||
1015 | /** | |
1016 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1017 | * @q: request_queue new bio is being queued at | |
1018 | * @bio: new bio being queued | |
1019 | * @nr_segs: number of segments in @bio | |
1020 | * @same_queue_rq: pointer to &struct request that gets filled in when | |
1021 | * another request associated with @q is found on the plug list | |
1022 | * (optional, may be %NULL) | |
1023 | * | |
1024 | * Determine whether @bio being queued on @q can be merged with a request | |
1025 | * on %current's plugged list. Returns %true if merge was successful, | |
1026 | * otherwise %false. | |
1027 | * | |
1028 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1029 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1030 | * than scheduling, and the request, while may have elvpriv data, is not | |
1031 | * added on the elevator at this point. In addition, we don't have | |
1032 | * reliable access to the elevator outside queue lock. Only check basic | |
1033 | * merging parameters without querying the elevator. | |
1034 | * | |
1035 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1036 | */ | |
1037 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
1038 | unsigned int nr_segs, struct request **same_queue_rq) | |
1039 | { | |
1040 | struct blk_plug *plug; | |
1041 | struct request *rq; | |
1042 | struct list_head *plug_list; | |
1043 | ||
1044 | plug = blk_mq_plug(q, bio); | |
1045 | if (!plug) | |
1046 | return false; | |
1047 | ||
1048 | plug_list = &plug->mq_list; | |
1049 | ||
1050 | list_for_each_entry_reverse(rq, plug_list, queuelist) { | |
8e756373 BW |
1051 | if (rq->q == q && same_queue_rq) { |
1052 | /* | |
1053 | * Only blk-mq multiple hardware queues case checks the | |
1054 | * rq in the same queue, there should be only one such | |
1055 | * rq in a queue | |
1056 | **/ | |
1057 | *same_queue_rq = rq; | |
1058 | } | |
1059 | ||
7d7ca7c5 | 1060 | if (rq->q != q) |
8e756373 BW |
1061 | continue; |
1062 | ||
7d7ca7c5 BW |
1063 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == |
1064 | BIO_MERGE_OK) | |
8e756373 BW |
1065 | return true; |
1066 | } | |
1067 | ||
1068 | return false; | |
1069 | } | |
bdc6a287 BW |
1070 | |
1071 | /* | |
1072 | * Iterate list of requests and see if we can merge this bio with any | |
1073 | * of them. | |
1074 | */ | |
1075 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1076 | struct bio *bio, unsigned int nr_segs) | |
1077 | { | |
1078 | struct request *rq; | |
1079 | int checked = 8; | |
1080 | ||
1081 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1082 | if (!checked--) |
1083 | break; | |
1084 | ||
7d7ca7c5 BW |
1085 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1086 | case BIO_MERGE_NONE: | |
bdc6a287 | 1087 | continue; |
7d7ca7c5 BW |
1088 | case BIO_MERGE_OK: |
1089 | return true; | |
1090 | case BIO_MERGE_FAILED: | |
1091 | return false; | |
bdc6a287 BW |
1092 | } |
1093 | ||
bdc6a287 BW |
1094 | } |
1095 | ||
1096 | return false; | |
1097 | } | |
1098 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); |