Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
d6d48196 JA |
2 | /* |
3 | * Functions related to segment and merge handling | |
4 | */ | |
5 | #include <linux/kernel.h> | |
6 | #include <linux/module.h> | |
7 | #include <linux/bio.h> | |
8 | #include <linux/blkdev.h> | |
9 | #include <linux/scatterlist.h> | |
10 | ||
cda22646 MK |
11 | #include <trace/events/block.h> |
12 | ||
d6d48196 JA |
13 | #include "blk.h" |
14 | ||
e9907009 CH |
15 | static inline bool bio_will_gap(struct request_queue *q, |
16 | struct request *prev_rq, struct bio *prev, struct bio *next) | |
17 | { | |
18 | struct bio_vec pb, nb; | |
19 | ||
20 | if (!bio_has_data(prev) || !queue_virt_boundary(q)) | |
21 | return false; | |
22 | ||
23 | /* | |
24 | * Don't merge if the 1st bio starts with non-zero offset, otherwise it | |
25 | * is quite difficult to respect the sg gap limit. We work hard to | |
26 | * merge a huge number of small single bios in case of mkfs. | |
27 | */ | |
28 | if (prev_rq) | |
29 | bio_get_first_bvec(prev_rq->bio, &pb); | |
30 | else | |
31 | bio_get_first_bvec(prev, &pb); | |
df376b2e | 32 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
33 | return true; |
34 | ||
35 | /* | |
36 | * We don't need to worry about the situation that the merged segment | |
37 | * ends in unaligned virt boundary: | |
38 | * | |
39 | * - if 'pb' ends aligned, the merged segment ends aligned | |
40 | * - if 'pb' ends unaligned, the next bio must include | |
41 | * one single bvec of 'nb', otherwise the 'nb' can't | |
42 | * merge with 'pb' | |
43 | */ | |
44 | bio_get_last_bvec(prev, &pb); | |
45 | bio_get_first_bvec(next, &nb); | |
200a9aff | 46 | if (biovec_phys_mergeable(q, &pb, &nb)) |
e9907009 CH |
47 | return false; |
48 | return __bvec_gap_to_prev(q, &pb, nb.bv_offset); | |
49 | } | |
50 | ||
51 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) | |
52 | { | |
53 | return bio_will_gap(req->q, req, req->biotail, bio); | |
54 | } | |
55 | ||
56 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) | |
57 | { | |
58 | return bio_will_gap(req->q, NULL, bio, req->bio); | |
59 | } | |
60 | ||
54efd50b KO |
61 | static struct bio *blk_bio_discard_split(struct request_queue *q, |
62 | struct bio *bio, | |
bdced438 ML |
63 | struct bio_set *bs, |
64 | unsigned *nsegs) | |
54efd50b KO |
65 | { |
66 | unsigned int max_discard_sectors, granularity; | |
67 | int alignment; | |
68 | sector_t tmp; | |
69 | unsigned split_sectors; | |
70 | ||
bdced438 ML |
71 | *nsegs = 1; |
72 | ||
54efd50b KO |
73 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
74 | granularity = max(q->limits.discard_granularity >> 9, 1U); | |
75 | ||
1adfc5e4 ML |
76 | max_discard_sectors = min(q->limits.max_discard_sectors, |
77 | bio_allowed_max_sectors(q)); | |
54efd50b KO |
78 | max_discard_sectors -= max_discard_sectors % granularity; |
79 | ||
80 | if (unlikely(!max_discard_sectors)) { | |
81 | /* XXX: warn */ | |
82 | return NULL; | |
83 | } | |
84 | ||
85 | if (bio_sectors(bio) <= max_discard_sectors) | |
86 | return NULL; | |
87 | ||
88 | split_sectors = max_discard_sectors; | |
89 | ||
90 | /* | |
91 | * If the next starting sector would be misaligned, stop the discard at | |
92 | * the previous aligned sector. | |
93 | */ | |
94 | alignment = (q->limits.discard_alignment >> 9) % granularity; | |
95 | ||
96 | tmp = bio->bi_iter.bi_sector + split_sectors - alignment; | |
97 | tmp = sector_div(tmp, granularity); | |
98 | ||
99 | if (split_sectors > tmp) | |
100 | split_sectors -= tmp; | |
101 | ||
102 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
103 | } | |
104 | ||
885fa13f CH |
105 | static struct bio *blk_bio_write_zeroes_split(struct request_queue *q, |
106 | struct bio *bio, struct bio_set *bs, unsigned *nsegs) | |
107 | { | |
d665e12a | 108 | *nsegs = 0; |
885fa13f CH |
109 | |
110 | if (!q->limits.max_write_zeroes_sectors) | |
111 | return NULL; | |
112 | ||
113 | if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors) | |
114 | return NULL; | |
115 | ||
116 | return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs); | |
117 | } | |
118 | ||
54efd50b KO |
119 | static struct bio *blk_bio_write_same_split(struct request_queue *q, |
120 | struct bio *bio, | |
bdced438 ML |
121 | struct bio_set *bs, |
122 | unsigned *nsegs) | |
54efd50b | 123 | { |
bdced438 ML |
124 | *nsegs = 1; |
125 | ||
54efd50b KO |
126 | if (!q->limits.max_write_same_sectors) |
127 | return NULL; | |
128 | ||
129 | if (bio_sectors(bio) <= q->limits.max_write_same_sectors) | |
130 | return NULL; | |
131 | ||
132 | return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs); | |
133 | } | |
134 | ||
9cc5169c BVA |
135 | /* |
136 | * Return the maximum number of sectors from the start of a bio that may be | |
137 | * submitted as a single request to a block device. If enough sectors remain, | |
138 | * align the end to the physical block size. Otherwise align the end to the | |
139 | * logical block size. This approach minimizes the number of non-aligned | |
140 | * requests that are submitted to a block device if the start of a bio is not | |
141 | * aligned to a physical block boundary. | |
142 | */ | |
d0e5fbb0 ML |
143 | static inline unsigned get_max_io_size(struct request_queue *q, |
144 | struct bio *bio) | |
145 | { | |
146 | unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); | |
9cc5169c BVA |
147 | unsigned max_sectors = sectors; |
148 | unsigned pbs = queue_physical_block_size(q) >> SECTOR_SHIFT; | |
149 | unsigned lbs = queue_logical_block_size(q) >> SECTOR_SHIFT; | |
150 | unsigned start_offset = bio->bi_iter.bi_sector & (pbs - 1); | |
d0e5fbb0 | 151 | |
9cc5169c BVA |
152 | max_sectors += start_offset; |
153 | max_sectors &= ~(pbs - 1); | |
154 | if (max_sectors > start_offset) | |
155 | return max_sectors - start_offset; | |
d0e5fbb0 | 156 | |
e4b469c6 | 157 | return sectors & ~(lbs - 1); |
d0e5fbb0 ML |
158 | } |
159 | ||
429120f3 ML |
160 | static inline unsigned get_max_segment_size(const struct request_queue *q, |
161 | struct page *start_page, | |
162 | unsigned long offset) | |
dcebd755 ML |
163 | { |
164 | unsigned long mask = queue_segment_boundary(q); | |
165 | ||
429120f3 | 166 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
167 | |
168 | /* | |
169 | * overflow may be triggered in case of zero page physical address | |
170 | * on 32bit arch, use queue's max segment size when that happens. | |
171 | */ | |
172 | return min_not_zero(mask - offset + 1, | |
173 | (unsigned long)queue_max_segment_size(q)); | |
dcebd755 ML |
174 | } |
175 | ||
708b25b3 BVA |
176 | /** |
177 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
178 | * @q: [in] request queue associated with the bio associated with @bv | |
179 | * @bv: [in] bvec to examine | |
180 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
181 | * by the number of segments from @bv that may be appended to that | |
182 | * bio without exceeding @max_segs | |
183 | * @sectors: [in,out] Number of sectors in the bio being built. Incremented | |
184 | * by the number of sectors from @bv that may be appended to that | |
185 | * bio without exceeding @max_sectors | |
186 | * @max_segs: [in] upper bound for *@nsegs | |
187 | * @max_sectors: [in] upper bound for *@sectors | |
188 | * | |
189 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
190 | * big to fit in a single segment and hence that it has to be split in the | |
191 | * middle. This function verifies whether or not that should happen. The value | |
192 | * %true is returned if and only if appending the entire @bv to a bio with | |
193 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
194 | * the block driver. | |
dcebd755 | 195 | */ |
af2c68fe BVA |
196 | static bool bvec_split_segs(const struct request_queue *q, |
197 | const struct bio_vec *bv, unsigned *nsegs, | |
708b25b3 BVA |
198 | unsigned *sectors, unsigned max_segs, |
199 | unsigned max_sectors) | |
dcebd755 | 200 | { |
708b25b3 BVA |
201 | unsigned max_len = (min(max_sectors, UINT_MAX >> 9) - *sectors) << 9; |
202 | unsigned len = min(bv->bv_len, max_len); | |
dcebd755 | 203 | unsigned total_len = 0; |
ff9811b3 | 204 | unsigned seg_size = 0; |
dcebd755 | 205 | |
ff9811b3 | 206 | while (len && *nsegs < max_segs) { |
429120f3 ML |
207 | seg_size = get_max_segment_size(q, bv->bv_page, |
208 | bv->bv_offset + total_len); | |
dcebd755 ML |
209 | seg_size = min(seg_size, len); |
210 | ||
ff9811b3 | 211 | (*nsegs)++; |
dcebd755 ML |
212 | total_len += seg_size; |
213 | len -= seg_size; | |
214 | ||
215 | if ((bv->bv_offset + total_len) & queue_virt_boundary(q)) | |
216 | break; | |
217 | } | |
218 | ||
ff9811b3 | 219 | *sectors += total_len >> 9; |
dcebd755 | 220 | |
708b25b3 BVA |
221 | /* tell the caller to split the bvec if it is too big to fit */ |
222 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
223 | } |
224 | ||
dad77584 BVA |
225 | /** |
226 | * blk_bio_segment_split - split a bio in two bios | |
227 | * @q: [in] request queue pointer | |
228 | * @bio: [in] bio to be split | |
229 | * @bs: [in] bio set to allocate the clone from | |
230 | * @segs: [out] number of segments in the bio with the first half of the sectors | |
231 | * | |
232 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
233 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
234 | * following is guaranteed for the cloned bio: | |
235 | * - That it has at most get_max_io_size(@q, @bio) sectors. | |
236 | * - That it has at most queue_max_segments(@q) segments. | |
237 | * | |
238 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
239 | * the original bio. It is the responsibility of the caller to ensure that the | |
240 | * original bio is not freed before the cloned bio. The caller is also | |
241 | * responsible for ensuring that @bs is only destroyed after processing of the | |
242 | * split bio has finished. | |
243 | */ | |
54efd50b KO |
244 | static struct bio *blk_bio_segment_split(struct request_queue *q, |
245 | struct bio *bio, | |
bdced438 ML |
246 | struct bio_set *bs, |
247 | unsigned *segs) | |
54efd50b | 248 | { |
5014c311 | 249 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 250 | struct bvec_iter iter; |
6869875f | 251 | unsigned nsegs = 0, sectors = 0; |
d0e5fbb0 | 252 | const unsigned max_sectors = get_max_io_size(q, bio); |
05b700ba | 253 | const unsigned max_segs = queue_max_segments(q); |
54efd50b | 254 | |
dcebd755 | 255 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
256 | /* |
257 | * If the queue doesn't support SG gaps and adding this | |
258 | * offset would create a gap, disallow it. | |
259 | */ | |
5014c311 | 260 | if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset)) |
54efd50b KO |
261 | goto split; |
262 | ||
708b25b3 BVA |
263 | if (nsegs < max_segs && |
264 | sectors + (bv.bv_len >> 9) <= max_sectors && | |
265 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { | |
266 | nsegs++; | |
267 | sectors += bv.bv_len >> 9; | |
268 | } else if (bvec_split_segs(q, &bv, &nsegs, §ors, max_segs, | |
269 | max_sectors)) { | |
cf8c0c6a | 270 | goto split; |
e36f6204 KB |
271 | } |
272 | ||
54efd50b | 273 | bvprv = bv; |
578270bf | 274 | bvprvp = &bvprv; |
54efd50b KO |
275 | } |
276 | ||
d627065d CH |
277 | *segs = nsegs; |
278 | return NULL; | |
54efd50b | 279 | split: |
bdced438 | 280 | *segs = nsegs; |
d627065d | 281 | return bio_split(bio, sectors, GFP_NOIO, bs); |
54efd50b KO |
282 | } |
283 | ||
dad77584 BVA |
284 | /** |
285 | * __blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
286 | * @bio: [in, out] bio to be split |
287 | * @nr_segs: [out] number of segments in the first bio | |
288 | * | |
289 | * Split a bio into two bios, chain the two bios, submit the second half and | |
290 | * store a pointer to the first half in *@bio. If the second bio is still too | |
291 | * big it will be split by a recursive call to this function. Since this | |
f695ca38 CH |
292 | * function may allocate a new bio from @bio->bi_disk->queue->bio_split, it is |
293 | * the responsibility of the caller to ensure that | |
294 | * @bio->bi_disk->queue->bio_split is only released after processing of the | |
dad77584 BVA |
295 | * split bio has finished. |
296 | */ | |
f695ca38 | 297 | void __blk_queue_split(struct bio **bio, unsigned int *nr_segs) |
54efd50b | 298 | { |
f695ca38 | 299 | struct request_queue *q = (*bio)->bi_disk->queue; |
fa532287 | 300 | struct bio *split = NULL; |
54efd50b | 301 | |
7afafc8a AH |
302 | switch (bio_op(*bio)) { |
303 | case REQ_OP_DISCARD: | |
304 | case REQ_OP_SECURE_ERASE: | |
14ccb66b | 305 | split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a | 306 | break; |
a6f0788e | 307 | case REQ_OP_WRITE_ZEROES: |
14ccb66b CH |
308 | split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split, |
309 | nr_segs); | |
a6f0788e | 310 | break; |
7afafc8a | 311 | case REQ_OP_WRITE_SAME: |
14ccb66b CH |
312 | split = blk_bio_write_same_split(q, *bio, &q->bio_split, |
313 | nr_segs); | |
7afafc8a AH |
314 | break; |
315 | default: | |
fa532287 CH |
316 | /* |
317 | * All drivers must accept single-segments bios that are <= | |
318 | * PAGE_SIZE. This is a quick and dirty check that relies on | |
319 | * the fact that bi_io_vec[0] is always valid if a bio has data. | |
320 | * The check might lead to occasional false negatives when bios | |
321 | * are cloned, but compared to the performance impact of cloned | |
322 | * bios themselves the loop below doesn't matter anyway. | |
323 | */ | |
324 | if (!q->limits.chunk_sectors && | |
325 | (*bio)->bi_vcnt == 1 && | |
59db8ba2 | 326 | ((*bio)->bi_io_vec[0].bv_len + |
1e279153 | 327 | (*bio)->bi_io_vec[0].bv_offset) <= PAGE_SIZE) { |
fa532287 CH |
328 | *nr_segs = 1; |
329 | break; | |
330 | } | |
14ccb66b | 331 | split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs); |
7afafc8a AH |
332 | break; |
333 | } | |
bdced438 | 334 | |
54efd50b | 335 | if (split) { |
6ac45aeb | 336 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 337 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 338 | |
54efd50b | 339 | bio_chain(split, *bio); |
cda22646 | 340 | trace_block_split(q, split, (*bio)->bi_iter.bi_sector); |
ed00aabd | 341 | submit_bio_noacct(*bio); |
54efd50b KO |
342 | *bio = split; |
343 | } | |
344 | } | |
14ccb66b | 345 | |
dad77584 BVA |
346 | /** |
347 | * blk_queue_split - split a bio and submit the second half | |
dad77584 BVA |
348 | * @bio: [in, out] bio to be split |
349 | * | |
350 | * Split a bio into two bios, chains the two bios, submit the second half and | |
351 | * store a pointer to the first half in *@bio. Since this function may allocate | |
f695ca38 CH |
352 | * a new bio from @bio->bi_disk->queue->bio_split, it is the responsibility of |
353 | * the caller to ensure that @bio->bi_disk->queue->bio_split is only released | |
354 | * after processing of the split bio has finished. | |
dad77584 | 355 | */ |
f695ca38 | 356 | void blk_queue_split(struct bio **bio) |
14ccb66b CH |
357 | { |
358 | unsigned int nr_segs; | |
359 | ||
f695ca38 | 360 | __blk_queue_split(bio, &nr_segs); |
14ccb66b | 361 | } |
54efd50b KO |
362 | EXPORT_SYMBOL(blk_queue_split); |
363 | ||
e9cd19c0 | 364 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 365 | { |
6869875f | 366 | unsigned int nr_phys_segs = 0; |
ff9811b3 | 367 | unsigned int nr_sectors = 0; |
e9cd19c0 | 368 | struct req_iterator iter; |
6869875f | 369 | struct bio_vec bv; |
d6d48196 | 370 | |
e9cd19c0 | 371 | if (!rq->bio) |
1e428079 | 372 | return 0; |
d6d48196 | 373 | |
e9cd19c0 | 374 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
375 | case REQ_OP_DISCARD: |
376 | case REQ_OP_SECURE_ERASE: | |
a6f0788e | 377 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 CH |
378 | return 0; |
379 | case REQ_OP_WRITE_SAME: | |
5cb8850c | 380 | return 1; |
a6f0788e | 381 | } |
5cb8850c | 382 | |
e9cd19c0 | 383 | rq_for_each_bvec(bv, rq, iter) |
ff9811b3 | 384 | bvec_split_segs(rq->q, &bv, &nr_phys_segs, &nr_sectors, |
708b25b3 | 385 | UINT_MAX, UINT_MAX); |
1e428079 JA |
386 | return nr_phys_segs; |
387 | } | |
388 | ||
48d7727c | 389 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
390 | struct scatterlist *sglist) |
391 | { | |
392 | if (!*sg) | |
393 | return sglist; | |
394 | ||
395 | /* | |
396 | * If the driver previously mapped a shorter list, we could see a | |
397 | * termination bit prematurely unless it fully inits the sg table | |
398 | * on each mapping. We KNOW that there must be more entries here | |
399 | * or the driver would be buggy, so force clear the termination bit | |
400 | * to avoid doing a full sg_init_table() in drivers for each command. | |
401 | */ | |
402 | sg_unmark_end(*sg); | |
403 | return sg_next(*sg); | |
404 | } | |
405 | ||
406 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
407 | struct bio_vec *bvec, struct scatterlist *sglist, | |
408 | struct scatterlist **sg) | |
409 | { | |
410 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 411 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
412 | |
413 | while (nbytes > 0) { | |
8a96a0e4 | 414 | unsigned offset = bvec->bv_offset + total; |
429120f3 ML |
415 | unsigned len = min(get_max_segment_size(q, bvec->bv_page, |
416 | offset), nbytes); | |
f9f76879 CH |
417 | struct page *page = bvec->bv_page; |
418 | ||
419 | /* | |
420 | * Unfortunately a fair number of drivers barf on scatterlists | |
421 | * that have an offset larger than PAGE_SIZE, despite other | |
422 | * subsystems dealing with that invariant just fine. For now | |
423 | * stick to the legacy format where we never present those from | |
424 | * the block layer, but the code below should be removed once | |
425 | * these offenders (mostly MMC/SD drivers) are fixed. | |
426 | */ | |
427 | page += (offset >> PAGE_SHIFT); | |
428 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
429 | |
430 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 431 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 432 | |
8a96a0e4 CH |
433 | total += len; |
434 | nbytes -= len; | |
862e5a5e ML |
435 | nsegs++; |
436 | } | |
437 | ||
438 | return nsegs; | |
439 | } | |
440 | ||
16e3e418 ML |
441 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
442 | struct scatterlist *sglist, struct scatterlist **sg) | |
443 | { | |
444 | *sg = blk_next_sg(sg, sglist); | |
445 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
446 | return 1; | |
447 | } | |
448 | ||
f6970f83 ML |
449 | /* only try to merge bvecs into one sg if they are from two bios */ |
450 | static inline bool | |
451 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
452 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
453 | { |
454 | ||
455 | int nbytes = bvec->bv_len; | |
456 | ||
f6970f83 ML |
457 | if (!*sg) |
458 | return false; | |
963ab9e5 | 459 | |
f6970f83 ML |
460 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
461 | return false; | |
462 | ||
463 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
464 | return false; | |
465 | ||
466 | (*sg)->length += nbytes; | |
467 | ||
468 | return true; | |
963ab9e5 AH |
469 | } |
470 | ||
5cb8850c KO |
471 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
472 | struct scatterlist *sglist, | |
473 | struct scatterlist **sg) | |
d6d48196 | 474 | { |
3f649ab7 | 475 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 476 | struct bvec_iter iter; |
38417468 | 477 | int nsegs = 0; |
f6970f83 | 478 | bool new_bio = false; |
5cb8850c | 479 | |
f6970f83 ML |
480 | for_each_bio(bio) { |
481 | bio_for_each_bvec(bvec, bio, iter) { | |
482 | /* | |
483 | * Only try to merge bvecs from two bios given we | |
484 | * have done bio internal merge when adding pages | |
485 | * to bio | |
486 | */ | |
487 | if (new_bio && | |
488 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
489 | goto next_bvec; | |
490 | ||
491 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
492 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
493 | else | |
494 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
495 | next_bvec: | |
496 | new_bio = false; | |
497 | } | |
b21e11c5 ML |
498 | if (likely(bio->bi_iter.bi_size)) { |
499 | bvprv = bvec; | |
500 | new_bio = true; | |
501 | } | |
f6970f83 | 502 | } |
d6d48196 | 503 | |
5cb8850c KO |
504 | return nsegs; |
505 | } | |
506 | ||
507 | /* | |
508 | * map a request to scatterlist, return number of sg entries setup. Caller | |
509 | * must make sure sg can hold rq->nr_phys_segments entries | |
510 | */ | |
89de1504 CH |
511 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
512 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 513 | { |
5cb8850c KO |
514 | int nsegs = 0; |
515 | ||
f9d03f96 | 516 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 517 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 518 | else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME) |
89de1504 | 519 | nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, last_sg); |
f9d03f96 | 520 | else if (rq->bio) |
89de1504 | 521 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 522 | |
89de1504 CH |
523 | if (*last_sg) |
524 | sg_mark_end(*last_sg); | |
d6d48196 | 525 | |
12e57f59 ML |
526 | /* |
527 | * Something must have been wrong if the figured number of | |
528 | * segment is bigger than number of req's physical segments | |
529 | */ | |
f9d03f96 | 530 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 531 | |
d6d48196 JA |
532 | return nsegs; |
533 | } | |
89de1504 | 534 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 535 | |
943b40c8 ML |
536 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
537 | { | |
538 | if (req_op(rq) == REQ_OP_DISCARD) | |
539 | return queue_max_discard_segments(rq->q); | |
540 | return queue_max_segments(rq->q); | |
541 | } | |
542 | ||
14ccb66b CH |
543 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
544 | unsigned int nr_phys_segs) | |
d6d48196 | 545 | { |
943b40c8 | 546 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) |
13f05c8d MP |
547 | goto no_merge; |
548 | ||
14ccb66b | 549 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d | 550 | goto no_merge; |
d6d48196 JA |
551 | |
552 | /* | |
553 | * This will form the start of a new hw segment. Bump both | |
554 | * counters. | |
555 | */ | |
d6d48196 JA |
556 | req->nr_phys_segments += nr_phys_segs; |
557 | return 1; | |
13f05c8d MP |
558 | |
559 | no_merge: | |
14ccb66b | 560 | req_set_nomerge(req->q, req); |
13f05c8d | 561 | return 0; |
d6d48196 JA |
562 | } |
563 | ||
14ccb66b | 564 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 565 | { |
5e7c4274 JA |
566 | if (req_gap_back_merge(req, bio)) |
567 | return 0; | |
7f39add3 SG |
568 | if (blk_integrity_rq(req) && |
569 | integrity_req_gap_back_merge(req, bio)) | |
570 | return 0; | |
a892c8d5 ST |
571 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
572 | return 0; | |
f31dc1cd | 573 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 574 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 575 | req_set_nomerge(req->q, req); |
d6d48196 JA |
576 | return 0; |
577 | } | |
d6d48196 | 578 | |
14ccb66b | 579 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
580 | } |
581 | ||
14ccb66b | 582 | int ll_front_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 583 | { |
5e7c4274 JA |
584 | if (req_gap_front_merge(req, bio)) |
585 | return 0; | |
7f39add3 SG |
586 | if (blk_integrity_rq(req) && |
587 | integrity_req_gap_front_merge(req, bio)) | |
588 | return 0; | |
a892c8d5 ST |
589 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
590 | return 0; | |
f31dc1cd | 591 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 592 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 593 | req_set_nomerge(req->q, req); |
d6d48196 JA |
594 | return 0; |
595 | } | |
d6d48196 | 596 | |
14ccb66b | 597 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
598 | } |
599 | ||
445251d0 JA |
600 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
601 | struct request *next) | |
602 | { | |
603 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
604 | ||
605 | if (segments >= queue_max_discard_segments(q)) | |
606 | goto no_merge; | |
607 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
608 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
609 | goto no_merge; | |
610 | ||
611 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
612 | return true; | |
613 | no_merge: | |
614 | req_set_nomerge(q, req); | |
615 | return false; | |
616 | } | |
617 | ||
d6d48196 JA |
618 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
619 | struct request *next) | |
620 | { | |
621 | int total_phys_segments; | |
d6d48196 | 622 | |
5e7c4274 | 623 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
624 | return 0; |
625 | ||
d6d48196 JA |
626 | /* |
627 | * Will it become too large? | |
628 | */ | |
f31dc1cd | 629 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 630 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
631 | return 0; |
632 | ||
633 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 634 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
635 | return 0; |
636 | ||
4eaf99be | 637 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
638 | return 0; |
639 | ||
a892c8d5 ST |
640 | if (!bio_crypt_ctx_merge_rq(req, next)) |
641 | return 0; | |
642 | ||
d6d48196 JA |
643 | /* Merge is OK... */ |
644 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
645 | return 1; |
646 | } | |
647 | ||
80a761fd TH |
648 | /** |
649 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
650 | * @rq: request to mark as mixed merge | |
651 | * | |
652 | * Description: | |
653 | * @rq is about to be mixed merged. Make sure the attributes | |
654 | * which can be mixed are set in each bio and mark @rq as mixed | |
655 | * merged. | |
656 | */ | |
657 | void blk_rq_set_mixed_merge(struct request *rq) | |
658 | { | |
659 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; | |
660 | struct bio *bio; | |
661 | ||
e8064021 | 662 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
663 | return; |
664 | ||
665 | /* | |
666 | * @rq will no longer represent mixable attributes for all the | |
667 | * contained bios. It will just track those of the first one. | |
668 | * Distributes the attributs to each bio. | |
669 | */ | |
670 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
671 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
672 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
673 | bio->bi_opf |= ff; | |
80a761fd | 674 | } |
e8064021 | 675 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
676 | } |
677 | ||
b9c54f56 | 678 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
679 | { |
680 | if (blk_do_io_stat(req)) { | |
112f158f | 681 | part_stat_lock(); |
b9c54f56 | 682 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab | 683 | part_stat_unlock(); |
524f9ffd CH |
684 | |
685 | hd_struct_put(req->part); | |
26308eab JM |
686 | } |
687 | } | |
b9c54f56 | 688 | |
69840466 JW |
689 | /* |
690 | * Two cases of handling DISCARD merge: | |
691 | * If max_discard_segments > 1, the driver takes every bio | |
692 | * as a range and send them to controller together. The ranges | |
693 | * needn't to be contiguous. | |
694 | * Otherwise, the bios/requests will be handled as same as | |
695 | * others which should be contiguous. | |
696 | */ | |
697 | static inline bool blk_discard_mergable(struct request *req) | |
698 | { | |
699 | if (req_op(req) == REQ_OP_DISCARD && | |
700 | queue_max_discard_segments(req->q) > 1) | |
701 | return true; | |
702 | return false; | |
703 | } | |
704 | ||
e96c0d83 EB |
705 | static enum elv_merge blk_try_req_merge(struct request *req, |
706 | struct request *next) | |
69840466 JW |
707 | { |
708 | if (blk_discard_mergable(req)) | |
709 | return ELEVATOR_DISCARD_MERGE; | |
710 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
711 | return ELEVATOR_BACK_MERGE; | |
712 | ||
713 | return ELEVATOR_NO_MERGE; | |
714 | } | |
26308eab | 715 | |
d6d48196 | 716 | /* |
b973cb7e JA |
717 | * For non-mq, this has to be called with the request spinlock acquired. |
718 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 719 | */ |
b973cb7e JA |
720 | static struct request *attempt_merge(struct request_queue *q, |
721 | struct request *req, struct request *next) | |
d6d48196 JA |
722 | { |
723 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 724 | return NULL; |
d6d48196 | 725 | |
288dab8a | 726 | if (req_op(req) != req_op(next)) |
b973cb7e | 727 | return NULL; |
f31dc1cd | 728 | |
d6d48196 | 729 | if (rq_data_dir(req) != rq_data_dir(next) |
2081a56b | 730 | || req->rq_disk != next->rq_disk) |
b973cb7e | 731 | return NULL; |
d6d48196 | 732 | |
8fe0d473 | 733 | if (req_op(req) == REQ_OP_WRITE_SAME && |
4363ac7c | 734 | !blk_write_same_mergeable(req->bio, next->bio)) |
b973cb7e | 735 | return NULL; |
4363ac7c | 736 | |
cb6934f8 JA |
737 | /* |
738 | * Don't allow merge of different write hints, or for a hint with | |
739 | * non-hint IO. | |
740 | */ | |
741 | if (req->write_hint != next->write_hint) | |
742 | return NULL; | |
743 | ||
668ffc03 DLM |
744 | if (req->ioprio != next->ioprio) |
745 | return NULL; | |
746 | ||
d6d48196 JA |
747 | /* |
748 | * If we are allowed to merge, then append bio list | |
749 | * from next to rq and release next. merge_requests_fn | |
750 | * will have updated segment counts, update sector | |
445251d0 JA |
751 | * counts here. Handle DISCARDs separately, as they |
752 | * have separate settings. | |
d6d48196 | 753 | */ |
69840466 JW |
754 | |
755 | switch (blk_try_req_merge(req, next)) { | |
756 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
757 | if (!req_attempt_discard_merge(q, req, next)) |
758 | return NULL; | |
69840466 JW |
759 | break; |
760 | case ELEVATOR_BACK_MERGE: | |
761 | if (!ll_merge_requests_fn(q, req, next)) | |
762 | return NULL; | |
763 | break; | |
764 | default: | |
b973cb7e | 765 | return NULL; |
69840466 | 766 | } |
d6d48196 | 767 | |
80a761fd TH |
768 | /* |
769 | * If failfast settings disagree or any of the two is already | |
770 | * a mixed merge, mark both as mixed before proceeding. This | |
771 | * makes sure that all involved bios have mixable attributes | |
772 | * set properly. | |
773 | */ | |
e8064021 | 774 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
775 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
776 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
777 | blk_rq_set_mixed_merge(req); | |
778 | blk_rq_set_mixed_merge(next); | |
779 | } | |
780 | ||
d6d48196 | 781 | /* |
522a7775 OS |
782 | * At this point we have either done a back merge or front merge. We |
783 | * need the smaller start_time_ns of the merged requests to be the | |
784 | * current request for accounting purposes. | |
d6d48196 | 785 | */ |
522a7775 OS |
786 | if (next->start_time_ns < req->start_time_ns) |
787 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
788 | |
789 | req->biotail->bi_next = next->bio; | |
790 | req->biotail = next->biotail; | |
791 | ||
a2dec7b3 | 792 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 793 | |
2a5cf35c | 794 | if (!blk_discard_mergable(req)) |
445251d0 | 795 | elv_merge_requests(q, req, next); |
d6d48196 | 796 | |
42dad764 JM |
797 | /* |
798 | * 'next' is going away, so update stats accordingly | |
799 | */ | |
b9c54f56 | 800 | blk_account_io_merge_request(next); |
d6d48196 | 801 | |
f3bdc62f JK |
802 | trace_block_rq_merge(q, next); |
803 | ||
e4d750c9 JA |
804 | /* |
805 | * ownership of bio passed from next to req, return 'next' for | |
806 | * the caller to free | |
807 | */ | |
1cd96c24 | 808 | next->bio = NULL; |
b973cb7e | 809 | return next; |
d6d48196 JA |
810 | } |
811 | ||
b973cb7e | 812 | struct request *attempt_back_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
813 | { |
814 | struct request *next = elv_latter_request(q, rq); | |
815 | ||
816 | if (next) | |
817 | return attempt_merge(q, rq, next); | |
818 | ||
b973cb7e | 819 | return NULL; |
d6d48196 JA |
820 | } |
821 | ||
b973cb7e | 822 | struct request *attempt_front_merge(struct request_queue *q, struct request *rq) |
d6d48196 JA |
823 | { |
824 | struct request *prev = elv_former_request(q, rq); | |
825 | ||
826 | if (prev) | |
827 | return attempt_merge(q, prev, rq); | |
828 | ||
b973cb7e | 829 | return NULL; |
d6d48196 | 830 | } |
5e84ea3a JA |
831 | |
832 | int blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
833 | struct request *next) | |
834 | { | |
e4d750c9 | 835 | struct request *free; |
72ef799b | 836 | |
e4d750c9 JA |
837 | free = attempt_merge(q, rq, next); |
838 | if (free) { | |
92bc5a24 | 839 | blk_put_request(free); |
e4d750c9 JA |
840 | return 1; |
841 | } | |
842 | ||
843 | return 0; | |
5e84ea3a | 844 | } |
050c8ea8 TH |
845 | |
846 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
847 | { | |
e2a60da7 | 848 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
849 | return false; |
850 | ||
288dab8a | 851 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
852 | return false; |
853 | ||
050c8ea8 TH |
854 | /* different data direction or already started, don't merge */ |
855 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
856 | return false; | |
857 | ||
2081a56b JA |
858 | /* must be same device */ |
859 | if (rq->rq_disk != bio->bi_disk) | |
050c8ea8 TH |
860 | return false; |
861 | ||
862 | /* only merge integrity protected bio into ditto rq */ | |
4eaf99be | 863 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
864 | return false; |
865 | ||
a892c8d5 ST |
866 | /* Only merge if the crypt contexts are compatible */ |
867 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
868 | return false; | |
869 | ||
4363ac7c | 870 | /* must be using the same buffer */ |
8fe0d473 | 871 | if (req_op(rq) == REQ_OP_WRITE_SAME && |
4363ac7c MP |
872 | !blk_write_same_mergeable(rq->bio, bio)) |
873 | return false; | |
874 | ||
cb6934f8 JA |
875 | /* |
876 | * Don't allow merge of different write hints, or for a hint with | |
877 | * non-hint IO. | |
878 | */ | |
879 | if (rq->write_hint != bio->bi_write_hint) | |
880 | return false; | |
881 | ||
668ffc03 DLM |
882 | if (rq->ioprio != bio_prio(bio)) |
883 | return false; | |
884 | ||
050c8ea8 TH |
885 | return true; |
886 | } | |
887 | ||
34fe7c05 | 888 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 889 | { |
69840466 | 890 | if (blk_discard_mergable(rq)) |
1e739730 CH |
891 | return ELEVATOR_DISCARD_MERGE; |
892 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 893 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 894 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
895 | return ELEVATOR_FRONT_MERGE; |
896 | return ELEVATOR_NO_MERGE; | |
897 | } |