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> | |
fe45e630 | 9 | #include <linux/blk-integrity.h> |
d6d48196 | 10 | #include <linux/scatterlist.h> |
82d981d4 | 11 | #include <linux/part_stat.h> |
6b2b0459 | 12 | #include <linux/blk-cgroup.h> |
d6d48196 | 13 | |
cda22646 MK |
14 | #include <trace/events/block.h> |
15 | ||
d6d48196 | 16 | #include "blk.h" |
2aa7745b | 17 | #include "blk-mq-sched.h" |
8e756373 | 18 | #include "blk-rq-qos.h" |
a7b36ee6 | 19 | #include "blk-throttle.h" |
d6d48196 | 20 | |
ff18d77b CH |
21 | static inline void bio_get_first_bvec(struct bio *bio, struct bio_vec *bv) |
22 | { | |
23 | *bv = mp_bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
24 | } | |
25 | ||
26 | static inline void bio_get_last_bvec(struct bio *bio, struct bio_vec *bv) | |
27 | { | |
28 | struct bvec_iter iter = bio->bi_iter; | |
29 | int idx; | |
30 | ||
31 | bio_get_first_bvec(bio, bv); | |
32 | if (bv->bv_len == bio->bi_iter.bi_size) | |
33 | return; /* this bio only has a single bvec */ | |
34 | ||
35 | bio_advance_iter(bio, &iter, iter.bi_size); | |
36 | ||
37 | if (!iter.bi_bvec_done) | |
38 | idx = iter.bi_idx - 1; | |
39 | else /* in the middle of bvec */ | |
40 | idx = iter.bi_idx; | |
41 | ||
42 | *bv = bio->bi_io_vec[idx]; | |
43 | ||
44 | /* | |
45 | * iter.bi_bvec_done records actual length of the last bvec | |
46 | * if this bio ends in the middle of one io vector | |
47 | */ | |
48 | if (iter.bi_bvec_done) | |
49 | bv->bv_len = iter.bi_bvec_done; | |
50 | } | |
51 | ||
e9907009 CH |
52 | static inline bool bio_will_gap(struct request_queue *q, |
53 | struct request *prev_rq, struct bio *prev, struct bio *next) | |
54 | { | |
55 | struct bio_vec pb, nb; | |
56 | ||
57 | if (!bio_has_data(prev) || !queue_virt_boundary(q)) | |
58 | return false; | |
59 | ||
60 | /* | |
61 | * Don't merge if the 1st bio starts with non-zero offset, otherwise it | |
62 | * is quite difficult to respect the sg gap limit. We work hard to | |
63 | * merge a huge number of small single bios in case of mkfs. | |
64 | */ | |
65 | if (prev_rq) | |
66 | bio_get_first_bvec(prev_rq->bio, &pb); | |
67 | else | |
68 | bio_get_first_bvec(prev, &pb); | |
df376b2e | 69 | if (pb.bv_offset & queue_virt_boundary(q)) |
e9907009 CH |
70 | return true; |
71 | ||
72 | /* | |
73 | * We don't need to worry about the situation that the merged segment | |
74 | * ends in unaligned virt boundary: | |
75 | * | |
76 | * - if 'pb' ends aligned, the merged segment ends aligned | |
77 | * - if 'pb' ends unaligned, the next bio must include | |
78 | * one single bvec of 'nb', otherwise the 'nb' can't | |
79 | * merge with 'pb' | |
80 | */ | |
81 | bio_get_last_bvec(prev, &pb); | |
82 | bio_get_first_bvec(next, &nb); | |
200a9aff | 83 | if (biovec_phys_mergeable(q, &pb, &nb)) |
e9907009 | 84 | return false; |
c55ddd90 | 85 | return __bvec_gap_to_prev(&q->limits, &pb, nb.bv_offset); |
e9907009 CH |
86 | } |
87 | ||
88 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) | |
89 | { | |
90 | return bio_will_gap(req->q, req, req->biotail, bio); | |
91 | } | |
92 | ||
93 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) | |
94 | { | |
95 | return bio_will_gap(req->q, NULL, bio, req->bio); | |
96 | } | |
97 | ||
b6dc6198 CH |
98 | /* |
99 | * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size | |
100 | * is defined as 'unsigned int', meantime it has to be aligned to with the | |
101 | * logical block size, which is the minimum accepted unit by hardware. | |
102 | */ | |
c55ddd90 | 103 | static unsigned int bio_allowed_max_sectors(struct queue_limits *lim) |
b6dc6198 | 104 | { |
c55ddd90 | 105 | return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT; |
b6dc6198 CH |
106 | } |
107 | ||
c55ddd90 | 108 | static struct bio *bio_split_discard(struct bio *bio, struct queue_limits *lim, |
5a97806f | 109 | unsigned *nsegs, struct bio_set *bs) |
54efd50b KO |
110 | { |
111 | unsigned int max_discard_sectors, granularity; | |
54efd50b KO |
112 | sector_t tmp; |
113 | unsigned split_sectors; | |
114 | ||
bdced438 ML |
115 | *nsegs = 1; |
116 | ||
54efd50b | 117 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
c55ddd90 | 118 | granularity = max(lim->discard_granularity >> 9, 1U); |
54efd50b | 119 | |
c55ddd90 CH |
120 | max_discard_sectors = |
121 | min(lim->max_discard_sectors, bio_allowed_max_sectors(lim)); | |
54efd50b KO |
122 | max_discard_sectors -= max_discard_sectors % granularity; |
123 | ||
124 | if (unlikely(!max_discard_sectors)) { | |
125 | /* XXX: warn */ | |
126 | return NULL; | |
127 | } | |
128 | ||
129 | if (bio_sectors(bio) <= max_discard_sectors) | |
130 | return NULL; | |
131 | ||
132 | split_sectors = max_discard_sectors; | |
133 | ||
134 | /* | |
135 | * If the next starting sector would be misaligned, stop the discard at | |
136 | * the previous aligned sector. | |
137 | */ | |
c55ddd90 CH |
138 | tmp = bio->bi_iter.bi_sector + split_sectors - |
139 | ((lim->discard_alignment >> 9) % granularity); | |
54efd50b KO |
140 | tmp = sector_div(tmp, granularity); |
141 | ||
142 | if (split_sectors > tmp) | |
143 | split_sectors -= tmp; | |
144 | ||
145 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
146 | } | |
147 | ||
5a97806f | 148 | static struct bio *bio_split_write_zeroes(struct bio *bio, |
c55ddd90 | 149 | struct queue_limits *lim, unsigned *nsegs, struct bio_set *bs) |
885fa13f | 150 | { |
d665e12a | 151 | *nsegs = 0; |
c55ddd90 | 152 | if (!lim->max_write_zeroes_sectors) |
885fa13f | 153 | return NULL; |
c55ddd90 | 154 | if (bio_sectors(bio) <= lim->max_write_zeroes_sectors) |
885fa13f | 155 | return NULL; |
c55ddd90 | 156 | return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs); |
885fa13f CH |
157 | } |
158 | ||
9cc5169c BVA |
159 | /* |
160 | * Return the maximum number of sectors from the start of a bio that may be | |
161 | * submitted as a single request to a block device. If enough sectors remain, | |
162 | * align the end to the physical block size. Otherwise align the end to the | |
163 | * logical block size. This approach minimizes the number of non-aligned | |
164 | * requests that are submitted to a block device if the start of a bio is not | |
165 | * aligned to a physical block boundary. | |
166 | */ | |
5a97806f | 167 | static inline unsigned get_max_io_size(struct bio *bio, |
c55ddd90 | 168 | struct queue_limits *lim) |
d0e5fbb0 | 169 | { |
c55ddd90 CH |
170 | unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT; |
171 | unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT; | |
172 | unsigned max_sectors = lim->max_sectors, start, end; | |
d0e5fbb0 | 173 | |
c55ddd90 | 174 | if (lim->chunk_sectors) { |
efef739d CH |
175 | max_sectors = min(max_sectors, |
176 | blk_chunk_sectors_left(bio->bi_iter.bi_sector, | |
c55ddd90 | 177 | lim->chunk_sectors)); |
efef739d | 178 | } |
d0e5fbb0 | 179 | |
84613bed CH |
180 | start = bio->bi_iter.bi_sector & (pbs - 1); |
181 | end = (start + max_sectors) & ~(pbs - 1); | |
182 | if (end > start) | |
183 | return end - start; | |
184 | return max_sectors & ~(lbs - 1); | |
d0e5fbb0 ML |
185 | } |
186 | ||
c55ddd90 CH |
187 | static inline unsigned get_max_segment_size(struct queue_limits *lim, |
188 | struct page *start_page, unsigned long offset) | |
dcebd755 | 189 | { |
c55ddd90 | 190 | unsigned long mask = lim->seg_boundary_mask; |
dcebd755 | 191 | |
429120f3 | 192 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
193 | |
194 | /* | |
195 | * overflow may be triggered in case of zero page physical address | |
196 | * on 32bit arch, use queue's max segment size when that happens. | |
197 | */ | |
198 | return min_not_zero(mask - offset + 1, | |
c55ddd90 | 199 | (unsigned long)lim->max_segment_size); |
dcebd755 ML |
200 | } |
201 | ||
708b25b3 BVA |
202 | /** |
203 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
c55ddd90 | 204 | * @lim: [in] queue limits to split based on |
708b25b3 BVA |
205 | * @bv: [in] bvec to examine |
206 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
207 | * by the number of segments from @bv that may be appended to that | |
208 | * bio without exceeding @max_segs | |
67927d22 KB |
209 | * @bytes: [in,out] Number of bytes in the bio being built. Incremented |
210 | * by the number of bytes from @bv that may be appended to that | |
211 | * bio without exceeding @max_bytes | |
708b25b3 | 212 | * @max_segs: [in] upper bound for *@nsegs |
67927d22 | 213 | * @max_bytes: [in] upper bound for *@bytes |
708b25b3 BVA |
214 | * |
215 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
216 | * big to fit in a single segment and hence that it has to be split in the | |
217 | * middle. This function verifies whether or not that should happen. The value | |
218 | * %true is returned if and only if appending the entire @bv to a bio with | |
219 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
220 | * the block driver. | |
dcebd755 | 221 | */ |
c55ddd90 CH |
222 | static bool bvec_split_segs(struct queue_limits *lim, const struct bio_vec *bv, |
223 | unsigned *nsegs, unsigned *bytes, unsigned max_segs, | |
224 | unsigned max_bytes) | |
dcebd755 | 225 | { |
67927d22 | 226 | unsigned max_len = min(max_bytes, UINT_MAX) - *bytes; |
708b25b3 | 227 | unsigned len = min(bv->bv_len, max_len); |
dcebd755 | 228 | unsigned total_len = 0; |
ff9811b3 | 229 | unsigned seg_size = 0; |
dcebd755 | 230 | |
ff9811b3 | 231 | while (len && *nsegs < max_segs) { |
c55ddd90 | 232 | seg_size = get_max_segment_size(lim, bv->bv_page, |
429120f3 | 233 | bv->bv_offset + total_len); |
dcebd755 ML |
234 | seg_size = min(seg_size, len); |
235 | ||
ff9811b3 | 236 | (*nsegs)++; |
dcebd755 ML |
237 | total_len += seg_size; |
238 | len -= seg_size; | |
239 | ||
c55ddd90 | 240 | if ((bv->bv_offset + total_len) & lim->virt_boundary_mask) |
dcebd755 ML |
241 | break; |
242 | } | |
243 | ||
67927d22 | 244 | *bytes += total_len; |
dcebd755 | 245 | |
708b25b3 BVA |
246 | /* tell the caller to split the bvec if it is too big to fit */ |
247 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
248 | } |
249 | ||
dad77584 | 250 | /** |
5a97806f | 251 | * bio_split_rw - split a bio in two bios |
dad77584 | 252 | * @bio: [in] bio to be split |
c55ddd90 | 253 | * @lim: [in] queue limits to split based on |
dad77584 | 254 | * @segs: [out] number of segments in the bio with the first half of the sectors |
5a97806f | 255 | * @bs: [in] bio set to allocate the clone from |
a85b3637 | 256 | * @max_bytes: [in] maximum number of bytes per bio |
dad77584 BVA |
257 | * |
258 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
259 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
260 | * following is guaranteed for the cloned bio: | |
a85b3637 | 261 | * - That it has at most @max_bytes worth of data |
dad77584 BVA |
262 | * - That it has at most queue_max_segments(@q) segments. |
263 | * | |
264 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
265 | * the original bio. It is the responsibility of the caller to ensure that the | |
266 | * original bio is not freed before the cloned bio. The caller is also | |
267 | * responsible for ensuring that @bs is only destroyed after processing of the | |
268 | * split bio has finished. | |
269 | */ | |
c55ddd90 | 270 | static struct bio *bio_split_rw(struct bio *bio, struct queue_limits *lim, |
a85b3637 | 271 | unsigned *segs, struct bio_set *bs, unsigned max_bytes) |
54efd50b | 272 | { |
5014c311 | 273 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 274 | struct bvec_iter iter; |
67927d22 | 275 | unsigned nsegs = 0, bytes = 0; |
54efd50b | 276 | |
dcebd755 | 277 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
278 | /* |
279 | * If the queue doesn't support SG gaps and adding this | |
280 | * offset would create a gap, disallow it. | |
281 | */ | |
c55ddd90 | 282 | if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset)) |
54efd50b KO |
283 | goto split; |
284 | ||
c55ddd90 | 285 | if (nsegs < lim->max_segments && |
67927d22 | 286 | bytes + bv.bv_len <= max_bytes && |
708b25b3 BVA |
287 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
288 | nsegs++; | |
67927d22 | 289 | bytes += bv.bv_len; |
c55ddd90 CH |
290 | } else { |
291 | if (bvec_split_segs(lim, &bv, &nsegs, &bytes, | |
292 | lim->max_segments, max_bytes)) | |
293 | goto split; | |
e36f6204 KB |
294 | } |
295 | ||
54efd50b | 296 | bvprv = bv; |
578270bf | 297 | bvprvp = &bvprv; |
54efd50b KO |
298 | } |
299 | ||
d627065d CH |
300 | *segs = nsegs; |
301 | return NULL; | |
54efd50b | 302 | split: |
bdced438 | 303 | *segs = nsegs; |
cc29e1bf | 304 | |
67927d22 KB |
305 | /* |
306 | * Individual bvecs might not be logical block aligned. Round down the | |
307 | * split size so that each bio is properly block size aligned, even if | |
308 | * we do not use the full hardware limits. | |
309 | */ | |
c55ddd90 | 310 | bytes = ALIGN_DOWN(bytes, lim->logical_block_size); |
67927d22 | 311 | |
cc29e1bf JX |
312 | /* |
313 | * Bio splitting may cause subtle trouble such as hang when doing sync | |
314 | * iopoll in direct IO routine. Given performance gain of iopoll for | |
315 | * big IO can be trival, disable iopoll when split needed. | |
316 | */ | |
6ce913fe | 317 | bio_clear_polled(bio); |
67927d22 | 318 | return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs); |
54efd50b KO |
319 | } |
320 | ||
dad77584 | 321 | /** |
5a97806f CH |
322 | * __bio_split_to_limits - split a bio to fit the queue limits |
323 | * @bio: bio to be split | |
c55ddd90 | 324 | * @lim: queue limits to split based on |
5a97806f CH |
325 | * @nr_segs: returns the number of segments in the returned bio |
326 | * | |
327 | * Check if @bio needs splitting based on the queue limits, and if so split off | |
328 | * a bio fitting the limits from the beginning of @bio and return it. @bio is | |
329 | * shortened to the remainder and re-submitted. | |
dad77584 | 330 | * |
5a97806f CH |
331 | * The split bio is allocated from @q->bio_split, which is provided by the |
332 | * block layer. | |
dad77584 | 333 | */ |
c55ddd90 | 334 | struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim, |
abd45c15 | 335 | unsigned int *nr_segs) |
54efd50b | 336 | { |
46754bd0 | 337 | struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split; |
5a97806f | 338 | struct bio *split; |
54efd50b | 339 | |
5a97806f | 340 | switch (bio_op(bio)) { |
7afafc8a AH |
341 | case REQ_OP_DISCARD: |
342 | case REQ_OP_SECURE_ERASE: | |
c55ddd90 | 343 | split = bio_split_discard(bio, lim, nr_segs, bs); |
7afafc8a | 344 | break; |
a6f0788e | 345 | case REQ_OP_WRITE_ZEROES: |
c55ddd90 | 346 | split = bio_split_write_zeroes(bio, lim, nr_segs, bs); |
a6f0788e | 347 | break; |
7afafc8a | 348 | default: |
c55ddd90 CH |
349 | split = bio_split_rw(bio, lim, nr_segs, bs, |
350 | get_max_io_size(bio, lim) << SECTOR_SHIFT); | |
7afafc8a AH |
351 | break; |
352 | } | |
bdced438 | 353 | |
54efd50b | 354 | if (split) { |
6ac45aeb | 355 | /* there isn't chance to merge the splitted bio */ |
1eff9d32 | 356 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 357 | |
957a2b34 | 358 | blkcg_bio_issue_init(split); |
5a97806f CH |
359 | bio_chain(split, bio); |
360 | trace_block_split(split, bio->bi_iter.bi_sector); | |
361 | submit_bio_noacct(bio); | |
362 | return split; | |
54efd50b | 363 | } |
5a97806f | 364 | return bio; |
54efd50b | 365 | } |
14ccb66b | 366 | |
dad77584 | 367 | /** |
5a97806f CH |
368 | * bio_split_to_limits - split a bio to fit the queue limits |
369 | * @bio: bio to be split | |
370 | * | |
371 | * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and | |
372 | * if so split off a bio fitting the limits from the beginning of @bio and | |
373 | * return it. @bio is shortened to the remainder and re-submitted. | |
dad77584 | 374 | * |
5a97806f CH |
375 | * The split bio is allocated from @q->bio_split, which is provided by the |
376 | * block layer. | |
dad77584 | 377 | */ |
5a97806f | 378 | struct bio *bio_split_to_limits(struct bio *bio) |
14ccb66b | 379 | { |
c55ddd90 | 380 | struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits; |
14ccb66b CH |
381 | unsigned int nr_segs; |
382 | ||
c55ddd90 CH |
383 | if (bio_may_exceed_limits(bio, lim)) |
384 | return __bio_split_to_limits(bio, lim, &nr_segs); | |
5a97806f | 385 | return bio; |
14ccb66b | 386 | } |
5a97806f | 387 | EXPORT_SYMBOL(bio_split_to_limits); |
54efd50b | 388 | |
e9cd19c0 | 389 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 390 | { |
6869875f | 391 | unsigned int nr_phys_segs = 0; |
67927d22 | 392 | unsigned int bytes = 0; |
e9cd19c0 | 393 | struct req_iterator iter; |
6869875f | 394 | struct bio_vec bv; |
d6d48196 | 395 | |
e9cd19c0 | 396 | if (!rq->bio) |
1e428079 | 397 | return 0; |
d6d48196 | 398 | |
e9cd19c0 | 399 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
400 | case REQ_OP_DISCARD: |
401 | case REQ_OP_SECURE_ERASE: | |
a958937f DJ |
402 | if (queue_max_discard_segments(rq->q) > 1) { |
403 | struct bio *bio = rq->bio; | |
404 | ||
405 | for_each_bio(bio) | |
406 | nr_phys_segs++; | |
407 | return nr_phys_segs; | |
408 | } | |
409 | return 1; | |
a6f0788e | 410 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 | 411 | return 0; |
2d9b02be BVA |
412 | default: |
413 | break; | |
a6f0788e | 414 | } |
5cb8850c | 415 | |
e9cd19c0 | 416 | rq_for_each_bvec(bv, rq, iter) |
c55ddd90 | 417 | bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes, |
708b25b3 | 418 | UINT_MAX, UINT_MAX); |
1e428079 JA |
419 | return nr_phys_segs; |
420 | } | |
421 | ||
48d7727c | 422 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
423 | struct scatterlist *sglist) |
424 | { | |
425 | if (!*sg) | |
426 | return sglist; | |
427 | ||
428 | /* | |
429 | * If the driver previously mapped a shorter list, we could see a | |
430 | * termination bit prematurely unless it fully inits the sg table | |
431 | * on each mapping. We KNOW that there must be more entries here | |
432 | * or the driver would be buggy, so force clear the termination bit | |
433 | * to avoid doing a full sg_init_table() in drivers for each command. | |
434 | */ | |
435 | sg_unmark_end(*sg); | |
436 | return sg_next(*sg); | |
437 | } | |
438 | ||
439 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
440 | struct bio_vec *bvec, struct scatterlist *sglist, | |
441 | struct scatterlist **sg) | |
442 | { | |
443 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 444 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
445 | |
446 | while (nbytes > 0) { | |
8a96a0e4 | 447 | unsigned offset = bvec->bv_offset + total; |
c55ddd90 CH |
448 | unsigned len = min(get_max_segment_size(&q->limits, |
449 | bvec->bv_page, offset), nbytes); | |
f9f76879 CH |
450 | struct page *page = bvec->bv_page; |
451 | ||
452 | /* | |
453 | * Unfortunately a fair number of drivers barf on scatterlists | |
454 | * that have an offset larger than PAGE_SIZE, despite other | |
455 | * subsystems dealing with that invariant just fine. For now | |
456 | * stick to the legacy format where we never present those from | |
457 | * the block layer, but the code below should be removed once | |
458 | * these offenders (mostly MMC/SD drivers) are fixed. | |
459 | */ | |
460 | page += (offset >> PAGE_SHIFT); | |
461 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
462 | |
463 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 464 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 465 | |
8a96a0e4 CH |
466 | total += len; |
467 | nbytes -= len; | |
862e5a5e ML |
468 | nsegs++; |
469 | } | |
470 | ||
471 | return nsegs; | |
472 | } | |
473 | ||
16e3e418 ML |
474 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
475 | struct scatterlist *sglist, struct scatterlist **sg) | |
476 | { | |
477 | *sg = blk_next_sg(sg, sglist); | |
478 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
479 | return 1; | |
480 | } | |
481 | ||
f6970f83 ML |
482 | /* only try to merge bvecs into one sg if they are from two bios */ |
483 | static inline bool | |
484 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
485 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
486 | { |
487 | ||
488 | int nbytes = bvec->bv_len; | |
489 | ||
f6970f83 ML |
490 | if (!*sg) |
491 | return false; | |
963ab9e5 | 492 | |
f6970f83 ML |
493 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
494 | return false; | |
495 | ||
496 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
497 | return false; | |
498 | ||
499 | (*sg)->length += nbytes; | |
500 | ||
501 | return true; | |
963ab9e5 AH |
502 | } |
503 | ||
5cb8850c KO |
504 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
505 | struct scatterlist *sglist, | |
506 | struct scatterlist **sg) | |
d6d48196 | 507 | { |
3f649ab7 | 508 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 509 | struct bvec_iter iter; |
38417468 | 510 | int nsegs = 0; |
f6970f83 | 511 | bool new_bio = false; |
5cb8850c | 512 | |
f6970f83 ML |
513 | for_each_bio(bio) { |
514 | bio_for_each_bvec(bvec, bio, iter) { | |
515 | /* | |
516 | * Only try to merge bvecs from two bios given we | |
517 | * have done bio internal merge when adding pages | |
518 | * to bio | |
519 | */ | |
520 | if (new_bio && | |
521 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
522 | goto next_bvec; | |
523 | ||
524 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
525 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
526 | else | |
527 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
528 | next_bvec: | |
529 | new_bio = false; | |
530 | } | |
b21e11c5 ML |
531 | if (likely(bio->bi_iter.bi_size)) { |
532 | bvprv = bvec; | |
533 | new_bio = true; | |
534 | } | |
f6970f83 | 535 | } |
d6d48196 | 536 | |
5cb8850c KO |
537 | return nsegs; |
538 | } | |
539 | ||
540 | /* | |
541 | * map a request to scatterlist, return number of sg entries setup. Caller | |
542 | * must make sure sg can hold rq->nr_phys_segments entries | |
543 | */ | |
89de1504 CH |
544 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
545 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 546 | { |
5cb8850c KO |
547 | int nsegs = 0; |
548 | ||
f9d03f96 | 549 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 550 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 551 | else if (rq->bio) |
89de1504 | 552 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 553 | |
89de1504 CH |
554 | if (*last_sg) |
555 | sg_mark_end(*last_sg); | |
d6d48196 | 556 | |
12e57f59 ML |
557 | /* |
558 | * Something must have been wrong if the figured number of | |
559 | * segment is bigger than number of req's physical segments | |
560 | */ | |
f9d03f96 | 561 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 562 | |
d6d48196 JA |
563 | return nsegs; |
564 | } | |
89de1504 | 565 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 566 | |
943b40c8 ML |
567 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
568 | { | |
569 | if (req_op(rq) == REQ_OP_DISCARD) | |
570 | return queue_max_discard_segments(rq->q); | |
571 | return queue_max_segments(rq->q); | |
572 | } | |
573 | ||
badf7f64 CH |
574 | static inline unsigned int blk_rq_get_max_sectors(struct request *rq, |
575 | sector_t offset) | |
576 | { | |
577 | struct request_queue *q = rq->q; | |
c8875190 | 578 | unsigned int max_sectors; |
badf7f64 CH |
579 | |
580 | if (blk_rq_is_passthrough(rq)) | |
581 | return q->limits.max_hw_sectors; | |
582 | ||
c8875190 | 583 | max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
badf7f64 CH |
584 | if (!q->limits.chunk_sectors || |
585 | req_op(rq) == REQ_OP_DISCARD || | |
586 | req_op(rq) == REQ_OP_SECURE_ERASE) | |
c8875190 CH |
587 | return max_sectors; |
588 | return min(max_sectors, | |
589 | blk_chunk_sectors_left(offset, q->limits.chunk_sectors)); | |
badf7f64 CH |
590 | } |
591 | ||
14ccb66b CH |
592 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
593 | unsigned int nr_phys_segs) | |
d6d48196 | 594 | { |
6b2b0459 TH |
595 | if (!blk_cgroup_mergeable(req, bio)) |
596 | goto no_merge; | |
597 | ||
2705dfb2 | 598 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d MP |
599 | goto no_merge; |
600 | ||
2705dfb2 ML |
601 | /* discard request merge won't add new segment */ |
602 | if (req_op(req) == REQ_OP_DISCARD) | |
603 | return 1; | |
604 | ||
605 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) | |
13f05c8d | 606 | goto no_merge; |
d6d48196 JA |
607 | |
608 | /* | |
609 | * This will form the start of a new hw segment. Bump both | |
610 | * counters. | |
611 | */ | |
d6d48196 JA |
612 | req->nr_phys_segments += nr_phys_segs; |
613 | return 1; | |
13f05c8d MP |
614 | |
615 | no_merge: | |
14ccb66b | 616 | req_set_nomerge(req->q, req); |
13f05c8d | 617 | return 0; |
d6d48196 JA |
618 | } |
619 | ||
14ccb66b | 620 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 621 | { |
5e7c4274 JA |
622 | if (req_gap_back_merge(req, bio)) |
623 | return 0; | |
7f39add3 SG |
624 | if (blk_integrity_rq(req) && |
625 | integrity_req_gap_back_merge(req, bio)) | |
626 | return 0; | |
a892c8d5 ST |
627 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
628 | return 0; | |
f31dc1cd | 629 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 630 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 631 | req_set_nomerge(req->q, req); |
d6d48196 JA |
632 | return 0; |
633 | } | |
d6d48196 | 634 | |
14ccb66b | 635 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
636 | } |
637 | ||
eda5cc99 CH |
638 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
639 | unsigned int nr_segs) | |
d6d48196 | 640 | { |
5e7c4274 JA |
641 | if (req_gap_front_merge(req, bio)) |
642 | return 0; | |
7f39add3 SG |
643 | if (blk_integrity_rq(req) && |
644 | integrity_req_gap_front_merge(req, bio)) | |
645 | return 0; | |
a892c8d5 ST |
646 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
647 | return 0; | |
f31dc1cd | 648 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 649 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 650 | req_set_nomerge(req->q, req); |
d6d48196 JA |
651 | return 0; |
652 | } | |
d6d48196 | 653 | |
14ccb66b | 654 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
655 | } |
656 | ||
445251d0 JA |
657 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
658 | struct request *next) | |
659 | { | |
660 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
661 | ||
662 | if (segments >= queue_max_discard_segments(q)) | |
663 | goto no_merge; | |
664 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
665 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
666 | goto no_merge; | |
667 | ||
668 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
669 | return true; | |
670 | no_merge: | |
671 | req_set_nomerge(q, req); | |
672 | return false; | |
673 | } | |
674 | ||
d6d48196 JA |
675 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
676 | struct request *next) | |
677 | { | |
678 | int total_phys_segments; | |
d6d48196 | 679 | |
5e7c4274 | 680 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
681 | return 0; |
682 | ||
d6d48196 JA |
683 | /* |
684 | * Will it become too large? | |
685 | */ | |
f31dc1cd | 686 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 687 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
688 | return 0; |
689 | ||
690 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 691 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
692 | return 0; |
693 | ||
6b2b0459 TH |
694 | if (!blk_cgroup_mergeable(req, next->bio)) |
695 | return 0; | |
696 | ||
4eaf99be | 697 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
698 | return 0; |
699 | ||
a892c8d5 ST |
700 | if (!bio_crypt_ctx_merge_rq(req, next)) |
701 | return 0; | |
702 | ||
d6d48196 JA |
703 | /* Merge is OK... */ |
704 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
705 | return 1; |
706 | } | |
707 | ||
80a761fd TH |
708 | /** |
709 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
710 | * @rq: request to mark as mixed merge | |
711 | * | |
712 | * Description: | |
713 | * @rq is about to be mixed merged. Make sure the attributes | |
714 | * which can be mixed are set in each bio and mark @rq as mixed | |
715 | * merged. | |
716 | */ | |
717 | void blk_rq_set_mixed_merge(struct request *rq) | |
718 | { | |
16458cf3 | 719 | blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
80a761fd TH |
720 | struct bio *bio; |
721 | ||
e8064021 | 722 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
723 | return; |
724 | ||
725 | /* | |
726 | * @rq will no longer represent mixable attributes for all the | |
727 | * contained bios. It will just track those of the first one. | |
728 | * Distributes the attributs to each bio. | |
729 | */ | |
730 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
731 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
732 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
733 | bio->bi_opf |= ff; | |
80a761fd | 734 | } |
e8064021 | 735 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
736 | } |
737 | ||
b9c54f56 | 738 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
739 | { |
740 | if (blk_do_io_stat(req)) { | |
112f158f | 741 | part_stat_lock(); |
b9c54f56 | 742 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab JM |
743 | part_stat_unlock(); |
744 | } | |
745 | } | |
b9c54f56 | 746 | |
e96c0d83 EB |
747 | static enum elv_merge blk_try_req_merge(struct request *req, |
748 | struct request *next) | |
69840466 JW |
749 | { |
750 | if (blk_discard_mergable(req)) | |
751 | return ELEVATOR_DISCARD_MERGE; | |
752 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
753 | return ELEVATOR_BACK_MERGE; | |
754 | ||
755 | return ELEVATOR_NO_MERGE; | |
756 | } | |
26308eab | 757 | |
d6d48196 | 758 | /* |
b973cb7e JA |
759 | * For non-mq, this has to be called with the request spinlock acquired. |
760 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 761 | */ |
b973cb7e JA |
762 | static struct request *attempt_merge(struct request_queue *q, |
763 | struct request *req, struct request *next) | |
d6d48196 JA |
764 | { |
765 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 766 | return NULL; |
d6d48196 | 767 | |
288dab8a | 768 | if (req_op(req) != req_op(next)) |
b973cb7e | 769 | return NULL; |
f31dc1cd | 770 | |
79bb1dbd | 771 | if (rq_data_dir(req) != rq_data_dir(next)) |
b973cb7e | 772 | return NULL; |
d6d48196 | 773 | |
668ffc03 DLM |
774 | if (req->ioprio != next->ioprio) |
775 | return NULL; | |
776 | ||
d6d48196 JA |
777 | /* |
778 | * If we are allowed to merge, then append bio list | |
779 | * from next to rq and release next. merge_requests_fn | |
780 | * will have updated segment counts, update sector | |
445251d0 JA |
781 | * counts here. Handle DISCARDs separately, as they |
782 | * have separate settings. | |
d6d48196 | 783 | */ |
69840466 JW |
784 | |
785 | switch (blk_try_req_merge(req, next)) { | |
786 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
787 | if (!req_attempt_discard_merge(q, req, next)) |
788 | return NULL; | |
69840466 JW |
789 | break; |
790 | case ELEVATOR_BACK_MERGE: | |
791 | if (!ll_merge_requests_fn(q, req, next)) | |
792 | return NULL; | |
793 | break; | |
794 | default: | |
b973cb7e | 795 | return NULL; |
69840466 | 796 | } |
d6d48196 | 797 | |
80a761fd TH |
798 | /* |
799 | * If failfast settings disagree or any of the two is already | |
800 | * a mixed merge, mark both as mixed before proceeding. This | |
801 | * makes sure that all involved bios have mixable attributes | |
802 | * set properly. | |
803 | */ | |
e8064021 | 804 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
805 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
806 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
807 | blk_rq_set_mixed_merge(req); | |
808 | blk_rq_set_mixed_merge(next); | |
809 | } | |
810 | ||
d6d48196 | 811 | /* |
522a7775 OS |
812 | * At this point we have either done a back merge or front merge. We |
813 | * need the smaller start_time_ns of the merged requests to be the | |
814 | * current request for accounting purposes. | |
d6d48196 | 815 | */ |
522a7775 OS |
816 | if (next->start_time_ns < req->start_time_ns) |
817 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
818 | |
819 | req->biotail->bi_next = next->bio; | |
820 | req->biotail = next->biotail; | |
821 | ||
a2dec7b3 | 822 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 823 | |
2a5cf35c | 824 | if (!blk_discard_mergable(req)) |
445251d0 | 825 | elv_merge_requests(q, req, next); |
d6d48196 | 826 | |
42dad764 JM |
827 | /* |
828 | * 'next' is going away, so update stats accordingly | |
829 | */ | |
b9c54f56 | 830 | blk_account_io_merge_request(next); |
d6d48196 | 831 | |
a54895fa | 832 | trace_block_rq_merge(next); |
f3bdc62f | 833 | |
e4d750c9 JA |
834 | /* |
835 | * ownership of bio passed from next to req, return 'next' for | |
836 | * the caller to free | |
837 | */ | |
1cd96c24 | 838 | next->bio = NULL; |
b973cb7e | 839 | return next; |
d6d48196 JA |
840 | } |
841 | ||
eda5cc99 CH |
842 | static struct request *attempt_back_merge(struct request_queue *q, |
843 | struct request *rq) | |
d6d48196 JA |
844 | { |
845 | struct request *next = elv_latter_request(q, rq); | |
846 | ||
847 | if (next) | |
848 | return attempt_merge(q, rq, next); | |
849 | ||
b973cb7e | 850 | return NULL; |
d6d48196 JA |
851 | } |
852 | ||
eda5cc99 CH |
853 | static struct request *attempt_front_merge(struct request_queue *q, |
854 | struct request *rq) | |
d6d48196 JA |
855 | { |
856 | struct request *prev = elv_former_request(q, rq); | |
857 | ||
858 | if (prev) | |
859 | return attempt_merge(q, prev, rq); | |
860 | ||
b973cb7e | 861 | return NULL; |
d6d48196 | 862 | } |
5e84ea3a | 863 | |
fd2ef39c JK |
864 | /* |
865 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false | |
866 | * otherwise. The caller is responsible for freeing 'next' if the merge | |
867 | * happened. | |
868 | */ | |
869 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
870 | struct request *next) | |
5e84ea3a | 871 | { |
fd2ef39c | 872 | return attempt_merge(q, rq, next); |
5e84ea3a | 873 | } |
050c8ea8 TH |
874 | |
875 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
876 | { | |
e2a60da7 | 877 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
878 | return false; |
879 | ||
288dab8a | 880 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
881 | return false; |
882 | ||
050c8ea8 TH |
883 | /* different data direction or already started, don't merge */ |
884 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
885 | return false; | |
886 | ||
6b2b0459 TH |
887 | /* don't merge across cgroup boundaries */ |
888 | if (!blk_cgroup_mergeable(rq, bio)) | |
889 | return false; | |
890 | ||
050c8ea8 | 891 | /* only merge integrity protected bio into ditto rq */ |
4eaf99be | 892 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
893 | return false; |
894 | ||
a892c8d5 ST |
895 | /* Only merge if the crypt contexts are compatible */ |
896 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
897 | return false; | |
898 | ||
668ffc03 DLM |
899 | if (rq->ioprio != bio_prio(bio)) |
900 | return false; | |
901 | ||
050c8ea8 TH |
902 | return true; |
903 | } | |
904 | ||
34fe7c05 | 905 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 906 | { |
69840466 | 907 | if (blk_discard_mergable(rq)) |
1e739730 CH |
908 | return ELEVATOR_DISCARD_MERGE; |
909 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 910 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 911 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
912 | return ELEVATOR_FRONT_MERGE; |
913 | return ELEVATOR_NO_MERGE; | |
914 | } | |
8e756373 BW |
915 | |
916 | static void blk_account_io_merge_bio(struct request *req) | |
917 | { | |
918 | if (!blk_do_io_stat(req)) | |
919 | return; | |
920 | ||
921 | part_stat_lock(); | |
922 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
923 | part_stat_unlock(); | |
924 | } | |
925 | ||
eda5cc99 CH |
926 | enum bio_merge_status { |
927 | BIO_MERGE_OK, | |
928 | BIO_MERGE_NONE, | |
929 | BIO_MERGE_FAILED, | |
930 | }; | |
931 | ||
932 | static enum bio_merge_status bio_attempt_back_merge(struct request *req, | |
933 | struct bio *bio, unsigned int nr_segs) | |
8e756373 | 934 | { |
16458cf3 | 935 | const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; |
8e756373 BW |
936 | |
937 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 938 | return BIO_MERGE_FAILED; |
8e756373 | 939 | |
e8a676d6 | 940 | trace_block_bio_backmerge(bio); |
8e756373 BW |
941 | rq_qos_merge(req->q, req, bio); |
942 | ||
943 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
944 | blk_rq_set_mixed_merge(req); | |
945 | ||
946 | req->biotail->bi_next = bio; | |
947 | req->biotail = bio; | |
948 | req->__data_len += bio->bi_iter.bi_size; | |
949 | ||
950 | bio_crypt_free_ctx(bio); | |
951 | ||
952 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 953 | return BIO_MERGE_OK; |
8e756373 BW |
954 | } |
955 | ||
eda5cc99 CH |
956 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
957 | struct bio *bio, unsigned int nr_segs) | |
8e756373 | 958 | { |
16458cf3 | 959 | const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; |
8e756373 BW |
960 | |
961 | if (!ll_front_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 962 | return BIO_MERGE_FAILED; |
8e756373 | 963 | |
e8a676d6 | 964 | trace_block_bio_frontmerge(bio); |
8e756373 BW |
965 | rq_qos_merge(req->q, req, bio); |
966 | ||
967 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
968 | blk_rq_set_mixed_merge(req); | |
969 | ||
970 | bio->bi_next = req->bio; | |
971 | req->bio = bio; | |
972 | ||
973 | req->__sector = bio->bi_iter.bi_sector; | |
974 | req->__data_len += bio->bi_iter.bi_size; | |
975 | ||
976 | bio_crypt_do_front_merge(req, bio); | |
977 | ||
978 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 979 | return BIO_MERGE_OK; |
8e756373 BW |
980 | } |
981 | ||
eda5cc99 CH |
982 | static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
983 | struct request *req, struct bio *bio) | |
8e756373 BW |
984 | { |
985 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
986 | ||
987 | if (segments >= queue_max_discard_segments(q)) | |
988 | goto no_merge; | |
989 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
990 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
991 | goto no_merge; | |
992 | ||
993 | rq_qos_merge(q, req, bio); | |
994 | ||
995 | req->biotail->bi_next = bio; | |
996 | req->biotail = bio; | |
997 | req->__data_len += bio->bi_iter.bi_size; | |
998 | req->nr_phys_segments = segments + 1; | |
999 | ||
1000 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1001 | return BIO_MERGE_OK; |
8e756373 BW |
1002 | no_merge: |
1003 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
1004 | return BIO_MERGE_FAILED; |
1005 | } | |
1006 | ||
1007 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
1008 | struct request *rq, | |
1009 | struct bio *bio, | |
1010 | unsigned int nr_segs, | |
1011 | bool sched_allow_merge) | |
1012 | { | |
1013 | if (!blk_rq_merge_ok(rq, bio)) | |
1014 | return BIO_MERGE_NONE; | |
1015 | ||
1016 | switch (blk_try_merge(rq, bio)) { | |
1017 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 1018 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1019 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1020 | break; | |
1021 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1022 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1023 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1024 | break; | |
1025 | case ELEVATOR_DISCARD_MERGE: | |
1026 | return bio_attempt_discard_merge(q, rq, bio); | |
1027 | default: | |
1028 | return BIO_MERGE_NONE; | |
1029 | } | |
1030 | ||
1031 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1032 | } |
1033 | ||
1034 | /** | |
1035 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1036 | * @q: request_queue new bio is being queued at | |
1037 | * @bio: new bio being queued | |
1038 | * @nr_segs: number of segments in @bio | |
87c037d1 | 1039 | * from the passed in @q already in the plug list |
8e756373 | 1040 | * |
d38a9c04 JA |
1041 | * Determine whether @bio being queued on @q can be merged with the previous |
1042 | * request on %current's plugged list. Returns %true if merge was successful, | |
8e756373 BW |
1043 | * otherwise %false. |
1044 | * | |
1045 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1046 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1047 | * than scheduling, and the request, while may have elvpriv data, is not | |
1048 | * added on the elevator at this point. In addition, we don't have | |
1049 | * reliable access to the elevator outside queue lock. Only check basic | |
1050 | * merging parameters without querying the elevator. | |
1051 | * | |
1052 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1053 | */ | |
1054 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
0c5bcc92 | 1055 | unsigned int nr_segs) |
8e756373 BW |
1056 | { |
1057 | struct blk_plug *plug; | |
1058 | struct request *rq; | |
8e756373 | 1059 | |
6deacb3b | 1060 | plug = blk_mq_plug(bio); |
bc490f81 | 1061 | if (!plug || rq_list_empty(plug->mq_list)) |
8e756373 BW |
1062 | return false; |
1063 | ||
5b205071 JA |
1064 | rq_list_for_each(&plug->mq_list, rq) { |
1065 | if (rq->q == q) { | |
1066 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == | |
1067 | BIO_MERGE_OK) | |
1068 | return true; | |
1069 | break; | |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * Only keep iterating plug list for merges if we have multiple | |
1074 | * queues | |
1075 | */ | |
1076 | if (!plug->multiple_queues) | |
1077 | break; | |
8e756373 | 1078 | } |
8e756373 BW |
1079 | return false; |
1080 | } | |
bdc6a287 BW |
1081 | |
1082 | /* | |
1083 | * Iterate list of requests and see if we can merge this bio with any | |
1084 | * of them. | |
1085 | */ | |
1086 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1087 | struct bio *bio, unsigned int nr_segs) | |
1088 | { | |
1089 | struct request *rq; | |
1090 | int checked = 8; | |
1091 | ||
1092 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1093 | if (!checked--) |
1094 | break; | |
1095 | ||
7d7ca7c5 BW |
1096 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1097 | case BIO_MERGE_NONE: | |
bdc6a287 | 1098 | continue; |
7d7ca7c5 BW |
1099 | case BIO_MERGE_OK: |
1100 | return true; | |
1101 | case BIO_MERGE_FAILED: | |
1102 | return false; | |
bdc6a287 BW |
1103 | } |
1104 | ||
bdc6a287 BW |
1105 | } |
1106 | ||
1107 | return false; | |
1108 | } | |
1109 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); | |
eda5cc99 CH |
1110 | |
1111 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, | |
1112 | unsigned int nr_segs, struct request **merged_request) | |
1113 | { | |
1114 | struct request *rq; | |
1115 | ||
1116 | switch (elv_merge(q, &rq, bio)) { | |
1117 | case ELEVATOR_BACK_MERGE: | |
1118 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1119 | return false; | |
1120 | if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1121 | return false; | |
1122 | *merged_request = attempt_back_merge(q, rq); | |
1123 | if (!*merged_request) | |
1124 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
1125 | return true; | |
1126 | case ELEVATOR_FRONT_MERGE: | |
1127 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1128 | return false; | |
1129 | if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1130 | return false; | |
1131 | *merged_request = attempt_front_merge(q, rq); | |
1132 | if (!*merged_request) | |
1133 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
1134 | return true; | |
1135 | case ELEVATOR_DISCARD_MERGE: | |
1136 | return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; | |
1137 | default: | |
1138 | return false; | |
1139 | } | |
1140 | } | |
1141 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |