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 | */ | |
aa261f20 | 103 | static unsigned int bio_allowed_max_sectors(const struct queue_limits *lim) |
b6dc6198 | 104 | { |
c55ddd90 | 105 | return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT; |
b6dc6198 CH |
106 | } |
107 | ||
aa261f20 BVA |
108 | static struct bio *bio_split_discard(struct bio *bio, |
109 | const struct queue_limits *lim, | |
110 | unsigned *nsegs, struct bio_set *bs) | |
54efd50b KO |
111 | { |
112 | unsigned int max_discard_sectors, granularity; | |
54efd50b KO |
113 | sector_t tmp; |
114 | unsigned split_sectors; | |
115 | ||
bdced438 ML |
116 | *nsegs = 1; |
117 | ||
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 | 122 | max_discard_sectors -= max_discard_sectors % granularity; |
928a5dd3 | 123 | if (unlikely(!max_discard_sectors)) |
54efd50b | 124 | return NULL; |
54efd50b KO |
125 | |
126 | if (bio_sectors(bio) <= max_discard_sectors) | |
127 | return NULL; | |
128 | ||
129 | split_sectors = max_discard_sectors; | |
130 | ||
131 | /* | |
132 | * If the next starting sector would be misaligned, stop the discard at | |
133 | * the previous aligned sector. | |
134 | */ | |
c55ddd90 CH |
135 | tmp = bio->bi_iter.bi_sector + split_sectors - |
136 | ((lim->discard_alignment >> 9) % granularity); | |
54efd50b KO |
137 | tmp = sector_div(tmp, granularity); |
138 | ||
139 | if (split_sectors > tmp) | |
140 | split_sectors -= tmp; | |
141 | ||
142 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
143 | } | |
144 | ||
5a97806f | 145 | static struct bio *bio_split_write_zeroes(struct bio *bio, |
aa261f20 BVA |
146 | const struct queue_limits *lim, |
147 | unsigned *nsegs, struct bio_set *bs) | |
885fa13f | 148 | { |
d665e12a | 149 | *nsegs = 0; |
c55ddd90 | 150 | if (!lim->max_write_zeroes_sectors) |
885fa13f | 151 | return NULL; |
c55ddd90 | 152 | if (bio_sectors(bio) <= lim->max_write_zeroes_sectors) |
885fa13f | 153 | return NULL; |
c55ddd90 | 154 | return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs); |
885fa13f CH |
155 | } |
156 | ||
9cc5169c BVA |
157 | /* |
158 | * Return the maximum number of sectors from the start of a bio that may be | |
159 | * submitted as a single request to a block device. If enough sectors remain, | |
160 | * align the end to the physical block size. Otherwise align the end to the | |
161 | * logical block size. This approach minimizes the number of non-aligned | |
162 | * requests that are submitted to a block device if the start of a bio is not | |
163 | * aligned to a physical block boundary. | |
164 | */ | |
5a97806f | 165 | static inline unsigned get_max_io_size(struct bio *bio, |
aa261f20 | 166 | const struct queue_limits *lim) |
d0e5fbb0 | 167 | { |
c55ddd90 CH |
168 | unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT; |
169 | unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT; | |
170 | unsigned max_sectors = lim->max_sectors, start, end; | |
d0e5fbb0 | 171 | |
c55ddd90 | 172 | if (lim->chunk_sectors) { |
efef739d CH |
173 | max_sectors = min(max_sectors, |
174 | blk_chunk_sectors_left(bio->bi_iter.bi_sector, | |
c55ddd90 | 175 | lim->chunk_sectors)); |
efef739d | 176 | } |
d0e5fbb0 | 177 | |
84613bed CH |
178 | start = bio->bi_iter.bi_sector & (pbs - 1); |
179 | end = (start + max_sectors) & ~(pbs - 1); | |
180 | if (end > start) | |
181 | return end - start; | |
182 | return max_sectors & ~(lbs - 1); | |
d0e5fbb0 ML |
183 | } |
184 | ||
95465318 BVA |
185 | /** |
186 | * get_max_segment_size() - maximum number of bytes to add as a single segment | |
187 | * @lim: Request queue limits. | |
188 | * @start_page: See below. | |
189 | * @offset: Offset from @start_page where to add a segment. | |
190 | * | |
191 | * Returns the maximum number of bytes that can be added as a single segment. | |
192 | */ | |
aa261f20 | 193 | static inline unsigned get_max_segment_size(const struct queue_limits *lim, |
c55ddd90 | 194 | struct page *start_page, unsigned long offset) |
dcebd755 | 195 | { |
c55ddd90 | 196 | unsigned long mask = lim->seg_boundary_mask; |
dcebd755 | 197 | |
429120f3 | 198 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
199 | |
200 | /* | |
95465318 BVA |
201 | * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1 |
202 | * after having calculated the minimum. | |
4a2f704e | 203 | */ |
95465318 | 204 | return min(mask - offset, (unsigned long)lim->max_segment_size - 1) + 1; |
dcebd755 ML |
205 | } |
206 | ||
708b25b3 BVA |
207 | /** |
208 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
c55ddd90 | 209 | * @lim: [in] queue limits to split based on |
708b25b3 BVA |
210 | * @bv: [in] bvec to examine |
211 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
212 | * by the number of segments from @bv that may be appended to that | |
213 | * bio without exceeding @max_segs | |
67927d22 KB |
214 | * @bytes: [in,out] Number of bytes in the bio being built. Incremented |
215 | * by the number of bytes from @bv that may be appended to that | |
216 | * bio without exceeding @max_bytes | |
708b25b3 | 217 | * @max_segs: [in] upper bound for *@nsegs |
67927d22 | 218 | * @max_bytes: [in] upper bound for *@bytes |
708b25b3 BVA |
219 | * |
220 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
221 | * big to fit in a single segment and hence that it has to be split in the | |
222 | * middle. This function verifies whether or not that should happen. The value | |
223 | * %true is returned if and only if appending the entire @bv to a bio with | |
224 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
225 | * the block driver. | |
dcebd755 | 226 | */ |
aa261f20 BVA |
227 | static bool bvec_split_segs(const struct queue_limits *lim, |
228 | const struct bio_vec *bv, unsigned *nsegs, unsigned *bytes, | |
229 | unsigned max_segs, unsigned max_bytes) | |
dcebd755 | 230 | { |
67927d22 | 231 | unsigned max_len = min(max_bytes, UINT_MAX) - *bytes; |
708b25b3 | 232 | unsigned len = min(bv->bv_len, max_len); |
dcebd755 | 233 | unsigned total_len = 0; |
ff9811b3 | 234 | unsigned seg_size = 0; |
dcebd755 | 235 | |
ff9811b3 | 236 | while (len && *nsegs < max_segs) { |
c55ddd90 | 237 | seg_size = get_max_segment_size(lim, bv->bv_page, |
429120f3 | 238 | bv->bv_offset + total_len); |
dcebd755 ML |
239 | seg_size = min(seg_size, len); |
240 | ||
ff9811b3 | 241 | (*nsegs)++; |
dcebd755 ML |
242 | total_len += seg_size; |
243 | len -= seg_size; | |
244 | ||
c55ddd90 | 245 | if ((bv->bv_offset + total_len) & lim->virt_boundary_mask) |
dcebd755 ML |
246 | break; |
247 | } | |
248 | ||
67927d22 | 249 | *bytes += total_len; |
dcebd755 | 250 | |
708b25b3 BVA |
251 | /* tell the caller to split the bvec if it is too big to fit */ |
252 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
253 | } |
254 | ||
dad77584 | 255 | /** |
5a97806f | 256 | * bio_split_rw - split a bio in two bios |
dad77584 | 257 | * @bio: [in] bio to be split |
c55ddd90 | 258 | * @lim: [in] queue limits to split based on |
dad77584 | 259 | * @segs: [out] number of segments in the bio with the first half of the sectors |
5a97806f | 260 | * @bs: [in] bio set to allocate the clone from |
a85b3637 | 261 | * @max_bytes: [in] maximum number of bytes per bio |
dad77584 BVA |
262 | * |
263 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
264 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
265 | * following is guaranteed for the cloned bio: | |
a85b3637 | 266 | * - That it has at most @max_bytes worth of data |
dad77584 BVA |
267 | * - That it has at most queue_max_segments(@q) segments. |
268 | * | |
269 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
270 | * the original bio. It is the responsibility of the caller to ensure that the | |
271 | * original bio is not freed before the cloned bio. The caller is also | |
272 | * responsible for ensuring that @bs is only destroyed after processing of the | |
273 | * split bio has finished. | |
274 | */ | |
fd8f8ede | 275 | struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim, |
a85b3637 | 276 | unsigned *segs, struct bio_set *bs, unsigned max_bytes) |
54efd50b | 277 | { |
5014c311 | 278 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 279 | struct bvec_iter iter; |
67927d22 | 280 | unsigned nsegs = 0, bytes = 0; |
54efd50b | 281 | |
dcebd755 | 282 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
283 | /* |
284 | * If the queue doesn't support SG gaps and adding this | |
285 | * offset would create a gap, disallow it. | |
286 | */ | |
c55ddd90 | 287 | if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset)) |
54efd50b KO |
288 | goto split; |
289 | ||
c55ddd90 | 290 | if (nsegs < lim->max_segments && |
67927d22 | 291 | bytes + bv.bv_len <= max_bytes && |
708b25b3 BVA |
292 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
293 | nsegs++; | |
67927d22 | 294 | bytes += bv.bv_len; |
c55ddd90 CH |
295 | } else { |
296 | if (bvec_split_segs(lim, &bv, &nsegs, &bytes, | |
297 | lim->max_segments, max_bytes)) | |
298 | goto split; | |
e36f6204 KB |
299 | } |
300 | ||
54efd50b | 301 | bvprv = bv; |
578270bf | 302 | bvprvp = &bvprv; |
54efd50b KO |
303 | } |
304 | ||
d627065d CH |
305 | *segs = nsegs; |
306 | return NULL; | |
54efd50b | 307 | split: |
9cea62b2 JA |
308 | /* |
309 | * We can't sanely support splitting for a REQ_NOWAIT bio. End it | |
310 | * with EAGAIN if splitting is required and return an error pointer. | |
311 | */ | |
312 | if (bio->bi_opf & REQ_NOWAIT) { | |
313 | bio->bi_status = BLK_STS_AGAIN; | |
314 | bio_endio(bio); | |
315 | return ERR_PTR(-EAGAIN); | |
316 | } | |
317 | ||
bdced438 | 318 | *segs = nsegs; |
cc29e1bf | 319 | |
67927d22 KB |
320 | /* |
321 | * Individual bvecs might not be logical block aligned. Round down the | |
322 | * split size so that each bio is properly block size aligned, even if | |
323 | * we do not use the full hardware limits. | |
324 | */ | |
c55ddd90 | 325 | bytes = ALIGN_DOWN(bytes, lim->logical_block_size); |
67927d22 | 326 | |
cc29e1bf JX |
327 | /* |
328 | * Bio splitting may cause subtle trouble such as hang when doing sync | |
329 | * iopoll in direct IO routine. Given performance gain of iopoll for | |
330 | * big IO can be trival, disable iopoll when split needed. | |
331 | */ | |
6ce913fe | 332 | bio_clear_polled(bio); |
67927d22 | 333 | return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs); |
54efd50b | 334 | } |
fd8f8ede | 335 | EXPORT_SYMBOL_GPL(bio_split_rw); |
54efd50b | 336 | |
dad77584 | 337 | /** |
5a97806f CH |
338 | * __bio_split_to_limits - split a bio to fit the queue limits |
339 | * @bio: bio to be split | |
c55ddd90 | 340 | * @lim: queue limits to split based on |
5a97806f CH |
341 | * @nr_segs: returns the number of segments in the returned bio |
342 | * | |
343 | * Check if @bio needs splitting based on the queue limits, and if so split off | |
344 | * a bio fitting the limits from the beginning of @bio and return it. @bio is | |
345 | * shortened to the remainder and re-submitted. | |
dad77584 | 346 | * |
5a97806f CH |
347 | * The split bio is allocated from @q->bio_split, which is provided by the |
348 | * block layer. | |
dad77584 | 349 | */ |
aa261f20 BVA |
350 | struct bio *__bio_split_to_limits(struct bio *bio, |
351 | const struct queue_limits *lim, | |
352 | unsigned int *nr_segs) | |
54efd50b | 353 | { |
46754bd0 | 354 | struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split; |
5a97806f | 355 | struct bio *split; |
54efd50b | 356 | |
5a97806f | 357 | switch (bio_op(bio)) { |
7afafc8a AH |
358 | case REQ_OP_DISCARD: |
359 | case REQ_OP_SECURE_ERASE: | |
c55ddd90 | 360 | split = bio_split_discard(bio, lim, nr_segs, bs); |
7afafc8a | 361 | break; |
a6f0788e | 362 | case REQ_OP_WRITE_ZEROES: |
c55ddd90 | 363 | split = bio_split_write_zeroes(bio, lim, nr_segs, bs); |
a6f0788e | 364 | break; |
7afafc8a | 365 | default: |
c55ddd90 CH |
366 | split = bio_split_rw(bio, lim, nr_segs, bs, |
367 | get_max_io_size(bio, lim) << SECTOR_SHIFT); | |
613b1488 JA |
368 | if (IS_ERR(split)) |
369 | return NULL; | |
7afafc8a AH |
370 | break; |
371 | } | |
bdced438 | 372 | |
54efd50b | 373 | if (split) { |
613b1488 | 374 | /* there isn't chance to merge the split bio */ |
1eff9d32 | 375 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 376 | |
957a2b34 | 377 | blkcg_bio_issue_init(split); |
5a97806f CH |
378 | bio_chain(split, bio); |
379 | trace_block_split(split, bio->bi_iter.bi_sector); | |
dd291d77 | 380 | WARN_ON_ONCE(bio_zone_write_plugging(bio)); |
5a97806f CH |
381 | submit_bio_noacct(bio); |
382 | return split; | |
54efd50b | 383 | } |
5a97806f | 384 | return bio; |
54efd50b | 385 | } |
14ccb66b | 386 | |
dad77584 | 387 | /** |
5a97806f CH |
388 | * bio_split_to_limits - split a bio to fit the queue limits |
389 | * @bio: bio to be split | |
390 | * | |
391 | * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and | |
392 | * if so split off a bio fitting the limits from the beginning of @bio and | |
393 | * return it. @bio is shortened to the remainder and re-submitted. | |
dad77584 | 394 | * |
5a97806f CH |
395 | * The split bio is allocated from @q->bio_split, which is provided by the |
396 | * block layer. | |
dad77584 | 397 | */ |
5a97806f | 398 | struct bio *bio_split_to_limits(struct bio *bio) |
14ccb66b | 399 | { |
aa261f20 | 400 | const struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits; |
14ccb66b CH |
401 | unsigned int nr_segs; |
402 | ||
c55ddd90 CH |
403 | if (bio_may_exceed_limits(bio, lim)) |
404 | return __bio_split_to_limits(bio, lim, &nr_segs); | |
5a97806f | 405 | return bio; |
14ccb66b | 406 | } |
5a97806f | 407 | EXPORT_SYMBOL(bio_split_to_limits); |
54efd50b | 408 | |
e9cd19c0 | 409 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 410 | { |
6869875f | 411 | unsigned int nr_phys_segs = 0; |
67927d22 | 412 | unsigned int bytes = 0; |
e9cd19c0 | 413 | struct req_iterator iter; |
6869875f | 414 | struct bio_vec bv; |
d6d48196 | 415 | |
e9cd19c0 | 416 | if (!rq->bio) |
1e428079 | 417 | return 0; |
d6d48196 | 418 | |
e9cd19c0 | 419 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
420 | case REQ_OP_DISCARD: |
421 | case REQ_OP_SECURE_ERASE: | |
a958937f DJ |
422 | if (queue_max_discard_segments(rq->q) > 1) { |
423 | struct bio *bio = rq->bio; | |
424 | ||
425 | for_each_bio(bio) | |
426 | nr_phys_segs++; | |
427 | return nr_phys_segs; | |
428 | } | |
429 | return 1; | |
a6f0788e | 430 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 | 431 | return 0; |
2d9b02be BVA |
432 | default: |
433 | break; | |
a6f0788e | 434 | } |
5cb8850c | 435 | |
e9cd19c0 | 436 | rq_for_each_bvec(bv, rq, iter) |
c55ddd90 | 437 | bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes, |
708b25b3 | 438 | UINT_MAX, UINT_MAX); |
1e428079 JA |
439 | return nr_phys_segs; |
440 | } | |
441 | ||
48d7727c | 442 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
443 | struct scatterlist *sglist) |
444 | { | |
445 | if (!*sg) | |
446 | return sglist; | |
447 | ||
448 | /* | |
449 | * If the driver previously mapped a shorter list, we could see a | |
450 | * termination bit prematurely unless it fully inits the sg table | |
451 | * on each mapping. We KNOW that there must be more entries here | |
452 | * or the driver would be buggy, so force clear the termination bit | |
453 | * to avoid doing a full sg_init_table() in drivers for each command. | |
454 | */ | |
455 | sg_unmark_end(*sg); | |
456 | return sg_next(*sg); | |
457 | } | |
458 | ||
459 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
460 | struct bio_vec *bvec, struct scatterlist *sglist, | |
461 | struct scatterlist **sg) | |
462 | { | |
463 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 464 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
465 | |
466 | while (nbytes > 0) { | |
8a96a0e4 | 467 | unsigned offset = bvec->bv_offset + total; |
c55ddd90 CH |
468 | unsigned len = min(get_max_segment_size(&q->limits, |
469 | bvec->bv_page, offset), nbytes); | |
f9f76879 CH |
470 | struct page *page = bvec->bv_page; |
471 | ||
472 | /* | |
473 | * Unfortunately a fair number of drivers barf on scatterlists | |
474 | * that have an offset larger than PAGE_SIZE, despite other | |
475 | * subsystems dealing with that invariant just fine. For now | |
476 | * stick to the legacy format where we never present those from | |
477 | * the block layer, but the code below should be removed once | |
478 | * these offenders (mostly MMC/SD drivers) are fixed. | |
479 | */ | |
480 | page += (offset >> PAGE_SHIFT); | |
481 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
482 | |
483 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 484 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 485 | |
8a96a0e4 CH |
486 | total += len; |
487 | nbytes -= len; | |
862e5a5e ML |
488 | nsegs++; |
489 | } | |
490 | ||
491 | return nsegs; | |
492 | } | |
493 | ||
16e3e418 ML |
494 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
495 | struct scatterlist *sglist, struct scatterlist **sg) | |
496 | { | |
497 | *sg = blk_next_sg(sg, sglist); | |
498 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
499 | return 1; | |
500 | } | |
501 | ||
f6970f83 ML |
502 | /* only try to merge bvecs into one sg if they are from two bios */ |
503 | static inline bool | |
504 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
505 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
506 | { |
507 | ||
508 | int nbytes = bvec->bv_len; | |
509 | ||
f6970f83 ML |
510 | if (!*sg) |
511 | return false; | |
963ab9e5 | 512 | |
f6970f83 ML |
513 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
514 | return false; | |
515 | ||
516 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
517 | return false; | |
518 | ||
519 | (*sg)->length += nbytes; | |
520 | ||
521 | return true; | |
963ab9e5 AH |
522 | } |
523 | ||
5cb8850c KO |
524 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
525 | struct scatterlist *sglist, | |
526 | struct scatterlist **sg) | |
d6d48196 | 527 | { |
3f649ab7 | 528 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 529 | struct bvec_iter iter; |
38417468 | 530 | int nsegs = 0; |
f6970f83 | 531 | bool new_bio = false; |
5cb8850c | 532 | |
f6970f83 ML |
533 | for_each_bio(bio) { |
534 | bio_for_each_bvec(bvec, bio, iter) { | |
535 | /* | |
536 | * Only try to merge bvecs from two bios given we | |
537 | * have done bio internal merge when adding pages | |
538 | * to bio | |
539 | */ | |
540 | if (new_bio && | |
541 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
542 | goto next_bvec; | |
543 | ||
544 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
545 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
546 | else | |
547 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
548 | next_bvec: | |
549 | new_bio = false; | |
550 | } | |
b21e11c5 ML |
551 | if (likely(bio->bi_iter.bi_size)) { |
552 | bvprv = bvec; | |
553 | new_bio = true; | |
554 | } | |
f6970f83 | 555 | } |
d6d48196 | 556 | |
5cb8850c KO |
557 | return nsegs; |
558 | } | |
559 | ||
560 | /* | |
561 | * map a request to scatterlist, return number of sg entries setup. Caller | |
562 | * must make sure sg can hold rq->nr_phys_segments entries | |
563 | */ | |
89de1504 CH |
564 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
565 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 566 | { |
5cb8850c KO |
567 | int nsegs = 0; |
568 | ||
f9d03f96 | 569 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 570 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 571 | else if (rq->bio) |
89de1504 | 572 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 573 | |
89de1504 CH |
574 | if (*last_sg) |
575 | sg_mark_end(*last_sg); | |
d6d48196 | 576 | |
12e57f59 ML |
577 | /* |
578 | * Something must have been wrong if the figured number of | |
579 | * segment is bigger than number of req's physical segments | |
580 | */ | |
f9d03f96 | 581 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 582 | |
d6d48196 JA |
583 | return nsegs; |
584 | } | |
89de1504 | 585 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 586 | |
badf7f64 CH |
587 | static inline unsigned int blk_rq_get_max_sectors(struct request *rq, |
588 | sector_t offset) | |
589 | { | |
590 | struct request_queue *q = rq->q; | |
c8875190 | 591 | unsigned int max_sectors; |
badf7f64 CH |
592 | |
593 | if (blk_rq_is_passthrough(rq)) | |
594 | return q->limits.max_hw_sectors; | |
595 | ||
c8875190 | 596 | max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
badf7f64 CH |
597 | if (!q->limits.chunk_sectors || |
598 | req_op(rq) == REQ_OP_DISCARD || | |
599 | req_op(rq) == REQ_OP_SECURE_ERASE) | |
c8875190 CH |
600 | return max_sectors; |
601 | return min(max_sectors, | |
602 | blk_chunk_sectors_left(offset, q->limits.chunk_sectors)); | |
badf7f64 CH |
603 | } |
604 | ||
14ccb66b CH |
605 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
606 | unsigned int nr_phys_segs) | |
d6d48196 | 607 | { |
6b2b0459 TH |
608 | if (!blk_cgroup_mergeable(req, bio)) |
609 | goto no_merge; | |
610 | ||
2705dfb2 | 611 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d MP |
612 | goto no_merge; |
613 | ||
2705dfb2 ML |
614 | /* discard request merge won't add new segment */ |
615 | if (req_op(req) == REQ_OP_DISCARD) | |
616 | return 1; | |
617 | ||
618 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) | |
13f05c8d | 619 | goto no_merge; |
d6d48196 JA |
620 | |
621 | /* | |
622 | * This will form the start of a new hw segment. Bump both | |
623 | * counters. | |
624 | */ | |
d6d48196 JA |
625 | req->nr_phys_segments += nr_phys_segs; |
626 | return 1; | |
13f05c8d MP |
627 | |
628 | no_merge: | |
14ccb66b | 629 | req_set_nomerge(req->q, req); |
13f05c8d | 630 | return 0; |
d6d48196 JA |
631 | } |
632 | ||
14ccb66b | 633 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 634 | { |
5e7c4274 JA |
635 | if (req_gap_back_merge(req, bio)) |
636 | return 0; | |
7f39add3 SG |
637 | if (blk_integrity_rq(req) && |
638 | integrity_req_gap_back_merge(req, bio)) | |
639 | return 0; | |
a892c8d5 ST |
640 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
641 | return 0; | |
f31dc1cd | 642 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 643 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 644 | req_set_nomerge(req->q, req); |
d6d48196 JA |
645 | return 0; |
646 | } | |
d6d48196 | 647 | |
14ccb66b | 648 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
649 | } |
650 | ||
eda5cc99 CH |
651 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
652 | unsigned int nr_segs) | |
d6d48196 | 653 | { |
5e7c4274 JA |
654 | if (req_gap_front_merge(req, bio)) |
655 | return 0; | |
7f39add3 SG |
656 | if (blk_integrity_rq(req) && |
657 | integrity_req_gap_front_merge(req, bio)) | |
658 | return 0; | |
a892c8d5 ST |
659 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
660 | return 0; | |
f31dc1cd | 661 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 662 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 663 | req_set_nomerge(req->q, req); |
d6d48196 JA |
664 | return 0; |
665 | } | |
d6d48196 | 666 | |
14ccb66b | 667 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
668 | } |
669 | ||
445251d0 JA |
670 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
671 | struct request *next) | |
672 | { | |
673 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
674 | ||
675 | if (segments >= queue_max_discard_segments(q)) | |
676 | goto no_merge; | |
677 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
678 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
679 | goto no_merge; | |
680 | ||
681 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
682 | return true; | |
683 | no_merge: | |
684 | req_set_nomerge(q, req); | |
685 | return false; | |
686 | } | |
687 | ||
d6d48196 JA |
688 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
689 | struct request *next) | |
690 | { | |
691 | int total_phys_segments; | |
d6d48196 | 692 | |
5e7c4274 | 693 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
694 | return 0; |
695 | ||
d6d48196 JA |
696 | /* |
697 | * Will it become too large? | |
698 | */ | |
f31dc1cd | 699 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 700 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
701 | return 0; |
702 | ||
703 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 704 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
705 | return 0; |
706 | ||
6b2b0459 TH |
707 | if (!blk_cgroup_mergeable(req, next->bio)) |
708 | return 0; | |
709 | ||
4eaf99be | 710 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
711 | return 0; |
712 | ||
a892c8d5 ST |
713 | if (!bio_crypt_ctx_merge_rq(req, next)) |
714 | return 0; | |
715 | ||
d6d48196 JA |
716 | /* Merge is OK... */ |
717 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
718 | return 1; |
719 | } | |
720 | ||
80a761fd TH |
721 | /** |
722 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
723 | * @rq: request to mark as mixed merge | |
724 | * | |
725 | * Description: | |
726 | * @rq is about to be mixed merged. Make sure the attributes | |
727 | * which can be mixed are set in each bio and mark @rq as mixed | |
728 | * merged. | |
729 | */ | |
dc53d9ea | 730 | static void blk_rq_set_mixed_merge(struct request *rq) |
80a761fd | 731 | { |
16458cf3 | 732 | blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
80a761fd TH |
733 | struct bio *bio; |
734 | ||
e8064021 | 735 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
736 | return; |
737 | ||
738 | /* | |
739 | * @rq will no longer represent mixable attributes for all the | |
740 | * contained bios. It will just track those of the first one. | |
741 | * Distributes the attributs to each bio. | |
742 | */ | |
743 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
744 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
745 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
746 | bio->bi_opf |= ff; | |
80a761fd | 747 | } |
e8064021 | 748 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
749 | } |
750 | ||
f3ca7386 | 751 | static inline blk_opf_t bio_failfast(const struct bio *bio) |
3ce6a115 ML |
752 | { |
753 | if (bio->bi_opf & REQ_RAHEAD) | |
754 | return REQ_FAILFAST_MASK; | |
755 | ||
756 | return bio->bi_opf & REQ_FAILFAST_MASK; | |
757 | } | |
758 | ||
759 | /* | |
760 | * After we are marked as MIXED_MERGE, any new RA bio has to be updated | |
761 | * as failfast, and request's failfast has to be updated in case of | |
762 | * front merge. | |
763 | */ | |
764 | static inline void blk_update_mixed_merge(struct request *req, | |
765 | struct bio *bio, bool front_merge) | |
766 | { | |
767 | if (req->rq_flags & RQF_MIXED_MERGE) { | |
768 | if (bio->bi_opf & REQ_RAHEAD) | |
769 | bio->bi_opf |= REQ_FAILFAST_MASK; | |
770 | ||
771 | if (front_merge) { | |
772 | req->cmd_flags &= ~REQ_FAILFAST_MASK; | |
773 | req->cmd_flags |= bio->bi_opf & REQ_FAILFAST_MASK; | |
774 | } | |
775 | } | |
776 | } | |
777 | ||
b9c54f56 | 778 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
779 | { |
780 | if (blk_do_io_stat(req)) { | |
112f158f | 781 | part_stat_lock(); |
b9c54f56 | 782 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab JM |
783 | part_stat_unlock(); |
784 | } | |
785 | } | |
b9c54f56 | 786 | |
e96c0d83 EB |
787 | static enum elv_merge blk_try_req_merge(struct request *req, |
788 | struct request *next) | |
69840466 JW |
789 | { |
790 | if (blk_discard_mergable(req)) | |
791 | return ELEVATOR_DISCARD_MERGE; | |
792 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
793 | return ELEVATOR_BACK_MERGE; | |
794 | ||
795 | return ELEVATOR_NO_MERGE; | |
796 | } | |
26308eab | 797 | |
d6d48196 | 798 | /* |
b973cb7e JA |
799 | * For non-mq, this has to be called with the request spinlock acquired. |
800 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 801 | */ |
b973cb7e JA |
802 | static struct request *attempt_merge(struct request_queue *q, |
803 | struct request *req, struct request *next) | |
d6d48196 JA |
804 | { |
805 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 806 | return NULL; |
d6d48196 | 807 | |
288dab8a | 808 | if (req_op(req) != req_op(next)) |
b973cb7e | 809 | return NULL; |
f31dc1cd | 810 | |
79bb1dbd | 811 | if (rq_data_dir(req) != rq_data_dir(next)) |
b973cb7e | 812 | return NULL; |
d6d48196 | 813 | |
44981351 BVA |
814 | /* Don't merge requests with different write hints. */ |
815 | if (req->write_hint != next->write_hint) | |
816 | return NULL; | |
817 | ||
668ffc03 DLM |
818 | if (req->ioprio != next->ioprio) |
819 | return NULL; | |
820 | ||
d6d48196 JA |
821 | /* |
822 | * If we are allowed to merge, then append bio list | |
823 | * from next to rq and release next. merge_requests_fn | |
824 | * will have updated segment counts, update sector | |
445251d0 JA |
825 | * counts here. Handle DISCARDs separately, as they |
826 | * have separate settings. | |
d6d48196 | 827 | */ |
69840466 JW |
828 | |
829 | switch (blk_try_req_merge(req, next)) { | |
830 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
831 | if (!req_attempt_discard_merge(q, req, next)) |
832 | return NULL; | |
69840466 JW |
833 | break; |
834 | case ELEVATOR_BACK_MERGE: | |
835 | if (!ll_merge_requests_fn(q, req, next)) | |
836 | return NULL; | |
837 | break; | |
838 | default: | |
b973cb7e | 839 | return NULL; |
69840466 | 840 | } |
d6d48196 | 841 | |
80a761fd TH |
842 | /* |
843 | * If failfast settings disagree or any of the two is already | |
844 | * a mixed merge, mark both as mixed before proceeding. This | |
845 | * makes sure that all involved bios have mixable attributes | |
846 | * set properly. | |
847 | */ | |
e8064021 | 848 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
849 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
850 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
851 | blk_rq_set_mixed_merge(req); | |
852 | blk_rq_set_mixed_merge(next); | |
853 | } | |
854 | ||
d6d48196 | 855 | /* |
522a7775 OS |
856 | * At this point we have either done a back merge or front merge. We |
857 | * need the smaller start_time_ns of the merged requests to be the | |
858 | * current request for accounting purposes. | |
d6d48196 | 859 | */ |
522a7775 OS |
860 | if (next->start_time_ns < req->start_time_ns) |
861 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
862 | |
863 | req->biotail->bi_next = next->bio; | |
864 | req->biotail = next->biotail; | |
865 | ||
a2dec7b3 | 866 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 867 | |
2a5cf35c | 868 | if (!blk_discard_mergable(req)) |
445251d0 | 869 | elv_merge_requests(q, req, next); |
d6d48196 | 870 | |
9cd1e566 EB |
871 | blk_crypto_rq_put_keyslot(next); |
872 | ||
42dad764 JM |
873 | /* |
874 | * 'next' is going away, so update stats accordingly | |
875 | */ | |
b9c54f56 | 876 | blk_account_io_merge_request(next); |
d6d48196 | 877 | |
a54895fa | 878 | trace_block_rq_merge(next); |
f3bdc62f | 879 | |
e4d750c9 JA |
880 | /* |
881 | * ownership of bio passed from next to req, return 'next' for | |
882 | * the caller to free | |
883 | */ | |
1cd96c24 | 884 | next->bio = NULL; |
b973cb7e | 885 | return next; |
d6d48196 JA |
886 | } |
887 | ||
eda5cc99 CH |
888 | static struct request *attempt_back_merge(struct request_queue *q, |
889 | struct request *rq) | |
d6d48196 JA |
890 | { |
891 | struct request *next = elv_latter_request(q, rq); | |
892 | ||
893 | if (next) | |
894 | return attempt_merge(q, rq, next); | |
895 | ||
b973cb7e | 896 | return NULL; |
d6d48196 JA |
897 | } |
898 | ||
eda5cc99 CH |
899 | static struct request *attempt_front_merge(struct request_queue *q, |
900 | struct request *rq) | |
d6d48196 JA |
901 | { |
902 | struct request *prev = elv_former_request(q, rq); | |
903 | ||
904 | if (prev) | |
905 | return attempt_merge(q, prev, rq); | |
906 | ||
b973cb7e | 907 | return NULL; |
d6d48196 | 908 | } |
5e84ea3a | 909 | |
fd2ef39c JK |
910 | /* |
911 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false | |
912 | * otherwise. The caller is responsible for freeing 'next' if the merge | |
913 | * happened. | |
914 | */ | |
915 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
916 | struct request *next) | |
5e84ea3a | 917 | { |
fd2ef39c | 918 | return attempt_merge(q, rq, next); |
5e84ea3a | 919 | } |
050c8ea8 TH |
920 | |
921 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
922 | { | |
e2a60da7 | 923 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
924 | return false; |
925 | ||
288dab8a | 926 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
927 | return false; |
928 | ||
050c8ea8 TH |
929 | /* different data direction or already started, don't merge */ |
930 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
931 | return false; | |
932 | ||
6b2b0459 TH |
933 | /* don't merge across cgroup boundaries */ |
934 | if (!blk_cgroup_mergeable(rq, bio)) | |
935 | return false; | |
936 | ||
050c8ea8 | 937 | /* only merge integrity protected bio into ditto rq */ |
4eaf99be | 938 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
939 | return false; |
940 | ||
a892c8d5 ST |
941 | /* Only merge if the crypt contexts are compatible */ |
942 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
943 | return false; | |
944 | ||
44981351 BVA |
945 | /* Don't merge requests with different write hints. */ |
946 | if (rq->write_hint != bio->bi_write_hint) | |
947 | return false; | |
948 | ||
668ffc03 DLM |
949 | if (rq->ioprio != bio_prio(bio)) |
950 | return false; | |
951 | ||
050c8ea8 TH |
952 | return true; |
953 | } | |
954 | ||
34fe7c05 | 955 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 956 | { |
69840466 | 957 | if (blk_discard_mergable(rq)) |
1e739730 CH |
958 | return ELEVATOR_DISCARD_MERGE; |
959 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 960 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 961 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
962 | return ELEVATOR_FRONT_MERGE; |
963 | return ELEVATOR_NO_MERGE; | |
964 | } | |
8e756373 BW |
965 | |
966 | static void blk_account_io_merge_bio(struct request *req) | |
967 | { | |
968 | if (!blk_do_io_stat(req)) | |
969 | return; | |
970 | ||
971 | part_stat_lock(); | |
972 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
973 | part_stat_unlock(); | |
974 | } | |
975 | ||
dd850ff3 | 976 | enum bio_merge_status bio_attempt_back_merge(struct request *req, |
eda5cc99 | 977 | struct bio *bio, unsigned int nr_segs) |
8e756373 | 978 | { |
3ce6a115 | 979 | const blk_opf_t ff = bio_failfast(bio); |
8e756373 BW |
980 | |
981 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 982 | return BIO_MERGE_FAILED; |
8e756373 | 983 | |
e8a676d6 | 984 | trace_block_bio_backmerge(bio); |
8e756373 BW |
985 | rq_qos_merge(req->q, req, bio); |
986 | ||
987 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
988 | blk_rq_set_mixed_merge(req); | |
989 | ||
3ce6a115 ML |
990 | blk_update_mixed_merge(req, bio, false); |
991 | ||
dd291d77 DLM |
992 | if (req->rq_flags & RQF_ZONE_WRITE_PLUGGING) |
993 | blk_zone_write_plug_bio_merged(bio); | |
994 | ||
8e756373 BW |
995 | req->biotail->bi_next = bio; |
996 | req->biotail = bio; | |
997 | req->__data_len += bio->bi_iter.bi_size; | |
998 | ||
999 | bio_crypt_free_ctx(bio); | |
1000 | ||
1001 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1002 | return BIO_MERGE_OK; |
8e756373 BW |
1003 | } |
1004 | ||
eda5cc99 CH |
1005 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
1006 | struct bio *bio, unsigned int nr_segs) | |
8e756373 | 1007 | { |
3ce6a115 | 1008 | const blk_opf_t ff = bio_failfast(bio); |
8e756373 | 1009 | |
dd291d77 DLM |
1010 | /* |
1011 | * A front merge for writes to sequential zones of a zoned block device | |
1012 | * can happen only if the user submitted writes out of order. Do not | |
1013 | * merge such write to let it fail. | |
1014 | */ | |
1015 | if (req->rq_flags & RQF_ZONE_WRITE_PLUGGING) | |
1016 | return BIO_MERGE_FAILED; | |
1017 | ||
8e756373 | 1018 | if (!ll_front_merge_fn(req, bio, nr_segs)) |
7d7ca7c5 | 1019 | return BIO_MERGE_FAILED; |
8e756373 | 1020 | |
e8a676d6 | 1021 | trace_block_bio_frontmerge(bio); |
8e756373 BW |
1022 | rq_qos_merge(req->q, req, bio); |
1023 | ||
1024 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
1025 | blk_rq_set_mixed_merge(req); | |
1026 | ||
3ce6a115 ML |
1027 | blk_update_mixed_merge(req, bio, true); |
1028 | ||
8e756373 BW |
1029 | bio->bi_next = req->bio; |
1030 | req->bio = bio; | |
1031 | ||
1032 | req->__sector = bio->bi_iter.bi_sector; | |
1033 | req->__data_len += bio->bi_iter.bi_size; | |
1034 | ||
1035 | bio_crypt_do_front_merge(req, bio); | |
1036 | ||
1037 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1038 | return BIO_MERGE_OK; |
8e756373 BW |
1039 | } |
1040 | ||
eda5cc99 CH |
1041 | static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
1042 | struct request *req, struct bio *bio) | |
8e756373 BW |
1043 | { |
1044 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1045 | ||
1046 | if (segments >= queue_max_discard_segments(q)) | |
1047 | goto no_merge; | |
1048 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1049 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1050 | goto no_merge; | |
1051 | ||
1052 | rq_qos_merge(q, req, bio); | |
1053 | ||
1054 | req->biotail->bi_next = bio; | |
1055 | req->biotail = bio; | |
1056 | req->__data_len += bio->bi_iter.bi_size; | |
1057 | req->nr_phys_segments = segments + 1; | |
1058 | ||
1059 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1060 | return BIO_MERGE_OK; |
8e756373 BW |
1061 | no_merge: |
1062 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
1063 | return BIO_MERGE_FAILED; |
1064 | } | |
1065 | ||
1066 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
1067 | struct request *rq, | |
1068 | struct bio *bio, | |
1069 | unsigned int nr_segs, | |
1070 | bool sched_allow_merge) | |
1071 | { | |
1072 | if (!blk_rq_merge_ok(rq, bio)) | |
1073 | return BIO_MERGE_NONE; | |
1074 | ||
1075 | switch (blk_try_merge(rq, bio)) { | |
1076 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 1077 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1078 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1079 | break; | |
1080 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1081 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1082 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1083 | break; | |
1084 | case ELEVATOR_DISCARD_MERGE: | |
1085 | return bio_attempt_discard_merge(q, rq, bio); | |
1086 | default: | |
1087 | return BIO_MERGE_NONE; | |
1088 | } | |
1089 | ||
1090 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1091 | } |
1092 | ||
1093 | /** | |
1094 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1095 | * @q: request_queue new bio is being queued at | |
1096 | * @bio: new bio being queued | |
1097 | * @nr_segs: number of segments in @bio | |
87c037d1 | 1098 | * from the passed in @q already in the plug list |
8e756373 | 1099 | * |
d38a9c04 JA |
1100 | * Determine whether @bio being queued on @q can be merged with the previous |
1101 | * request on %current's plugged list. Returns %true if merge was successful, | |
8e756373 BW |
1102 | * otherwise %false. |
1103 | * | |
1104 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1105 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1106 | * than scheduling, and the request, while may have elvpriv data, is not | |
1107 | * added on the elevator at this point. In addition, we don't have | |
1108 | * reliable access to the elevator outside queue lock. Only check basic | |
1109 | * merging parameters without querying the elevator. | |
1110 | * | |
1111 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1112 | */ | |
1113 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
0c5bcc92 | 1114 | unsigned int nr_segs) |
8e756373 | 1115 | { |
99a9476b | 1116 | struct blk_plug *plug = current->plug; |
8e756373 | 1117 | struct request *rq; |
8e756373 | 1118 | |
bc490f81 | 1119 | if (!plug || rq_list_empty(plug->mq_list)) |
8e756373 BW |
1120 | return false; |
1121 | ||
5b205071 JA |
1122 | rq_list_for_each(&plug->mq_list, rq) { |
1123 | if (rq->q == q) { | |
1124 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == | |
1125 | BIO_MERGE_OK) | |
1126 | return true; | |
1127 | break; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * Only keep iterating plug list for merges if we have multiple | |
1132 | * queues | |
1133 | */ | |
1134 | if (!plug->multiple_queues) | |
1135 | break; | |
8e756373 | 1136 | } |
8e756373 BW |
1137 | return false; |
1138 | } | |
bdc6a287 BW |
1139 | |
1140 | /* | |
1141 | * Iterate list of requests and see if we can merge this bio with any | |
1142 | * of them. | |
1143 | */ | |
1144 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1145 | struct bio *bio, unsigned int nr_segs) | |
1146 | { | |
1147 | struct request *rq; | |
1148 | int checked = 8; | |
1149 | ||
1150 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1151 | if (!checked--) |
1152 | break; | |
1153 | ||
7d7ca7c5 BW |
1154 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1155 | case BIO_MERGE_NONE: | |
bdc6a287 | 1156 | continue; |
7d7ca7c5 BW |
1157 | case BIO_MERGE_OK: |
1158 | return true; | |
1159 | case BIO_MERGE_FAILED: | |
1160 | return false; | |
bdc6a287 BW |
1161 | } |
1162 | ||
bdc6a287 BW |
1163 | } |
1164 | ||
1165 | return false; | |
1166 | } | |
1167 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); | |
eda5cc99 CH |
1168 | |
1169 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, | |
1170 | unsigned int nr_segs, struct request **merged_request) | |
1171 | { | |
1172 | struct request *rq; | |
1173 | ||
1174 | switch (elv_merge(q, &rq, bio)) { | |
1175 | case ELEVATOR_BACK_MERGE: | |
1176 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1177 | return false; | |
1178 | if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1179 | return false; | |
1180 | *merged_request = attempt_back_merge(q, rq); | |
1181 | if (!*merged_request) | |
1182 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
1183 | return true; | |
1184 | case ELEVATOR_FRONT_MERGE: | |
1185 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1186 | return false; | |
1187 | if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1188 | return false; | |
1189 | *merged_request = attempt_front_merge(q, rq); | |
1190 | if (!*merged_request) | |
1191 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
1192 | return true; | |
1193 | case ELEVATOR_DISCARD_MERGE: | |
1194 | return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; | |
1195 | default: | |
1196 | return false; | |
1197 | } | |
1198 | } | |
1199 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |