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 | ||
54efd50b | 118 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
c55ddd90 | 119 | granularity = max(lim->discard_granularity >> 9, 1U); |
54efd50b | 120 | |
c55ddd90 CH |
121 | max_discard_sectors = |
122 | min(lim->max_discard_sectors, bio_allowed_max_sectors(lim)); | |
54efd50b KO |
123 | max_discard_sectors -= max_discard_sectors % granularity; |
124 | ||
125 | if (unlikely(!max_discard_sectors)) { | |
126 | /* XXX: warn */ | |
127 | return NULL; | |
128 | } | |
129 | ||
130 | if (bio_sectors(bio) <= max_discard_sectors) | |
131 | return NULL; | |
132 | ||
133 | split_sectors = max_discard_sectors; | |
134 | ||
135 | /* | |
136 | * If the next starting sector would be misaligned, stop the discard at | |
137 | * the previous aligned sector. | |
138 | */ | |
c55ddd90 CH |
139 | tmp = bio->bi_iter.bi_sector + split_sectors - |
140 | ((lim->discard_alignment >> 9) % granularity); | |
54efd50b KO |
141 | tmp = sector_div(tmp, granularity); |
142 | ||
143 | if (split_sectors > tmp) | |
144 | split_sectors -= tmp; | |
145 | ||
146 | return bio_split(bio, split_sectors, GFP_NOIO, bs); | |
147 | } | |
148 | ||
5a97806f | 149 | static struct bio *bio_split_write_zeroes(struct bio *bio, |
aa261f20 BVA |
150 | const struct queue_limits *lim, |
151 | unsigned *nsegs, struct bio_set *bs) | |
885fa13f | 152 | { |
d665e12a | 153 | *nsegs = 0; |
c55ddd90 | 154 | if (!lim->max_write_zeroes_sectors) |
885fa13f | 155 | return NULL; |
c55ddd90 | 156 | if (bio_sectors(bio) <= lim->max_write_zeroes_sectors) |
885fa13f | 157 | return NULL; |
c55ddd90 | 158 | return bio_split(bio, lim->max_write_zeroes_sectors, GFP_NOIO, bs); |
885fa13f CH |
159 | } |
160 | ||
9cc5169c BVA |
161 | /* |
162 | * Return the maximum number of sectors from the start of a bio that may be | |
163 | * submitted as a single request to a block device. If enough sectors remain, | |
164 | * align the end to the physical block size. Otherwise align the end to the | |
165 | * logical block size. This approach minimizes the number of non-aligned | |
166 | * requests that are submitted to a block device if the start of a bio is not | |
167 | * aligned to a physical block boundary. | |
168 | */ | |
5a97806f | 169 | static inline unsigned get_max_io_size(struct bio *bio, |
aa261f20 | 170 | const struct queue_limits *lim) |
d0e5fbb0 | 171 | { |
c55ddd90 CH |
172 | unsigned pbs = lim->physical_block_size >> SECTOR_SHIFT; |
173 | unsigned lbs = lim->logical_block_size >> SECTOR_SHIFT; | |
174 | unsigned max_sectors = lim->max_sectors, start, end; | |
d0e5fbb0 | 175 | |
c55ddd90 | 176 | if (lim->chunk_sectors) { |
efef739d CH |
177 | max_sectors = min(max_sectors, |
178 | blk_chunk_sectors_left(bio->bi_iter.bi_sector, | |
c55ddd90 | 179 | lim->chunk_sectors)); |
efef739d | 180 | } |
d0e5fbb0 | 181 | |
84613bed CH |
182 | start = bio->bi_iter.bi_sector & (pbs - 1); |
183 | end = (start + max_sectors) & ~(pbs - 1); | |
184 | if (end > start) | |
185 | return end - start; | |
186 | return max_sectors & ~(lbs - 1); | |
d0e5fbb0 ML |
187 | } |
188 | ||
95465318 BVA |
189 | /** |
190 | * get_max_segment_size() - maximum number of bytes to add as a single segment | |
191 | * @lim: Request queue limits. | |
192 | * @start_page: See below. | |
193 | * @offset: Offset from @start_page where to add a segment. | |
194 | * | |
195 | * Returns the maximum number of bytes that can be added as a single segment. | |
196 | */ | |
aa261f20 | 197 | static inline unsigned get_max_segment_size(const struct queue_limits *lim, |
c55ddd90 | 198 | struct page *start_page, unsigned long offset) |
dcebd755 | 199 | { |
c55ddd90 | 200 | unsigned long mask = lim->seg_boundary_mask; |
dcebd755 | 201 | |
429120f3 | 202 | offset = mask & (page_to_phys(start_page) + offset); |
4a2f704e ML |
203 | |
204 | /* | |
95465318 BVA |
205 | * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1 |
206 | * after having calculated the minimum. | |
4a2f704e | 207 | */ |
95465318 | 208 | return min(mask - offset, (unsigned long)lim->max_segment_size - 1) + 1; |
dcebd755 ML |
209 | } |
210 | ||
708b25b3 BVA |
211 | /** |
212 | * bvec_split_segs - verify whether or not a bvec should be split in the middle | |
c55ddd90 | 213 | * @lim: [in] queue limits to split based on |
708b25b3 BVA |
214 | * @bv: [in] bvec to examine |
215 | * @nsegs: [in,out] Number of segments in the bio being built. Incremented | |
216 | * by the number of segments from @bv that may be appended to that | |
217 | * bio without exceeding @max_segs | |
67927d22 KB |
218 | * @bytes: [in,out] Number of bytes in the bio being built. Incremented |
219 | * by the number of bytes from @bv that may be appended to that | |
220 | * bio without exceeding @max_bytes | |
708b25b3 | 221 | * @max_segs: [in] upper bound for *@nsegs |
67927d22 | 222 | * @max_bytes: [in] upper bound for *@bytes |
708b25b3 BVA |
223 | * |
224 | * When splitting a bio, it can happen that a bvec is encountered that is too | |
225 | * big to fit in a single segment and hence that it has to be split in the | |
226 | * middle. This function verifies whether or not that should happen. The value | |
227 | * %true is returned if and only if appending the entire @bv to a bio with | |
228 | * *@nsegs segments and *@sectors sectors would make that bio unacceptable for | |
229 | * the block driver. | |
dcebd755 | 230 | */ |
aa261f20 BVA |
231 | static bool bvec_split_segs(const struct queue_limits *lim, |
232 | const struct bio_vec *bv, unsigned *nsegs, unsigned *bytes, | |
233 | unsigned max_segs, unsigned max_bytes) | |
dcebd755 | 234 | { |
67927d22 | 235 | unsigned max_len = min(max_bytes, UINT_MAX) - *bytes; |
708b25b3 | 236 | unsigned len = min(bv->bv_len, max_len); |
dcebd755 | 237 | unsigned total_len = 0; |
ff9811b3 | 238 | unsigned seg_size = 0; |
dcebd755 | 239 | |
ff9811b3 | 240 | while (len && *nsegs < max_segs) { |
c55ddd90 | 241 | seg_size = get_max_segment_size(lim, bv->bv_page, |
429120f3 | 242 | bv->bv_offset + total_len); |
dcebd755 ML |
243 | seg_size = min(seg_size, len); |
244 | ||
ff9811b3 | 245 | (*nsegs)++; |
dcebd755 ML |
246 | total_len += seg_size; |
247 | len -= seg_size; | |
248 | ||
c55ddd90 | 249 | if ((bv->bv_offset + total_len) & lim->virt_boundary_mask) |
dcebd755 ML |
250 | break; |
251 | } | |
252 | ||
67927d22 | 253 | *bytes += total_len; |
dcebd755 | 254 | |
708b25b3 BVA |
255 | /* tell the caller to split the bvec if it is too big to fit */ |
256 | return len > 0 || bv->bv_len > max_len; | |
dcebd755 ML |
257 | } |
258 | ||
dad77584 | 259 | /** |
5a97806f | 260 | * bio_split_rw - split a bio in two bios |
dad77584 | 261 | * @bio: [in] bio to be split |
c55ddd90 | 262 | * @lim: [in] queue limits to split based on |
dad77584 | 263 | * @segs: [out] number of segments in the bio with the first half of the sectors |
5a97806f | 264 | * @bs: [in] bio set to allocate the clone from |
a85b3637 | 265 | * @max_bytes: [in] maximum number of bytes per bio |
dad77584 BVA |
266 | * |
267 | * Clone @bio, update the bi_iter of the clone to represent the first sectors | |
268 | * of @bio and update @bio->bi_iter to represent the remaining sectors. The | |
269 | * following is guaranteed for the cloned bio: | |
a85b3637 | 270 | * - That it has at most @max_bytes worth of data |
dad77584 BVA |
271 | * - That it has at most queue_max_segments(@q) segments. |
272 | * | |
273 | * Except for discard requests the cloned bio will point at the bi_io_vec of | |
274 | * the original bio. It is the responsibility of the caller to ensure that the | |
275 | * original bio is not freed before the cloned bio. The caller is also | |
276 | * responsible for ensuring that @bs is only destroyed after processing of the | |
277 | * split bio has finished. | |
278 | */ | |
fd8f8ede | 279 | struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim, |
a85b3637 | 280 | unsigned *segs, struct bio_set *bs, unsigned max_bytes) |
54efd50b | 281 | { |
5014c311 | 282 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
54efd50b | 283 | struct bvec_iter iter; |
67927d22 | 284 | unsigned nsegs = 0, bytes = 0; |
54efd50b | 285 | |
dcebd755 | 286 | bio_for_each_bvec(bv, bio, iter) { |
54efd50b KO |
287 | /* |
288 | * If the queue doesn't support SG gaps and adding this | |
289 | * offset would create a gap, disallow it. | |
290 | */ | |
c55ddd90 | 291 | if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv.bv_offset)) |
54efd50b KO |
292 | goto split; |
293 | ||
c55ddd90 | 294 | if (nsegs < lim->max_segments && |
67927d22 | 295 | bytes + bv.bv_len <= max_bytes && |
708b25b3 BVA |
296 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
297 | nsegs++; | |
67927d22 | 298 | bytes += bv.bv_len; |
c55ddd90 CH |
299 | } else { |
300 | if (bvec_split_segs(lim, &bv, &nsegs, &bytes, | |
301 | lim->max_segments, max_bytes)) | |
302 | goto split; | |
e36f6204 KB |
303 | } |
304 | ||
54efd50b | 305 | bvprv = bv; |
578270bf | 306 | bvprvp = &bvprv; |
54efd50b KO |
307 | } |
308 | ||
d627065d CH |
309 | *segs = nsegs; |
310 | return NULL; | |
54efd50b | 311 | split: |
9cea62b2 JA |
312 | /* |
313 | * We can't sanely support splitting for a REQ_NOWAIT bio. End it | |
314 | * with EAGAIN if splitting is required and return an error pointer. | |
315 | */ | |
316 | if (bio->bi_opf & REQ_NOWAIT) { | |
317 | bio->bi_status = BLK_STS_AGAIN; | |
318 | bio_endio(bio); | |
319 | return ERR_PTR(-EAGAIN); | |
320 | } | |
321 | ||
bdced438 | 322 | *segs = nsegs; |
cc29e1bf | 323 | |
67927d22 KB |
324 | /* |
325 | * Individual bvecs might not be logical block aligned. Round down the | |
326 | * split size so that each bio is properly block size aligned, even if | |
327 | * we do not use the full hardware limits. | |
328 | */ | |
c55ddd90 | 329 | bytes = ALIGN_DOWN(bytes, lim->logical_block_size); |
67927d22 | 330 | |
cc29e1bf JX |
331 | /* |
332 | * Bio splitting may cause subtle trouble such as hang when doing sync | |
333 | * iopoll in direct IO routine. Given performance gain of iopoll for | |
334 | * big IO can be trival, disable iopoll when split needed. | |
335 | */ | |
6ce913fe | 336 | bio_clear_polled(bio); |
67927d22 | 337 | return bio_split(bio, bytes >> SECTOR_SHIFT, GFP_NOIO, bs); |
54efd50b | 338 | } |
fd8f8ede | 339 | EXPORT_SYMBOL_GPL(bio_split_rw); |
54efd50b | 340 | |
dad77584 | 341 | /** |
5a97806f CH |
342 | * __bio_split_to_limits - split a bio to fit the queue limits |
343 | * @bio: bio to be split | |
c55ddd90 | 344 | * @lim: queue limits to split based on |
5a97806f CH |
345 | * @nr_segs: returns the number of segments in the returned bio |
346 | * | |
347 | * Check if @bio needs splitting based on the queue limits, and if so split off | |
348 | * a bio fitting the limits from the beginning of @bio and return it. @bio is | |
349 | * shortened to the remainder and re-submitted. | |
dad77584 | 350 | * |
5a97806f CH |
351 | * The split bio is allocated from @q->bio_split, which is provided by the |
352 | * block layer. | |
dad77584 | 353 | */ |
aa261f20 BVA |
354 | struct bio *__bio_split_to_limits(struct bio *bio, |
355 | const struct queue_limits *lim, | |
356 | unsigned int *nr_segs) | |
54efd50b | 357 | { |
46754bd0 | 358 | struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split; |
5a97806f | 359 | struct bio *split; |
54efd50b | 360 | |
5a97806f | 361 | switch (bio_op(bio)) { |
7afafc8a AH |
362 | case REQ_OP_DISCARD: |
363 | case REQ_OP_SECURE_ERASE: | |
c55ddd90 | 364 | split = bio_split_discard(bio, lim, nr_segs, bs); |
7afafc8a | 365 | break; |
a6f0788e | 366 | case REQ_OP_WRITE_ZEROES: |
c55ddd90 | 367 | split = bio_split_write_zeroes(bio, lim, nr_segs, bs); |
a6f0788e | 368 | break; |
7afafc8a | 369 | default: |
c55ddd90 CH |
370 | split = bio_split_rw(bio, lim, nr_segs, bs, |
371 | get_max_io_size(bio, lim) << SECTOR_SHIFT); | |
613b1488 JA |
372 | if (IS_ERR(split)) |
373 | return NULL; | |
7afafc8a AH |
374 | break; |
375 | } | |
bdced438 | 376 | |
54efd50b | 377 | if (split) { |
613b1488 | 378 | /* there isn't chance to merge the split bio */ |
1eff9d32 | 379 | split->bi_opf |= REQ_NOMERGE; |
6ac45aeb | 380 | |
957a2b34 | 381 | blkcg_bio_issue_init(split); |
5a97806f CH |
382 | bio_chain(split, bio); |
383 | trace_block_split(split, bio->bi_iter.bi_sector); | |
384 | submit_bio_noacct(bio); | |
385 | return split; | |
54efd50b | 386 | } |
5a97806f | 387 | return bio; |
54efd50b | 388 | } |
14ccb66b | 389 | |
dad77584 | 390 | /** |
5a97806f CH |
391 | * bio_split_to_limits - split a bio to fit the queue limits |
392 | * @bio: bio to be split | |
393 | * | |
394 | * Check if @bio needs splitting based on the queue limits of @bio->bi_bdev, and | |
395 | * if so split off a bio fitting the limits from the beginning of @bio and | |
396 | * return it. @bio is shortened to the remainder and re-submitted. | |
dad77584 | 397 | * |
5a97806f CH |
398 | * The split bio is allocated from @q->bio_split, which is provided by the |
399 | * block layer. | |
dad77584 | 400 | */ |
5a97806f | 401 | struct bio *bio_split_to_limits(struct bio *bio) |
14ccb66b | 402 | { |
aa261f20 | 403 | const struct queue_limits *lim = &bdev_get_queue(bio->bi_bdev)->limits; |
14ccb66b CH |
404 | unsigned int nr_segs; |
405 | ||
c55ddd90 CH |
406 | if (bio_may_exceed_limits(bio, lim)) |
407 | return __bio_split_to_limits(bio, lim, &nr_segs); | |
5a97806f | 408 | return bio; |
14ccb66b | 409 | } |
5a97806f | 410 | EXPORT_SYMBOL(bio_split_to_limits); |
54efd50b | 411 | |
e9cd19c0 | 412 | unsigned int blk_recalc_rq_segments(struct request *rq) |
d6d48196 | 413 | { |
6869875f | 414 | unsigned int nr_phys_segs = 0; |
67927d22 | 415 | unsigned int bytes = 0; |
e9cd19c0 | 416 | struct req_iterator iter; |
6869875f | 417 | struct bio_vec bv; |
d6d48196 | 418 | |
e9cd19c0 | 419 | if (!rq->bio) |
1e428079 | 420 | return 0; |
d6d48196 | 421 | |
e9cd19c0 | 422 | switch (bio_op(rq->bio)) { |
a6f0788e CK |
423 | case REQ_OP_DISCARD: |
424 | case REQ_OP_SECURE_ERASE: | |
a958937f DJ |
425 | if (queue_max_discard_segments(rq->q) > 1) { |
426 | struct bio *bio = rq->bio; | |
427 | ||
428 | for_each_bio(bio) | |
429 | nr_phys_segs++; | |
430 | return nr_phys_segs; | |
431 | } | |
432 | return 1; | |
a6f0788e | 433 | case REQ_OP_WRITE_ZEROES: |
f9d03f96 | 434 | return 0; |
2d9b02be BVA |
435 | default: |
436 | break; | |
a6f0788e | 437 | } |
5cb8850c | 438 | |
e9cd19c0 | 439 | rq_for_each_bvec(bv, rq, iter) |
c55ddd90 | 440 | bvec_split_segs(&rq->q->limits, &bv, &nr_phys_segs, &bytes, |
708b25b3 | 441 | UINT_MAX, UINT_MAX); |
1e428079 JA |
442 | return nr_phys_segs; |
443 | } | |
444 | ||
48d7727c | 445 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
862e5a5e ML |
446 | struct scatterlist *sglist) |
447 | { | |
448 | if (!*sg) | |
449 | return sglist; | |
450 | ||
451 | /* | |
452 | * If the driver previously mapped a shorter list, we could see a | |
453 | * termination bit prematurely unless it fully inits the sg table | |
454 | * on each mapping. We KNOW that there must be more entries here | |
455 | * or the driver would be buggy, so force clear the termination bit | |
456 | * to avoid doing a full sg_init_table() in drivers for each command. | |
457 | */ | |
458 | sg_unmark_end(*sg); | |
459 | return sg_next(*sg); | |
460 | } | |
461 | ||
462 | static unsigned blk_bvec_map_sg(struct request_queue *q, | |
463 | struct bio_vec *bvec, struct scatterlist *sglist, | |
464 | struct scatterlist **sg) | |
465 | { | |
466 | unsigned nbytes = bvec->bv_len; | |
8a96a0e4 | 467 | unsigned nsegs = 0, total = 0; |
862e5a5e ML |
468 | |
469 | while (nbytes > 0) { | |
8a96a0e4 | 470 | unsigned offset = bvec->bv_offset + total; |
c55ddd90 CH |
471 | unsigned len = min(get_max_segment_size(&q->limits, |
472 | bvec->bv_page, offset), nbytes); | |
f9f76879 CH |
473 | struct page *page = bvec->bv_page; |
474 | ||
475 | /* | |
476 | * Unfortunately a fair number of drivers barf on scatterlists | |
477 | * that have an offset larger than PAGE_SIZE, despite other | |
478 | * subsystems dealing with that invariant just fine. For now | |
479 | * stick to the legacy format where we never present those from | |
480 | * the block layer, but the code below should be removed once | |
481 | * these offenders (mostly MMC/SD drivers) are fixed. | |
482 | */ | |
483 | page += (offset >> PAGE_SHIFT); | |
484 | offset &= ~PAGE_MASK; | |
862e5a5e ML |
485 | |
486 | *sg = blk_next_sg(sg, sglist); | |
f9f76879 | 487 | sg_set_page(*sg, page, len, offset); |
862e5a5e | 488 | |
8a96a0e4 CH |
489 | total += len; |
490 | nbytes -= len; | |
862e5a5e ML |
491 | nsegs++; |
492 | } | |
493 | ||
494 | return nsegs; | |
495 | } | |
496 | ||
16e3e418 ML |
497 | static inline int __blk_bvec_map_sg(struct bio_vec bv, |
498 | struct scatterlist *sglist, struct scatterlist **sg) | |
499 | { | |
500 | *sg = blk_next_sg(sg, sglist); | |
501 | sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset); | |
502 | return 1; | |
503 | } | |
504 | ||
f6970f83 ML |
505 | /* only try to merge bvecs into one sg if they are from two bios */ |
506 | static inline bool | |
507 | __blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec, | |
508 | struct bio_vec *bvprv, struct scatterlist **sg) | |
963ab9e5 AH |
509 | { |
510 | ||
511 | int nbytes = bvec->bv_len; | |
512 | ||
f6970f83 ML |
513 | if (!*sg) |
514 | return false; | |
963ab9e5 | 515 | |
f6970f83 ML |
516 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
517 | return false; | |
518 | ||
519 | if (!biovec_phys_mergeable(q, bvprv, bvec)) | |
520 | return false; | |
521 | ||
522 | (*sg)->length += nbytes; | |
523 | ||
524 | return true; | |
963ab9e5 AH |
525 | } |
526 | ||
5cb8850c KO |
527 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
528 | struct scatterlist *sglist, | |
529 | struct scatterlist **sg) | |
d6d48196 | 530 | { |
3f649ab7 | 531 | struct bio_vec bvec, bvprv = { NULL }; |
5cb8850c | 532 | struct bvec_iter iter; |
38417468 | 533 | int nsegs = 0; |
f6970f83 | 534 | bool new_bio = false; |
5cb8850c | 535 | |
f6970f83 ML |
536 | for_each_bio(bio) { |
537 | bio_for_each_bvec(bvec, bio, iter) { | |
538 | /* | |
539 | * Only try to merge bvecs from two bios given we | |
540 | * have done bio internal merge when adding pages | |
541 | * to bio | |
542 | */ | |
543 | if (new_bio && | |
544 | __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg)) | |
545 | goto next_bvec; | |
546 | ||
547 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) | |
548 | nsegs += __blk_bvec_map_sg(bvec, sglist, sg); | |
549 | else | |
550 | nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg); | |
551 | next_bvec: | |
552 | new_bio = false; | |
553 | } | |
b21e11c5 ML |
554 | if (likely(bio->bi_iter.bi_size)) { |
555 | bvprv = bvec; | |
556 | new_bio = true; | |
557 | } | |
f6970f83 | 558 | } |
d6d48196 | 559 | |
5cb8850c KO |
560 | return nsegs; |
561 | } | |
562 | ||
563 | /* | |
564 | * map a request to scatterlist, return number of sg entries setup. Caller | |
565 | * must make sure sg can hold rq->nr_phys_segments entries | |
566 | */ | |
89de1504 CH |
567 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
568 | struct scatterlist *sglist, struct scatterlist **last_sg) | |
5cb8850c | 569 | { |
5cb8850c KO |
570 | int nsegs = 0; |
571 | ||
f9d03f96 | 572 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
89de1504 | 573 | nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, last_sg); |
f9d03f96 | 574 | else if (rq->bio) |
89de1504 | 575 | nsegs = __blk_bios_map_sg(q, rq->bio, sglist, last_sg); |
f18573ab | 576 | |
89de1504 CH |
577 | if (*last_sg) |
578 | sg_mark_end(*last_sg); | |
d6d48196 | 579 | |
12e57f59 ML |
580 | /* |
581 | * Something must have been wrong if the figured number of | |
582 | * segment is bigger than number of req's physical segments | |
583 | */ | |
f9d03f96 | 584 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
12e57f59 | 585 | |
d6d48196 JA |
586 | return nsegs; |
587 | } | |
89de1504 | 588 | EXPORT_SYMBOL(__blk_rq_map_sg); |
d6d48196 | 589 | |
943b40c8 ML |
590 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
591 | { | |
592 | if (req_op(rq) == REQ_OP_DISCARD) | |
593 | return queue_max_discard_segments(rq->q); | |
594 | return queue_max_segments(rq->q); | |
595 | } | |
596 | ||
badf7f64 CH |
597 | static inline unsigned int blk_rq_get_max_sectors(struct request *rq, |
598 | sector_t offset) | |
599 | { | |
600 | struct request_queue *q = rq->q; | |
c8875190 | 601 | unsigned int max_sectors; |
badf7f64 CH |
602 | |
603 | if (blk_rq_is_passthrough(rq)) | |
604 | return q->limits.max_hw_sectors; | |
605 | ||
c8875190 | 606 | max_sectors = blk_queue_get_max_sectors(q, req_op(rq)); |
badf7f64 CH |
607 | if (!q->limits.chunk_sectors || |
608 | req_op(rq) == REQ_OP_DISCARD || | |
609 | req_op(rq) == REQ_OP_SECURE_ERASE) | |
c8875190 CH |
610 | return max_sectors; |
611 | return min(max_sectors, | |
612 | blk_chunk_sectors_left(offset, q->limits.chunk_sectors)); | |
badf7f64 CH |
613 | } |
614 | ||
14ccb66b CH |
615 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
616 | unsigned int nr_phys_segs) | |
d6d48196 | 617 | { |
6b2b0459 TH |
618 | if (!blk_cgroup_mergeable(req, bio)) |
619 | goto no_merge; | |
620 | ||
2705dfb2 | 621 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
13f05c8d MP |
622 | goto no_merge; |
623 | ||
2705dfb2 ML |
624 | /* discard request merge won't add new segment */ |
625 | if (req_op(req) == REQ_OP_DISCARD) | |
626 | return 1; | |
627 | ||
628 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(req)) | |
13f05c8d | 629 | goto no_merge; |
d6d48196 JA |
630 | |
631 | /* | |
632 | * This will form the start of a new hw segment. Bump both | |
633 | * counters. | |
634 | */ | |
d6d48196 JA |
635 | req->nr_phys_segments += nr_phys_segs; |
636 | return 1; | |
13f05c8d MP |
637 | |
638 | no_merge: | |
14ccb66b | 639 | req_set_nomerge(req->q, req); |
13f05c8d | 640 | return 0; |
d6d48196 JA |
641 | } |
642 | ||
14ccb66b | 643 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
d6d48196 | 644 | { |
5e7c4274 JA |
645 | if (req_gap_back_merge(req, bio)) |
646 | return 0; | |
7f39add3 SG |
647 | if (blk_integrity_rq(req) && |
648 | integrity_req_gap_back_merge(req, bio)) | |
649 | return 0; | |
a892c8d5 ST |
650 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
651 | return 0; | |
f31dc1cd | 652 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 653 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) { |
14ccb66b | 654 | req_set_nomerge(req->q, req); |
d6d48196 JA |
655 | return 0; |
656 | } | |
d6d48196 | 657 | |
14ccb66b | 658 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
659 | } |
660 | ||
eda5cc99 CH |
661 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
662 | unsigned int nr_segs) | |
d6d48196 | 663 | { |
5e7c4274 JA |
664 | if (req_gap_front_merge(req, bio)) |
665 | return 0; | |
7f39add3 SG |
666 | if (blk_integrity_rq(req) && |
667 | integrity_req_gap_front_merge(req, bio)) | |
668 | return 0; | |
a892c8d5 ST |
669 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
670 | return 0; | |
f31dc1cd | 671 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
17007f39 | 672 | blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) { |
14ccb66b | 673 | req_set_nomerge(req->q, req); |
d6d48196 JA |
674 | return 0; |
675 | } | |
d6d48196 | 676 | |
14ccb66b | 677 | return ll_new_hw_segment(req, bio, nr_segs); |
d6d48196 JA |
678 | } |
679 | ||
445251d0 JA |
680 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
681 | struct request *next) | |
682 | { | |
683 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
684 | ||
685 | if (segments >= queue_max_discard_segments(q)) | |
686 | goto no_merge; | |
687 | if (blk_rq_sectors(req) + bio_sectors(next->bio) > | |
688 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
689 | goto no_merge; | |
690 | ||
691 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next); | |
692 | return true; | |
693 | no_merge: | |
694 | req_set_nomerge(q, req); | |
695 | return false; | |
696 | } | |
697 | ||
d6d48196 JA |
698 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
699 | struct request *next) | |
700 | { | |
701 | int total_phys_segments; | |
d6d48196 | 702 | |
5e7c4274 | 703 | if (req_gap_back_merge(req, next->bio)) |
854fbb9c KB |
704 | return 0; |
705 | ||
d6d48196 JA |
706 | /* |
707 | * Will it become too large? | |
708 | */ | |
f31dc1cd | 709 | if ((blk_rq_sectors(req) + blk_rq_sectors(next)) > |
17007f39 | 710 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
d6d48196 JA |
711 | return 0; |
712 | ||
713 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | |
943b40c8 | 714 | if (total_phys_segments > blk_rq_get_max_segments(req)) |
d6d48196 JA |
715 | return 0; |
716 | ||
6b2b0459 TH |
717 | if (!blk_cgroup_mergeable(req, next->bio)) |
718 | return 0; | |
719 | ||
4eaf99be | 720 | if (blk_integrity_merge_rq(q, req, next) == false) |
13f05c8d MP |
721 | return 0; |
722 | ||
a892c8d5 ST |
723 | if (!bio_crypt_ctx_merge_rq(req, next)) |
724 | return 0; | |
725 | ||
d6d48196 JA |
726 | /* Merge is OK... */ |
727 | req->nr_phys_segments = total_phys_segments; | |
d6d48196 JA |
728 | return 1; |
729 | } | |
730 | ||
80a761fd TH |
731 | /** |
732 | * blk_rq_set_mixed_merge - mark a request as mixed merge | |
733 | * @rq: request to mark as mixed merge | |
734 | * | |
735 | * Description: | |
736 | * @rq is about to be mixed merged. Make sure the attributes | |
737 | * which can be mixed are set in each bio and mark @rq as mixed | |
738 | * merged. | |
739 | */ | |
740 | void blk_rq_set_mixed_merge(struct request *rq) | |
741 | { | |
16458cf3 | 742 | blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
80a761fd TH |
743 | struct bio *bio; |
744 | ||
e8064021 | 745 | if (rq->rq_flags & RQF_MIXED_MERGE) |
80a761fd TH |
746 | return; |
747 | ||
748 | /* | |
749 | * @rq will no longer represent mixable attributes for all the | |
750 | * contained bios. It will just track those of the first one. | |
751 | * Distributes the attributs to each bio. | |
752 | */ | |
753 | for (bio = rq->bio; bio; bio = bio->bi_next) { | |
1eff9d32 JA |
754 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
755 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); | |
756 | bio->bi_opf |= ff; | |
80a761fd | 757 | } |
e8064021 | 758 | rq->rq_flags |= RQF_MIXED_MERGE; |
80a761fd TH |
759 | } |
760 | ||
b9c54f56 | 761 | static void blk_account_io_merge_request(struct request *req) |
26308eab JM |
762 | { |
763 | if (blk_do_io_stat(req)) { | |
112f158f | 764 | part_stat_lock(); |
b9c54f56 | 765 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
26308eab JM |
766 | part_stat_unlock(); |
767 | } | |
768 | } | |
b9c54f56 | 769 | |
e96c0d83 EB |
770 | static enum elv_merge blk_try_req_merge(struct request *req, |
771 | struct request *next) | |
69840466 JW |
772 | { |
773 | if (blk_discard_mergable(req)) | |
774 | return ELEVATOR_DISCARD_MERGE; | |
775 | else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next)) | |
776 | return ELEVATOR_BACK_MERGE; | |
777 | ||
778 | return ELEVATOR_NO_MERGE; | |
779 | } | |
26308eab | 780 | |
d6d48196 | 781 | /* |
b973cb7e JA |
782 | * For non-mq, this has to be called with the request spinlock acquired. |
783 | * For mq with scheduling, the appropriate queue wide lock should be held. | |
d6d48196 | 784 | */ |
b973cb7e JA |
785 | static struct request *attempt_merge(struct request_queue *q, |
786 | struct request *req, struct request *next) | |
d6d48196 JA |
787 | { |
788 | if (!rq_mergeable(req) || !rq_mergeable(next)) | |
b973cb7e | 789 | return NULL; |
d6d48196 | 790 | |
288dab8a | 791 | if (req_op(req) != req_op(next)) |
b973cb7e | 792 | return NULL; |
f31dc1cd | 793 | |
79bb1dbd | 794 | if (rq_data_dir(req) != rq_data_dir(next)) |
b973cb7e | 795 | return NULL; |
d6d48196 | 796 | |
668ffc03 DLM |
797 | if (req->ioprio != next->ioprio) |
798 | return NULL; | |
799 | ||
d6d48196 JA |
800 | /* |
801 | * If we are allowed to merge, then append bio list | |
802 | * from next to rq and release next. merge_requests_fn | |
803 | * will have updated segment counts, update sector | |
445251d0 JA |
804 | * counts here. Handle DISCARDs separately, as they |
805 | * have separate settings. | |
d6d48196 | 806 | */ |
69840466 JW |
807 | |
808 | switch (blk_try_req_merge(req, next)) { | |
809 | case ELEVATOR_DISCARD_MERGE: | |
445251d0 JA |
810 | if (!req_attempt_discard_merge(q, req, next)) |
811 | return NULL; | |
69840466 JW |
812 | break; |
813 | case ELEVATOR_BACK_MERGE: | |
814 | if (!ll_merge_requests_fn(q, req, next)) | |
815 | return NULL; | |
816 | break; | |
817 | default: | |
b973cb7e | 818 | return NULL; |
69840466 | 819 | } |
d6d48196 | 820 | |
80a761fd TH |
821 | /* |
822 | * If failfast settings disagree or any of the two is already | |
823 | * a mixed merge, mark both as mixed before proceeding. This | |
824 | * makes sure that all involved bios have mixable attributes | |
825 | * set properly. | |
826 | */ | |
e8064021 | 827 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
80a761fd TH |
828 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
829 | (next->cmd_flags & REQ_FAILFAST_MASK)) { | |
830 | blk_rq_set_mixed_merge(req); | |
831 | blk_rq_set_mixed_merge(next); | |
832 | } | |
833 | ||
d6d48196 | 834 | /* |
522a7775 OS |
835 | * At this point we have either done a back merge or front merge. We |
836 | * need the smaller start_time_ns of the merged requests to be the | |
837 | * current request for accounting purposes. | |
d6d48196 | 838 | */ |
522a7775 OS |
839 | if (next->start_time_ns < req->start_time_ns) |
840 | req->start_time_ns = next->start_time_ns; | |
d6d48196 JA |
841 | |
842 | req->biotail->bi_next = next->bio; | |
843 | req->biotail = next->biotail; | |
844 | ||
a2dec7b3 | 845 | req->__data_len += blk_rq_bytes(next); |
d6d48196 | 846 | |
2a5cf35c | 847 | if (!blk_discard_mergable(req)) |
445251d0 | 848 | elv_merge_requests(q, req, next); |
d6d48196 | 849 | |
42dad764 JM |
850 | /* |
851 | * 'next' is going away, so update stats accordingly | |
852 | */ | |
b9c54f56 | 853 | blk_account_io_merge_request(next); |
d6d48196 | 854 | |
a54895fa | 855 | trace_block_rq_merge(next); |
f3bdc62f | 856 | |
e4d750c9 JA |
857 | /* |
858 | * ownership of bio passed from next to req, return 'next' for | |
859 | * the caller to free | |
860 | */ | |
1cd96c24 | 861 | next->bio = NULL; |
b973cb7e | 862 | return next; |
d6d48196 JA |
863 | } |
864 | ||
eda5cc99 CH |
865 | static struct request *attempt_back_merge(struct request_queue *q, |
866 | struct request *rq) | |
d6d48196 JA |
867 | { |
868 | struct request *next = elv_latter_request(q, rq); | |
869 | ||
870 | if (next) | |
871 | return attempt_merge(q, rq, next); | |
872 | ||
b973cb7e | 873 | return NULL; |
d6d48196 JA |
874 | } |
875 | ||
eda5cc99 CH |
876 | static struct request *attempt_front_merge(struct request_queue *q, |
877 | struct request *rq) | |
d6d48196 JA |
878 | { |
879 | struct request *prev = elv_former_request(q, rq); | |
880 | ||
881 | if (prev) | |
882 | return attempt_merge(q, prev, rq); | |
883 | ||
b973cb7e | 884 | return NULL; |
d6d48196 | 885 | } |
5e84ea3a | 886 | |
fd2ef39c JK |
887 | /* |
888 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false | |
889 | * otherwise. The caller is responsible for freeing 'next' if the merge | |
890 | * happened. | |
891 | */ | |
892 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, | |
893 | struct request *next) | |
5e84ea3a | 894 | { |
fd2ef39c | 895 | return attempt_merge(q, rq, next); |
5e84ea3a | 896 | } |
050c8ea8 TH |
897 | |
898 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) | |
899 | { | |
e2a60da7 | 900 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
050c8ea8 TH |
901 | return false; |
902 | ||
288dab8a | 903 | if (req_op(rq) != bio_op(bio)) |
f31dc1cd MP |
904 | return false; |
905 | ||
050c8ea8 TH |
906 | /* different data direction or already started, don't merge */ |
907 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
908 | return false; | |
909 | ||
6b2b0459 TH |
910 | /* don't merge across cgroup boundaries */ |
911 | if (!blk_cgroup_mergeable(rq, bio)) | |
912 | return false; | |
913 | ||
050c8ea8 | 914 | /* only merge integrity protected bio into ditto rq */ |
4eaf99be | 915 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
050c8ea8 TH |
916 | return false; |
917 | ||
a892c8d5 ST |
918 | /* Only merge if the crypt contexts are compatible */ |
919 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) | |
920 | return false; | |
921 | ||
668ffc03 DLM |
922 | if (rq->ioprio != bio_prio(bio)) |
923 | return false; | |
924 | ||
050c8ea8 TH |
925 | return true; |
926 | } | |
927 | ||
34fe7c05 | 928 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
050c8ea8 | 929 | { |
69840466 | 930 | if (blk_discard_mergable(rq)) |
1e739730 CH |
931 | return ELEVATOR_DISCARD_MERGE; |
932 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) | |
050c8ea8 | 933 | return ELEVATOR_BACK_MERGE; |
4f024f37 | 934 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
050c8ea8 TH |
935 | return ELEVATOR_FRONT_MERGE; |
936 | return ELEVATOR_NO_MERGE; | |
937 | } | |
8e756373 BW |
938 | |
939 | static void blk_account_io_merge_bio(struct request *req) | |
940 | { | |
941 | if (!blk_do_io_stat(req)) | |
942 | return; | |
943 | ||
944 | part_stat_lock(); | |
945 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); | |
946 | part_stat_unlock(); | |
947 | } | |
948 | ||
eda5cc99 CH |
949 | enum bio_merge_status { |
950 | BIO_MERGE_OK, | |
951 | BIO_MERGE_NONE, | |
952 | BIO_MERGE_FAILED, | |
953 | }; | |
954 | ||
955 | static enum bio_merge_status bio_attempt_back_merge(struct request *req, | |
956 | struct bio *bio, unsigned int nr_segs) | |
8e756373 | 957 | { |
16458cf3 | 958 | const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; |
8e756373 BW |
959 | |
960 | if (!ll_back_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 961 | return BIO_MERGE_FAILED; |
8e756373 | 962 | |
e8a676d6 | 963 | trace_block_bio_backmerge(bio); |
8e756373 BW |
964 | rq_qos_merge(req->q, req, bio); |
965 | ||
966 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
967 | blk_rq_set_mixed_merge(req); | |
968 | ||
969 | req->biotail->bi_next = bio; | |
970 | req->biotail = bio; | |
971 | req->__data_len += bio->bi_iter.bi_size; | |
972 | ||
973 | bio_crypt_free_ctx(bio); | |
974 | ||
975 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 976 | return BIO_MERGE_OK; |
8e756373 BW |
977 | } |
978 | ||
eda5cc99 CH |
979 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
980 | struct bio *bio, unsigned int nr_segs) | |
8e756373 | 981 | { |
16458cf3 | 982 | const blk_opf_t ff = bio->bi_opf & REQ_FAILFAST_MASK; |
8e756373 BW |
983 | |
984 | if (!ll_front_merge_fn(req, bio, nr_segs)) | |
7d7ca7c5 | 985 | return BIO_MERGE_FAILED; |
8e756373 | 986 | |
e8a676d6 | 987 | trace_block_bio_frontmerge(bio); |
8e756373 BW |
988 | rq_qos_merge(req->q, req, bio); |
989 | ||
990 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) | |
991 | blk_rq_set_mixed_merge(req); | |
992 | ||
993 | bio->bi_next = req->bio; | |
994 | req->bio = bio; | |
995 | ||
996 | req->__sector = bio->bi_iter.bi_sector; | |
997 | req->__data_len += bio->bi_iter.bi_size; | |
998 | ||
999 | bio_crypt_do_front_merge(req, 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_discard_merge(struct request_queue *q, |
1006 | struct request *req, struct bio *bio) | |
8e756373 BW |
1007 | { |
1008 | unsigned short segments = blk_rq_nr_discard_segments(req); | |
1009 | ||
1010 | if (segments >= queue_max_discard_segments(q)) | |
1011 | goto no_merge; | |
1012 | if (blk_rq_sectors(req) + bio_sectors(bio) > | |
1013 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) | |
1014 | goto no_merge; | |
1015 | ||
1016 | rq_qos_merge(q, req, bio); | |
1017 | ||
1018 | req->biotail->bi_next = bio; | |
1019 | req->biotail = bio; | |
1020 | req->__data_len += bio->bi_iter.bi_size; | |
1021 | req->nr_phys_segments = segments + 1; | |
1022 | ||
1023 | blk_account_io_merge_bio(req); | |
7d7ca7c5 | 1024 | return BIO_MERGE_OK; |
8e756373 BW |
1025 | no_merge: |
1026 | req_set_nomerge(q, req); | |
7d7ca7c5 BW |
1027 | return BIO_MERGE_FAILED; |
1028 | } | |
1029 | ||
1030 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, | |
1031 | struct request *rq, | |
1032 | struct bio *bio, | |
1033 | unsigned int nr_segs, | |
1034 | bool sched_allow_merge) | |
1035 | { | |
1036 | if (!blk_rq_merge_ok(rq, bio)) | |
1037 | return BIO_MERGE_NONE; | |
1038 | ||
1039 | switch (blk_try_merge(rq, bio)) { | |
1040 | case ELEVATOR_BACK_MERGE: | |
265600b7 | 1041 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1042 | return bio_attempt_back_merge(rq, bio, nr_segs); |
1043 | break; | |
1044 | case ELEVATOR_FRONT_MERGE: | |
265600b7 | 1045 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
7d7ca7c5 BW |
1046 | return bio_attempt_front_merge(rq, bio, nr_segs); |
1047 | break; | |
1048 | case ELEVATOR_DISCARD_MERGE: | |
1049 | return bio_attempt_discard_merge(q, rq, bio); | |
1050 | default: | |
1051 | return BIO_MERGE_NONE; | |
1052 | } | |
1053 | ||
1054 | return BIO_MERGE_FAILED; | |
8e756373 BW |
1055 | } |
1056 | ||
1057 | /** | |
1058 | * blk_attempt_plug_merge - try to merge with %current's plugged list | |
1059 | * @q: request_queue new bio is being queued at | |
1060 | * @bio: new bio being queued | |
1061 | * @nr_segs: number of segments in @bio | |
87c037d1 | 1062 | * from the passed in @q already in the plug list |
8e756373 | 1063 | * |
d38a9c04 JA |
1064 | * Determine whether @bio being queued on @q can be merged with the previous |
1065 | * request on %current's plugged list. Returns %true if merge was successful, | |
8e756373 BW |
1066 | * otherwise %false. |
1067 | * | |
1068 | * Plugging coalesces IOs from the same issuer for the same purpose without | |
1069 | * going through @q->queue_lock. As such it's more of an issuing mechanism | |
1070 | * than scheduling, and the request, while may have elvpriv data, is not | |
1071 | * added on the elevator at this point. In addition, we don't have | |
1072 | * reliable access to the elevator outside queue lock. Only check basic | |
1073 | * merging parameters without querying the elevator. | |
1074 | * | |
1075 | * Caller must ensure !blk_queue_nomerges(q) beforehand. | |
1076 | */ | |
1077 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, | |
0c5bcc92 | 1078 | unsigned int nr_segs) |
8e756373 BW |
1079 | { |
1080 | struct blk_plug *plug; | |
1081 | struct request *rq; | |
8e756373 | 1082 | |
6deacb3b | 1083 | plug = blk_mq_plug(bio); |
bc490f81 | 1084 | if (!plug || rq_list_empty(plug->mq_list)) |
8e756373 BW |
1085 | return false; |
1086 | ||
5b205071 JA |
1087 | rq_list_for_each(&plug->mq_list, rq) { |
1088 | if (rq->q == q) { | |
1089 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == | |
1090 | BIO_MERGE_OK) | |
1091 | return true; | |
1092 | break; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Only keep iterating plug list for merges if we have multiple | |
1097 | * queues | |
1098 | */ | |
1099 | if (!plug->multiple_queues) | |
1100 | break; | |
8e756373 | 1101 | } |
8e756373 BW |
1102 | return false; |
1103 | } | |
bdc6a287 BW |
1104 | |
1105 | /* | |
1106 | * Iterate list of requests and see if we can merge this bio with any | |
1107 | * of them. | |
1108 | */ | |
1109 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, | |
1110 | struct bio *bio, unsigned int nr_segs) | |
1111 | { | |
1112 | struct request *rq; | |
1113 | int checked = 8; | |
1114 | ||
1115 | list_for_each_entry_reverse(rq, list, queuelist) { | |
bdc6a287 BW |
1116 | if (!checked--) |
1117 | break; | |
1118 | ||
7d7ca7c5 BW |
1119 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, true)) { |
1120 | case BIO_MERGE_NONE: | |
bdc6a287 | 1121 | continue; |
7d7ca7c5 BW |
1122 | case BIO_MERGE_OK: |
1123 | return true; | |
1124 | case BIO_MERGE_FAILED: | |
1125 | return false; | |
bdc6a287 BW |
1126 | } |
1127 | ||
bdc6a287 BW |
1128 | } |
1129 | ||
1130 | return false; | |
1131 | } | |
1132 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); | |
eda5cc99 CH |
1133 | |
1134 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, | |
1135 | unsigned int nr_segs, struct request **merged_request) | |
1136 | { | |
1137 | struct request *rq; | |
1138 | ||
1139 | switch (elv_merge(q, &rq, bio)) { | |
1140 | case ELEVATOR_BACK_MERGE: | |
1141 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1142 | return false; | |
1143 | if (bio_attempt_back_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1144 | return false; | |
1145 | *merged_request = attempt_back_merge(q, rq); | |
1146 | if (!*merged_request) | |
1147 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
1148 | return true; | |
1149 | case ELEVATOR_FRONT_MERGE: | |
1150 | if (!blk_mq_sched_allow_merge(q, rq, bio)) | |
1151 | return false; | |
1152 | if (bio_attempt_front_merge(rq, bio, nr_segs) != BIO_MERGE_OK) | |
1153 | return false; | |
1154 | *merged_request = attempt_front_merge(q, rq); | |
1155 | if (!*merged_request) | |
1156 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
1157 | return true; | |
1158 | case ELEVATOR_DISCARD_MERGE: | |
1159 | return bio_attempt_discard_merge(q, rq, bio) == BIO_MERGE_OK; | |
1160 | default: | |
1161 | return false; | |
1162 | } | |
1163 | } | |
1164 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |