1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Functions related to segment and merge handling |
4 | */ |
5 | #include <linux/kernel.h> |
6 | #include <linux/module.h> |
7 | #include <linux/bio.h> |
8 | #include <linux/blkdev.h> |
9 | #include <linux/blk-integrity.h> |
10 | #include <linux/scatterlist.h> |
11 | #include <linux/part_stat.h> |
12 | #include <linux/blk-cgroup.h> |
13 | |
14 | #include <trace/events/block.h> |
15 | |
16 | #include "blk.h" |
17 | #include "blk-mq-sched.h" |
18 | #include "blk-rq-qos.h" |
19 | #include "blk-throttle.h" |
20 | |
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, bytes: 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 | |
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(bio: 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(bio: prev_rq->bio, bv: &pb); |
67 | else |
68 | bio_get_first_bvec(bio: prev, bv: &pb); |
69 | if (pb.bv_offset & queue_virt_boundary(q)) |
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(bio: prev, bv: &pb); |
82 | bio_get_first_bvec(bio: next, bv: &nb); |
83 | if (biovec_phys_mergeable(q, vec1: &pb, vec2: &nb)) |
84 | return false; |
85 | return __bvec_gap_to_prev(lim: &q->limits, bprv: &pb, offset: nb.bv_offset); |
86 | } |
87 | |
88 | static inline bool req_gap_back_merge(struct request *req, struct bio *bio) |
89 | { |
90 | return bio_will_gap(q: req->q, prev_rq: req, prev: req->biotail, next: bio); |
91 | } |
92 | |
93 | static inline bool req_gap_front_merge(struct request *req, struct bio *bio) |
94 | { |
95 | return bio_will_gap(q: req->q, NULL, prev: bio, next: req->bio); |
96 | } |
97 | |
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 | */ |
103 | static unsigned int bio_allowed_max_sectors(const struct queue_limits *lim) |
104 | { |
105 | return round_down(UINT_MAX, lim->logical_block_size) >> SECTOR_SHIFT; |
106 | } |
107 | |
108 | static struct bio *bio_split_discard(struct bio *bio, |
109 | const struct queue_limits *lim, |
110 | unsigned *nsegs, struct bio_set *bs) |
111 | { |
112 | unsigned int max_discard_sectors, granularity; |
113 | sector_t tmp; |
114 | unsigned split_sectors; |
115 | |
116 | *nsegs = 1; |
117 | |
118 | /* Zero-sector (unknown) and one-sector granularities are the same. */ |
119 | granularity = max(lim->discard_granularity >> 9, 1U); |
120 | |
121 | max_discard_sectors = |
122 | min(lim->max_discard_sectors, bio_allowed_max_sectors(lim)); |
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 | */ |
139 | tmp = bio->bi_iter.bi_sector + split_sectors - |
140 | ((lim->discard_alignment >> 9) % granularity); |
141 | tmp = sector_div(tmp, granularity); |
142 | |
143 | if (split_sectors > tmp) |
144 | split_sectors -= tmp; |
145 | |
146 | return bio_split(bio, sectors: split_sectors, GFP_NOIO, bs); |
147 | } |
148 | |
149 | static struct bio *bio_split_write_zeroes(struct bio *bio, |
150 | const struct queue_limits *lim, |
151 | unsigned *nsegs, struct bio_set *bs) |
152 | { |
153 | *nsegs = 0; |
154 | if (!lim->max_write_zeroes_sectors) |
155 | return NULL; |
156 | if (bio_sectors(bio) <= lim->max_write_zeroes_sectors) |
157 | return NULL; |
158 | return bio_split(bio, sectors: lim->max_write_zeroes_sectors, GFP_NOIO, bs); |
159 | } |
160 | |
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 | */ |
169 | static inline unsigned get_max_io_size(struct bio *bio, |
170 | const struct queue_limits *lim) |
171 | { |
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; |
175 | |
176 | if (lim->chunk_sectors) { |
177 | max_sectors = min(max_sectors, |
178 | blk_chunk_sectors_left(bio->bi_iter.bi_sector, |
179 | lim->chunk_sectors)); |
180 | } |
181 | |
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); |
187 | } |
188 | |
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 | */ |
197 | static inline unsigned get_max_segment_size(const struct queue_limits *lim, |
198 | struct page *start_page, unsigned long offset) |
199 | { |
200 | unsigned long mask = lim->seg_boundary_mask; |
201 | |
202 | offset = mask & (page_to_phys(start_page) + offset); |
203 | |
204 | /* |
205 | * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1 |
206 | * after having calculated the minimum. |
207 | */ |
208 | return min(mask - offset, (unsigned long)lim->max_segment_size - 1) + 1; |
209 | } |
210 | |
211 | /** |
212 | * bvec_split_segs - verify whether or not a bvec should be split in the middle |
213 | * @lim: [in] queue limits to split based on |
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 |
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 |
221 | * @max_segs: [in] upper bound for *@nsegs |
222 | * @max_bytes: [in] upper bound for *@bytes |
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. |
230 | */ |
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) |
234 | { |
235 | unsigned max_len = min(max_bytes, UINT_MAX) - *bytes; |
236 | unsigned len = min(bv->bv_len, max_len); |
237 | unsigned total_len = 0; |
238 | unsigned seg_size = 0; |
239 | |
240 | while (len && *nsegs < max_segs) { |
241 | seg_size = get_max_segment_size(lim, start_page: bv->bv_page, |
242 | offset: bv->bv_offset + total_len); |
243 | seg_size = min(seg_size, len); |
244 | |
245 | (*nsegs)++; |
246 | total_len += seg_size; |
247 | len -= seg_size; |
248 | |
249 | if ((bv->bv_offset + total_len) & lim->virt_boundary_mask) |
250 | break; |
251 | } |
252 | |
253 | *bytes += total_len; |
254 | |
255 | /* tell the caller to split the bvec if it is too big to fit */ |
256 | return len > 0 || bv->bv_len > max_len; |
257 | } |
258 | |
259 | /** |
260 | * bio_split_rw - split a bio in two bios |
261 | * @bio: [in] bio to be split |
262 | * @lim: [in] queue limits to split based on |
263 | * @segs: [out] number of segments in the bio with the first half of the sectors |
264 | * @bs: [in] bio set to allocate the clone from |
265 | * @max_bytes: [in] maximum number of bytes per bio |
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: |
270 | * - That it has at most @max_bytes worth of data |
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 | */ |
279 | struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim, |
280 | unsigned *segs, struct bio_set *bs, unsigned max_bytes) |
281 | { |
282 | struct bio_vec bv, bvprv, *bvprvp = NULL; |
283 | struct bvec_iter iter; |
284 | unsigned nsegs = 0, bytes = 0; |
285 | |
286 | bio_for_each_bvec(bv, bio, iter) { |
287 | /* |
288 | * If the queue doesn't support SG gaps and adding this |
289 | * offset would create a gap, disallow it. |
290 | */ |
291 | if (bvprvp && bvec_gap_to_prev(lim, bprv: bvprvp, offset: bv.bv_offset)) |
292 | goto split; |
293 | |
294 | if (nsegs < lim->max_segments && |
295 | bytes + bv.bv_len <= max_bytes && |
296 | bv.bv_offset + bv.bv_len <= PAGE_SIZE) { |
297 | nsegs++; |
298 | bytes += bv.bv_len; |
299 | } else { |
300 | if (bvec_split_segs(lim, bv: &bv, nsegs: &nsegs, bytes: &bytes, |
301 | max_segs: lim->max_segments, max_bytes)) |
302 | goto split; |
303 | } |
304 | |
305 | bvprv = bv; |
306 | bvprvp = &bvprv; |
307 | } |
308 | |
309 | *segs = nsegs; |
310 | return NULL; |
311 | split: |
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(error: -EAGAIN); |
320 | } |
321 | |
322 | *segs = nsegs; |
323 | |
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 | */ |
329 | bytes = ALIGN_DOWN(bytes, lim->logical_block_size); |
330 | |
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 | */ |
336 | bio_clear_polled(bio); |
337 | return bio_split(bio, sectors: bytes >> SECTOR_SHIFT, GFP_NOIO, bs); |
338 | } |
339 | EXPORT_SYMBOL_GPL(bio_split_rw); |
340 | |
341 | /** |
342 | * __bio_split_to_limits - split a bio to fit the queue limits |
343 | * @bio: bio to be split |
344 | * @lim: queue limits to split based on |
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. |
350 | * |
351 | * The split bio is allocated from @q->bio_split, which is provided by the |
352 | * block layer. |
353 | */ |
354 | struct bio *__bio_split_to_limits(struct bio *bio, |
355 | const struct queue_limits *lim, |
356 | unsigned int *nr_segs) |
357 | { |
358 | struct bio_set *bs = &bio->bi_bdev->bd_disk->bio_split; |
359 | struct bio *split; |
360 | |
361 | switch (bio_op(bio)) { |
362 | case REQ_OP_DISCARD: |
363 | case REQ_OP_SECURE_ERASE: |
364 | split = bio_split_discard(bio, lim, nsegs: nr_segs, bs); |
365 | break; |
366 | case REQ_OP_WRITE_ZEROES: |
367 | split = bio_split_write_zeroes(bio, lim, nsegs: nr_segs, bs); |
368 | break; |
369 | default: |
370 | split = bio_split_rw(bio, lim, nr_segs, bs, |
371 | get_max_io_size(bio, lim) << SECTOR_SHIFT); |
372 | if (IS_ERR(ptr: split)) |
373 | return NULL; |
374 | break; |
375 | } |
376 | |
377 | if (split) { |
378 | /* there isn't chance to merge the split bio */ |
379 | split->bi_opf |= REQ_NOMERGE; |
380 | |
381 | blkcg_bio_issue_init(bio: split); |
382 | bio_chain(split, bio); |
383 | trace_block_split(bio: split, new_sector: bio->bi_iter.bi_sector); |
384 | submit_bio_noacct(bio); |
385 | return split; |
386 | } |
387 | return bio; |
388 | } |
389 | |
390 | /** |
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. |
397 | * |
398 | * The split bio is allocated from @q->bio_split, which is provided by the |
399 | * block layer. |
400 | */ |
401 | struct bio *bio_split_to_limits(struct bio *bio) |
402 | { |
403 | const struct queue_limits *lim = &bdev_get_queue(bdev: bio->bi_bdev)->limits; |
404 | unsigned int nr_segs; |
405 | |
406 | if (bio_may_exceed_limits(bio, lim)) |
407 | return __bio_split_to_limits(bio, lim, nr_segs: &nr_segs); |
408 | return bio; |
409 | } |
410 | EXPORT_SYMBOL(bio_split_to_limits); |
411 | |
412 | unsigned int blk_recalc_rq_segments(struct request *rq) |
413 | { |
414 | unsigned int nr_phys_segs = 0; |
415 | unsigned int bytes = 0; |
416 | struct req_iterator iter; |
417 | struct bio_vec bv; |
418 | |
419 | if (!rq->bio) |
420 | return 0; |
421 | |
422 | switch (bio_op(bio: rq->bio)) { |
423 | case REQ_OP_DISCARD: |
424 | case REQ_OP_SECURE_ERASE: |
425 | if (queue_max_discard_segments(q: 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; |
433 | case REQ_OP_WRITE_ZEROES: |
434 | return 0; |
435 | default: |
436 | break; |
437 | } |
438 | |
439 | rq_for_each_bvec(bv, rq, iter) |
440 | bvec_split_segs(lim: &rq->q->limits, bv: &bv, nsegs: &nr_phys_segs, bytes: &bytes, |
441 | UINT_MAX, UINT_MAX); |
442 | return nr_phys_segs; |
443 | } |
444 | |
445 | static inline struct scatterlist *blk_next_sg(struct scatterlist **sg, |
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: *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; |
467 | unsigned nsegs = 0, total = 0; |
468 | |
469 | while (nbytes > 0) { |
470 | unsigned offset = bvec->bv_offset + total; |
471 | unsigned len = min(get_max_segment_size(&q->limits, |
472 | bvec->bv_page, offset), nbytes); |
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; |
485 | |
486 | *sg = blk_next_sg(sg, sglist); |
487 | sg_set_page(sg: *sg, page, len, offset); |
488 | |
489 | total += len; |
490 | nbytes -= len; |
491 | nsegs++; |
492 | } |
493 | |
494 | return nsegs; |
495 | } |
496 | |
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: *sg, page: bv.bv_page, len: bv.bv_len, offset: bv.bv_offset); |
502 | return 1; |
503 | } |
504 | |
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) |
509 | { |
510 | |
511 | int nbytes = bvec->bv_len; |
512 | |
513 | if (!*sg) |
514 | return false; |
515 | |
516 | if ((*sg)->length + nbytes > queue_max_segment_size(q)) |
517 | return false; |
518 | |
519 | if (!biovec_phys_mergeable(q, vec1: bvprv, vec2: bvec)) |
520 | return false; |
521 | |
522 | (*sg)->length += nbytes; |
523 | |
524 | return true; |
525 | } |
526 | |
527 | static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio, |
528 | struct scatterlist *sglist, |
529 | struct scatterlist **sg) |
530 | { |
531 | struct bio_vec bvec, bvprv = { NULL }; |
532 | struct bvec_iter iter; |
533 | int nsegs = 0; |
534 | bool new_bio = false; |
535 | |
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: &bvec, bvprv: &bvprv, sg)) |
545 | goto next_bvec; |
546 | |
547 | if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE) |
548 | nsegs += __blk_bvec_map_sg(bv: bvec, sglist, sg); |
549 | else |
550 | nsegs += blk_bvec_map_sg(q, bvec: &bvec, sglist, sg); |
551 | next_bvec: |
552 | new_bio = false; |
553 | } |
554 | if (likely(bio->bi_iter.bi_size)) { |
555 | bvprv = bvec; |
556 | new_bio = true; |
557 | } |
558 | } |
559 | |
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 | */ |
567 | int __blk_rq_map_sg(struct request_queue *q, struct request *rq, |
568 | struct scatterlist *sglist, struct scatterlist **last_sg) |
569 | { |
570 | int nsegs = 0; |
571 | |
572 | if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) |
573 | nsegs = __blk_bvec_map_sg(bv: rq->special_vec, sglist, sg: last_sg); |
574 | else if (rq->bio) |
575 | nsegs = __blk_bios_map_sg(q, bio: rq->bio, sglist, sg: last_sg); |
576 | |
577 | if (*last_sg) |
578 | sg_mark_end(sg: *last_sg); |
579 | |
580 | /* |
581 | * Something must have been wrong if the figured number of |
582 | * segment is bigger than number of req's physical segments |
583 | */ |
584 | WARN_ON(nsegs > blk_rq_nr_phys_segments(rq)); |
585 | |
586 | return nsegs; |
587 | } |
588 | EXPORT_SYMBOL(__blk_rq_map_sg); |
589 | |
590 | static inline unsigned int blk_rq_get_max_sectors(struct request *rq, |
591 | sector_t offset) |
592 | { |
593 | struct request_queue *q = rq->q; |
594 | unsigned int max_sectors; |
595 | |
596 | if (blk_rq_is_passthrough(rq)) |
597 | return q->limits.max_hw_sectors; |
598 | |
599 | max_sectors = blk_queue_get_max_sectors(q, op: req_op(req: rq)); |
600 | if (!q->limits.chunk_sectors || |
601 | req_op(req: rq) == REQ_OP_DISCARD || |
602 | req_op(req: rq) == REQ_OP_SECURE_ERASE) |
603 | return max_sectors; |
604 | return min(max_sectors, |
605 | blk_chunk_sectors_left(offset, q->limits.chunk_sectors)); |
606 | } |
607 | |
608 | static inline int ll_new_hw_segment(struct request *req, struct bio *bio, |
609 | unsigned int nr_phys_segs) |
610 | { |
611 | if (!blk_cgroup_mergeable(rq: req, bio)) |
612 | goto no_merge; |
613 | |
614 | if (blk_integrity_merge_bio(req->q, req, bio) == false) |
615 | goto no_merge; |
616 | |
617 | /* discard request merge won't add new segment */ |
618 | if (req_op(req) == REQ_OP_DISCARD) |
619 | return 1; |
620 | |
621 | if (req->nr_phys_segments + nr_phys_segs > blk_rq_get_max_segments(rq: req)) |
622 | goto no_merge; |
623 | |
624 | /* |
625 | * This will form the start of a new hw segment. Bump both |
626 | * counters. |
627 | */ |
628 | req->nr_phys_segments += nr_phys_segs; |
629 | return 1; |
630 | |
631 | no_merge: |
632 | req_set_nomerge(q: req->q, req); |
633 | return 0; |
634 | } |
635 | |
636 | int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs) |
637 | { |
638 | if (req_gap_back_merge(req, bio)) |
639 | return 0; |
640 | if (blk_integrity_rq(rq: req) && |
641 | integrity_req_gap_back_merge(req, next: bio)) |
642 | return 0; |
643 | if (!bio_crypt_ctx_back_mergeable(req, bio)) |
644 | return 0; |
645 | if (blk_rq_sectors(rq: req) + bio_sectors(bio) > |
646 | blk_rq_get_max_sectors(rq: req, offset: blk_rq_pos(rq: req))) { |
647 | req_set_nomerge(q: req->q, req); |
648 | return 0; |
649 | } |
650 | |
651 | return ll_new_hw_segment(req, bio, nr_phys_segs: nr_segs); |
652 | } |
653 | |
654 | static int ll_front_merge_fn(struct request *req, struct bio *bio, |
655 | unsigned int nr_segs) |
656 | { |
657 | if (req_gap_front_merge(req, bio)) |
658 | return 0; |
659 | if (blk_integrity_rq(rq: req) && |
660 | integrity_req_gap_front_merge(req, bio)) |
661 | return 0; |
662 | if (!bio_crypt_ctx_front_mergeable(req, bio)) |
663 | return 0; |
664 | if (blk_rq_sectors(rq: req) + bio_sectors(bio) > |
665 | blk_rq_get_max_sectors(rq: req, offset: bio->bi_iter.bi_sector)) { |
666 | req_set_nomerge(q: req->q, req); |
667 | return 0; |
668 | } |
669 | |
670 | return ll_new_hw_segment(req, bio, nr_phys_segs: nr_segs); |
671 | } |
672 | |
673 | static bool req_attempt_discard_merge(struct request_queue *q, struct request *req, |
674 | struct request *next) |
675 | { |
676 | unsigned short segments = blk_rq_nr_discard_segments(rq: req); |
677 | |
678 | if (segments >= queue_max_discard_segments(q)) |
679 | goto no_merge; |
680 | if (blk_rq_sectors(rq: req) + bio_sectors(next->bio) > |
681 | blk_rq_get_max_sectors(rq: req, offset: blk_rq_pos(rq: req))) |
682 | goto no_merge; |
683 | |
684 | req->nr_phys_segments = segments + blk_rq_nr_discard_segments(rq: next); |
685 | return true; |
686 | no_merge: |
687 | req_set_nomerge(q, req); |
688 | return false; |
689 | } |
690 | |
691 | static int ll_merge_requests_fn(struct request_queue *q, struct request *req, |
692 | struct request *next) |
693 | { |
694 | int total_phys_segments; |
695 | |
696 | if (req_gap_back_merge(req, bio: next->bio)) |
697 | return 0; |
698 | |
699 | /* |
700 | * Will it become too large? |
701 | */ |
702 | if ((blk_rq_sectors(rq: req) + blk_rq_sectors(rq: next)) > |
703 | blk_rq_get_max_sectors(rq: req, offset: blk_rq_pos(rq: req))) |
704 | return 0; |
705 | |
706 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; |
707 | if (total_phys_segments > blk_rq_get_max_segments(rq: req)) |
708 | return 0; |
709 | |
710 | if (!blk_cgroup_mergeable(rq: req, bio: next->bio)) |
711 | return 0; |
712 | |
713 | if (blk_integrity_merge_rq(q, req, next) == false) |
714 | return 0; |
715 | |
716 | if (!bio_crypt_ctx_merge_rq(req, next)) |
717 | return 0; |
718 | |
719 | /* Merge is OK... */ |
720 | req->nr_phys_segments = total_phys_segments; |
721 | return 1; |
722 | } |
723 | |
724 | /** |
725 | * blk_rq_set_mixed_merge - mark a request as mixed merge |
726 | * @rq: request to mark as mixed merge |
727 | * |
728 | * Description: |
729 | * @rq is about to be mixed merged. Make sure the attributes |
730 | * which can be mixed are set in each bio and mark @rq as mixed |
731 | * merged. |
732 | */ |
733 | void blk_rq_set_mixed_merge(struct request *rq) |
734 | { |
735 | blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
736 | struct bio *bio; |
737 | |
738 | if (rq->rq_flags & RQF_MIXED_MERGE) |
739 | return; |
740 | |
741 | /* |
742 | * @rq will no longer represent mixable attributes for all the |
743 | * contained bios. It will just track those of the first one. |
744 | * Distributes the attributs to each bio. |
745 | */ |
746 | for (bio = rq->bio; bio; bio = bio->bi_next) { |
747 | WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) && |
748 | (bio->bi_opf & REQ_FAILFAST_MASK) != ff); |
749 | bio->bi_opf |= ff; |
750 | } |
751 | rq->rq_flags |= RQF_MIXED_MERGE; |
752 | } |
753 | |
754 | static inline blk_opf_t bio_failfast(const struct bio *bio) |
755 | { |
756 | if (bio->bi_opf & REQ_RAHEAD) |
757 | return REQ_FAILFAST_MASK; |
758 | |
759 | return bio->bi_opf & REQ_FAILFAST_MASK; |
760 | } |
761 | |
762 | /* |
763 | * After we are marked as MIXED_MERGE, any new RA bio has to be updated |
764 | * as failfast, and request's failfast has to be updated in case of |
765 | * front merge. |
766 | */ |
767 | static inline void blk_update_mixed_merge(struct request *req, |
768 | struct bio *bio, bool front_merge) |
769 | { |
770 | if (req->rq_flags & RQF_MIXED_MERGE) { |
771 | if (bio->bi_opf & REQ_RAHEAD) |
772 | bio->bi_opf |= REQ_FAILFAST_MASK; |
773 | |
774 | if (front_merge) { |
775 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
776 | req->cmd_flags |= bio->bi_opf & REQ_FAILFAST_MASK; |
777 | } |
778 | } |
779 | } |
780 | |
781 | static void blk_account_io_merge_request(struct request *req) |
782 | { |
783 | if (blk_do_io_stat(rq: req)) { |
784 | part_stat_lock(); |
785 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
786 | part_stat_unlock(); |
787 | } |
788 | } |
789 | |
790 | static enum elv_merge blk_try_req_merge(struct request *req, |
791 | struct request *next) |
792 | { |
793 | if (blk_discard_mergable(req)) |
794 | return ELEVATOR_DISCARD_MERGE; |
795 | else if (blk_rq_pos(rq: req) + blk_rq_sectors(rq: req) == blk_rq_pos(rq: next)) |
796 | return ELEVATOR_BACK_MERGE; |
797 | |
798 | return ELEVATOR_NO_MERGE; |
799 | } |
800 | |
801 | /* |
802 | * For non-mq, this has to be called with the request spinlock acquired. |
803 | * For mq with scheduling, the appropriate queue wide lock should be held. |
804 | */ |
805 | static struct request *attempt_merge(struct request_queue *q, |
806 | struct request *req, struct request *next) |
807 | { |
808 | if (!rq_mergeable(rq: req) || !rq_mergeable(rq: next)) |
809 | return NULL; |
810 | |
811 | if (req_op(req) != req_op(req: next)) |
812 | return NULL; |
813 | |
814 | if (rq_data_dir(req) != rq_data_dir(next)) |
815 | return NULL; |
816 | |
817 | if (req->ioprio != next->ioprio) |
818 | return NULL; |
819 | |
820 | /* |
821 | * If we are allowed to merge, then append bio list |
822 | * from next to rq and release next. merge_requests_fn |
823 | * will have updated segment counts, update sector |
824 | * counts here. Handle DISCARDs separately, as they |
825 | * have separate settings. |
826 | */ |
827 | |
828 | switch (blk_try_req_merge(req, next)) { |
829 | case ELEVATOR_DISCARD_MERGE: |
830 | if (!req_attempt_discard_merge(q, req, next)) |
831 | return NULL; |
832 | break; |
833 | case ELEVATOR_BACK_MERGE: |
834 | if (!ll_merge_requests_fn(q, req, next)) |
835 | return NULL; |
836 | break; |
837 | default: |
838 | return NULL; |
839 | } |
840 | |
841 | /* |
842 | * If failfast settings disagree or any of the two is already |
843 | * a mixed merge, mark both as mixed before proceeding. This |
844 | * makes sure that all involved bios have mixable attributes |
845 | * set properly. |
846 | */ |
847 | if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) || |
848 | (req->cmd_flags & REQ_FAILFAST_MASK) != |
849 | (next->cmd_flags & REQ_FAILFAST_MASK)) { |
850 | blk_rq_set_mixed_merge(rq: req); |
851 | blk_rq_set_mixed_merge(rq: next); |
852 | } |
853 | |
854 | /* |
855 | * At this point we have either done a back merge or front merge. We |
856 | * need the smaller start_time_ns of the merged requests to be the |
857 | * current request for accounting purposes. |
858 | */ |
859 | if (next->start_time_ns < req->start_time_ns) |
860 | req->start_time_ns = next->start_time_ns; |
861 | |
862 | req->biotail->bi_next = next->bio; |
863 | req->biotail = next->biotail; |
864 | |
865 | req->__data_len += blk_rq_bytes(rq: next); |
866 | |
867 | if (!blk_discard_mergable(req)) |
868 | elv_merge_requests(q, req, next); |
869 | |
870 | blk_crypto_rq_put_keyslot(rq: next); |
871 | |
872 | /* |
873 | * 'next' is going away, so update stats accordingly |
874 | */ |
875 | blk_account_io_merge_request(req: next); |
876 | |
877 | trace_block_rq_merge(rq: next); |
878 | |
879 | /* |
880 | * ownership of bio passed from next to req, return 'next' for |
881 | * the caller to free |
882 | */ |
883 | next->bio = NULL; |
884 | return next; |
885 | } |
886 | |
887 | static struct request *attempt_back_merge(struct request_queue *q, |
888 | struct request *rq) |
889 | { |
890 | struct request *next = elv_latter_request(q, rq); |
891 | |
892 | if (next) |
893 | return attempt_merge(q, req: rq, next); |
894 | |
895 | return NULL; |
896 | } |
897 | |
898 | static struct request *attempt_front_merge(struct request_queue *q, |
899 | struct request *rq) |
900 | { |
901 | struct request *prev = elv_former_request(q, rq); |
902 | |
903 | if (prev) |
904 | return attempt_merge(q, req: prev, next: rq); |
905 | |
906 | return NULL; |
907 | } |
908 | |
909 | /* |
910 | * Try to merge 'next' into 'rq'. Return true if the merge happened, false |
911 | * otherwise. The caller is responsible for freeing 'next' if the merge |
912 | * happened. |
913 | */ |
914 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, |
915 | struct request *next) |
916 | { |
917 | return attempt_merge(q, req: rq, next); |
918 | } |
919 | |
920 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio) |
921 | { |
922 | if (!rq_mergeable(rq) || !bio_mergeable(bio)) |
923 | return false; |
924 | |
925 | if (req_op(req: rq) != bio_op(bio)) |
926 | return false; |
927 | |
928 | /* different data direction or already started, don't merge */ |
929 | if (bio_data_dir(bio) != rq_data_dir(rq)) |
930 | return false; |
931 | |
932 | /* don't merge across cgroup boundaries */ |
933 | if (!blk_cgroup_mergeable(rq, bio)) |
934 | return false; |
935 | |
936 | /* only merge integrity protected bio into ditto rq */ |
937 | if (blk_integrity_merge_bio(rq->q, rq, bio) == false) |
938 | return false; |
939 | |
940 | /* Only merge if the crypt contexts are compatible */ |
941 | if (!bio_crypt_rq_ctx_compatible(rq, bio)) |
942 | return false; |
943 | |
944 | if (rq->ioprio != bio_prio(bio)) |
945 | return false; |
946 | |
947 | return true; |
948 | } |
949 | |
950 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio) |
951 | { |
952 | if (blk_discard_mergable(req: rq)) |
953 | return ELEVATOR_DISCARD_MERGE; |
954 | else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector) |
955 | return ELEVATOR_BACK_MERGE; |
956 | else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector) |
957 | return ELEVATOR_FRONT_MERGE; |
958 | return ELEVATOR_NO_MERGE; |
959 | } |
960 | |
961 | static void blk_account_io_merge_bio(struct request *req) |
962 | { |
963 | if (!blk_do_io_stat(rq: req)) |
964 | return; |
965 | |
966 | part_stat_lock(); |
967 | part_stat_inc(req->part, merges[op_stat_group(req_op(req))]); |
968 | part_stat_unlock(); |
969 | } |
970 | |
971 | enum bio_merge_status { |
972 | BIO_MERGE_OK, |
973 | BIO_MERGE_NONE, |
974 | BIO_MERGE_FAILED, |
975 | }; |
976 | |
977 | static enum bio_merge_status bio_attempt_back_merge(struct request *req, |
978 | struct bio *bio, unsigned int nr_segs) |
979 | { |
980 | const blk_opf_t ff = bio_failfast(bio); |
981 | |
982 | if (!ll_back_merge_fn(req, bio, nr_segs)) |
983 | return BIO_MERGE_FAILED; |
984 | |
985 | trace_block_bio_backmerge(bio); |
986 | rq_qos_merge(q: req->q, rq: req, bio); |
987 | |
988 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
989 | blk_rq_set_mixed_merge(rq: req); |
990 | |
991 | blk_update_mixed_merge(req, bio, front_merge: false); |
992 | |
993 | req->biotail->bi_next = bio; |
994 | req->biotail = bio; |
995 | req->__data_len += bio->bi_iter.bi_size; |
996 | |
997 | bio_crypt_free_ctx(bio); |
998 | |
999 | blk_account_io_merge_bio(req); |
1000 | return BIO_MERGE_OK; |
1001 | } |
1002 | |
1003 | static enum bio_merge_status bio_attempt_front_merge(struct request *req, |
1004 | struct bio *bio, unsigned int nr_segs) |
1005 | { |
1006 | const blk_opf_t ff = bio_failfast(bio); |
1007 | |
1008 | if (!ll_front_merge_fn(req, bio, nr_segs)) |
1009 | return BIO_MERGE_FAILED; |
1010 | |
1011 | trace_block_bio_frontmerge(bio); |
1012 | rq_qos_merge(q: req->q, rq: req, bio); |
1013 | |
1014 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
1015 | blk_rq_set_mixed_merge(rq: req); |
1016 | |
1017 | blk_update_mixed_merge(req, bio, front_merge: true); |
1018 | |
1019 | bio->bi_next = req->bio; |
1020 | req->bio = bio; |
1021 | |
1022 | req->__sector = bio->bi_iter.bi_sector; |
1023 | req->__data_len += bio->bi_iter.bi_size; |
1024 | |
1025 | bio_crypt_do_front_merge(rq: req, bio); |
1026 | |
1027 | blk_account_io_merge_bio(req); |
1028 | return BIO_MERGE_OK; |
1029 | } |
1030 | |
1031 | static enum bio_merge_status bio_attempt_discard_merge(struct request_queue *q, |
1032 | struct request *req, struct bio *bio) |
1033 | { |
1034 | unsigned short segments = blk_rq_nr_discard_segments(rq: req); |
1035 | |
1036 | if (segments >= queue_max_discard_segments(q)) |
1037 | goto no_merge; |
1038 | if (blk_rq_sectors(rq: req) + bio_sectors(bio) > |
1039 | blk_rq_get_max_sectors(rq: req, offset: blk_rq_pos(rq: req))) |
1040 | goto no_merge; |
1041 | |
1042 | rq_qos_merge(q, rq: req, bio); |
1043 | |
1044 | req->biotail->bi_next = bio; |
1045 | req->biotail = bio; |
1046 | req->__data_len += bio->bi_iter.bi_size; |
1047 | req->nr_phys_segments = segments + 1; |
1048 | |
1049 | blk_account_io_merge_bio(req); |
1050 | return BIO_MERGE_OK; |
1051 | no_merge: |
1052 | req_set_nomerge(q, req); |
1053 | return BIO_MERGE_FAILED; |
1054 | } |
1055 | |
1056 | static enum bio_merge_status blk_attempt_bio_merge(struct request_queue *q, |
1057 | struct request *rq, |
1058 | struct bio *bio, |
1059 | unsigned int nr_segs, |
1060 | bool sched_allow_merge) |
1061 | { |
1062 | if (!blk_rq_merge_ok(rq, bio)) |
1063 | return BIO_MERGE_NONE; |
1064 | |
1065 | switch (blk_try_merge(rq, bio)) { |
1066 | case ELEVATOR_BACK_MERGE: |
1067 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
1068 | return bio_attempt_back_merge(req: rq, bio, nr_segs); |
1069 | break; |
1070 | case ELEVATOR_FRONT_MERGE: |
1071 | if (!sched_allow_merge || blk_mq_sched_allow_merge(q, rq, bio)) |
1072 | return bio_attempt_front_merge(req: rq, bio, nr_segs); |
1073 | break; |
1074 | case ELEVATOR_DISCARD_MERGE: |
1075 | return bio_attempt_discard_merge(q, req: rq, bio); |
1076 | default: |
1077 | return BIO_MERGE_NONE; |
1078 | } |
1079 | |
1080 | return BIO_MERGE_FAILED; |
1081 | } |
1082 | |
1083 | /** |
1084 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
1085 | * @q: request_queue new bio is being queued at |
1086 | * @bio: new bio being queued |
1087 | * @nr_segs: number of segments in @bio |
1088 | * from the passed in @q already in the plug list |
1089 | * |
1090 | * Determine whether @bio being queued on @q can be merged with the previous |
1091 | * request on %current's plugged list. Returns %true if merge was successful, |
1092 | * otherwise %false. |
1093 | * |
1094 | * Plugging coalesces IOs from the same issuer for the same purpose without |
1095 | * going through @q->queue_lock. As such it's more of an issuing mechanism |
1096 | * than scheduling, and the request, while may have elvpriv data, is not |
1097 | * added on the elevator at this point. In addition, we don't have |
1098 | * reliable access to the elevator outside queue lock. Only check basic |
1099 | * merging parameters without querying the elevator. |
1100 | * |
1101 | * Caller must ensure !blk_queue_nomerges(q) beforehand. |
1102 | */ |
1103 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
1104 | unsigned int nr_segs) |
1105 | { |
1106 | struct blk_plug *plug; |
1107 | struct request *rq; |
1108 | |
1109 | plug = blk_mq_plug(bio); |
1110 | if (!plug || rq_list_empty(plug->mq_list)) |
1111 | return false; |
1112 | |
1113 | rq_list_for_each(&plug->mq_list, rq) { |
1114 | if (rq->q == q) { |
1115 | if (blk_attempt_bio_merge(q, rq, bio, nr_segs, sched_allow_merge: false) == |
1116 | BIO_MERGE_OK) |
1117 | return true; |
1118 | break; |
1119 | } |
1120 | |
1121 | /* |
1122 | * Only keep iterating plug list for merges if we have multiple |
1123 | * queues |
1124 | */ |
1125 | if (!plug->multiple_queues) |
1126 | break; |
1127 | } |
1128 | return false; |
1129 | } |
1130 | |
1131 | /* |
1132 | * Iterate list of requests and see if we can merge this bio with any |
1133 | * of them. |
1134 | */ |
1135 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, |
1136 | struct bio *bio, unsigned int nr_segs) |
1137 | { |
1138 | struct request *rq; |
1139 | int checked = 8; |
1140 | |
1141 | list_for_each_entry_reverse(rq, list, queuelist) { |
1142 | if (!checked--) |
1143 | break; |
1144 | |
1145 | switch (blk_attempt_bio_merge(q, rq, bio, nr_segs, sched_allow_merge: true)) { |
1146 | case BIO_MERGE_NONE: |
1147 | continue; |
1148 | case BIO_MERGE_OK: |
1149 | return true; |
1150 | case BIO_MERGE_FAILED: |
1151 | return false; |
1152 | } |
1153 | |
1154 | } |
1155 | |
1156 | return false; |
1157 | } |
1158 | EXPORT_SYMBOL_GPL(blk_bio_list_merge); |
1159 | |
1160 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, |
1161 | unsigned int nr_segs, struct request **merged_request) |
1162 | { |
1163 | struct request *rq; |
1164 | |
1165 | switch (elv_merge(q, &rq, bio)) { |
1166 | case ELEVATOR_BACK_MERGE: |
1167 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
1168 | return false; |
1169 | if (bio_attempt_back_merge(req: rq, bio, nr_segs) != BIO_MERGE_OK) |
1170 | return false; |
1171 | *merged_request = attempt_back_merge(q, rq); |
1172 | if (!*merged_request) |
1173 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); |
1174 | return true; |
1175 | case ELEVATOR_FRONT_MERGE: |
1176 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
1177 | return false; |
1178 | if (bio_attempt_front_merge(req: rq, bio, nr_segs) != BIO_MERGE_OK) |
1179 | return false; |
1180 | *merged_request = attempt_front_merge(q, rq); |
1181 | if (!*merged_request) |
1182 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); |
1183 | return true; |
1184 | case ELEVATOR_DISCARD_MERGE: |
1185 | return bio_attempt_discard_merge(q, req: rq, bio) == BIO_MERGE_OK; |
1186 | default: |
1187 | return false; |
1188 | } |
1189 | } |
1190 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); |
1191 | |