blob: 57f7990b342dd1957077ae48baf71283663c0e73 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Jens Axboed6d48192008-01-29 14:04:06 +01002/*
3 * Functions related to segment and merge handling
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/scatterlist.h>
10
Mike Krinkincda22642015-12-03 17:32:30 +030011#include <trace/events/block.h>
12
Jens Axboed6d48192008-01-29 14:04:06 +010013#include "blk.h"
14
Christoph Hellwige9907002018-09-24 09:43:48 +020015static inline bool bio_will_gap(struct request_queue *q,
16 struct request *prev_rq, struct bio *prev, struct bio *next)
17{
18 struct bio_vec pb, nb;
19
20 if (!bio_has_data(prev) || !queue_virt_boundary(q))
21 return false;
22
23 /*
24 * Don't merge if the 1st bio starts with non-zero offset, otherwise it
25 * is quite difficult to respect the sg gap limit. We work hard to
26 * merge a huge number of small single bios in case of mkfs.
27 */
28 if (prev_rq)
29 bio_get_first_bvec(prev_rq->bio, &pb);
30 else
31 bio_get_first_bvec(prev, &pb);
Johannes Thumshirndf376b22018-11-07 14:58:14 +010032 if (pb.bv_offset & queue_virt_boundary(q))
Christoph Hellwige9907002018-09-24 09:43:48 +020033 return true;
34
35 /*
36 * We don't need to worry about the situation that the merged segment
37 * ends in unaligned virt boundary:
38 *
39 * - if 'pb' ends aligned, the merged segment ends aligned
40 * - if 'pb' ends unaligned, the next bio must include
41 * one single bvec of 'nb', otherwise the 'nb' can't
42 * merge with 'pb'
43 */
44 bio_get_last_bvec(prev, &pb);
45 bio_get_first_bvec(next, &nb);
Christoph Hellwig200a9af2019-05-21 09:01:42 +020046 if (biovec_phys_mergeable(q, &pb, &nb))
Christoph Hellwige9907002018-09-24 09:43:48 +020047 return false;
48 return __bvec_gap_to_prev(q, &pb, nb.bv_offset);
49}
50
51static inline bool req_gap_back_merge(struct request *req, struct bio *bio)
52{
53 return bio_will_gap(req->q, req, req->biotail, bio);
54}
55
56static inline bool req_gap_front_merge(struct request *req, struct bio *bio)
57{
58 return bio_will_gap(req->q, NULL, bio, req->bio);
59}
60
Kent Overstreet54efd502015-04-23 22:37:18 -070061static struct bio *blk_bio_discard_split(struct request_queue *q,
62 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +080063 struct bio_set *bs,
64 unsigned *nsegs)
Kent Overstreet54efd502015-04-23 22:37:18 -070065{
66 unsigned int max_discard_sectors, granularity;
67 int alignment;
68 sector_t tmp;
69 unsigned split_sectors;
70
Ming Leibdced432015-10-20 23:13:52 +080071 *nsegs = 1;
72
Kent Overstreet54efd502015-04-23 22:37:18 -070073 /* Zero-sector (unknown) and one-sector granularities are the same. */
74 granularity = max(q->limits.discard_granularity >> 9, 1U);
75
Ming Lei1adfc5e2018-10-29 20:57:17 +080076 max_discard_sectors = min(q->limits.max_discard_sectors,
77 bio_allowed_max_sectors(q));
Kent Overstreet54efd502015-04-23 22:37:18 -070078 max_discard_sectors -= max_discard_sectors % granularity;
79
80 if (unlikely(!max_discard_sectors)) {
81 /* XXX: warn */
82 return NULL;
83 }
84
85 if (bio_sectors(bio) <= max_discard_sectors)
86 return NULL;
87
88 split_sectors = max_discard_sectors;
89
90 /*
91 * If the next starting sector would be misaligned, stop the discard at
92 * the previous aligned sector.
93 */
94 alignment = (q->limits.discard_alignment >> 9) % granularity;
95
96 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
97 tmp = sector_div(tmp, granularity);
98
99 if (split_sectors > tmp)
100 split_sectors -= tmp;
101
102 return bio_split(bio, split_sectors, GFP_NOIO, bs);
103}
104
Christoph Hellwig885fa132017-04-05 19:21:01 +0200105static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
106 struct bio *bio, struct bio_set *bs, unsigned *nsegs)
107{
Christoph Hellwigd665e122019-07-03 05:24:35 -0700108 *nsegs = 0;
Christoph Hellwig885fa132017-04-05 19:21:01 +0200109
110 if (!q->limits.max_write_zeroes_sectors)
111 return NULL;
112
113 if (bio_sectors(bio) <= q->limits.max_write_zeroes_sectors)
114 return NULL;
115
116 return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
117}
118
Kent Overstreet54efd502015-04-23 22:37:18 -0700119static struct bio *blk_bio_write_same_split(struct request_queue *q,
120 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +0800121 struct bio_set *bs,
122 unsigned *nsegs)
Kent Overstreet54efd502015-04-23 22:37:18 -0700123{
Ming Leibdced432015-10-20 23:13:52 +0800124 *nsegs = 1;
125
Kent Overstreet54efd502015-04-23 22:37:18 -0700126 if (!q->limits.max_write_same_sectors)
127 return NULL;
128
129 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
130 return NULL;
131
132 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
133}
134
Ming Leid0e5fbb2016-01-23 08:05:33 +0800135static inline unsigned get_max_io_size(struct request_queue *q,
136 struct bio *bio)
137{
138 unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
139 unsigned mask = queue_logical_block_size(q) - 1;
140
141 /* aligned to logical block size */
142 sectors &= ~(mask >> 9);
143
144 return sectors;
145}
146
Ming Leidcebd752019-02-15 19:13:12 +0800147static unsigned get_max_segment_size(struct request_queue *q,
148 unsigned offset)
149{
150 unsigned long mask = queue_segment_boundary(q);
151
152 /* default segment boundary mask means no boundary limit */
153 if (mask == BLK_SEG_BOUNDARY_MASK)
154 return queue_max_segment_size(q);
155
156 return min_t(unsigned long, mask - (mask & offset) + 1,
157 queue_max_segment_size(q));
158}
159
160/*
161 * Split the bvec @bv into segments, and update all kinds of
162 * variables.
163 */
164static bool bvec_split_segs(struct request_queue *q, struct bio_vec *bv,
Christoph Hellwig68698752019-05-21 09:01:43 +0200165 unsigned *nsegs, unsigned *sectors, unsigned max_segs)
Ming Leidcebd752019-02-15 19:13:12 +0800166{
167 unsigned len = bv->bv_len;
168 unsigned total_len = 0;
169 unsigned new_nsegs = 0, seg_size = 0;
170
171 /*
172 * Multi-page bvec may be too big to hold in one segment, so the
173 * current bvec has to be splitted as multiple segments.
174 */
Ming Lei05b700b2019-03-03 21:17:48 +0800175 while (len && new_nsegs + *nsegs < max_segs) {
Ming Leidcebd752019-02-15 19:13:12 +0800176 seg_size = get_max_segment_size(q, bv->bv_offset + total_len);
177 seg_size = min(seg_size, len);
178
179 new_nsegs++;
180 total_len += seg_size;
181 len -= seg_size;
182
183 if ((bv->bv_offset + total_len) & queue_virt_boundary(q))
184 break;
185 }
186
Christoph Hellwig68698752019-05-21 09:01:43 +0200187 if (new_nsegs) {
188 *nsegs += new_nsegs;
189 if (sectors)
190 *sectors += total_len >> 9;
Ming Leidcebd752019-02-15 19:13:12 +0800191 }
192
Ming Leidcebd752019-02-15 19:13:12 +0800193 /* split in the middle of the bvec if len != 0 */
194 return !!len;
195}
196
Kent Overstreet54efd502015-04-23 22:37:18 -0700197static struct bio *blk_bio_segment_split(struct request_queue *q,
198 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +0800199 struct bio_set *bs,
200 unsigned *segs)
Kent Overstreet54efd502015-04-23 22:37:18 -0700201{
Jens Axboe5014c312015-09-02 16:46:02 -0600202 struct bio_vec bv, bvprv, *bvprvp = NULL;
Kent Overstreet54efd502015-04-23 22:37:18 -0700203 struct bvec_iter iter;
Christoph Hellwig68698752019-05-21 09:01:43 +0200204 unsigned nsegs = 0, sectors = 0;
Ming Leid0e5fbb2016-01-23 08:05:33 +0800205 const unsigned max_sectors = get_max_io_size(q, bio);
Ming Lei05b700b2019-03-03 21:17:48 +0800206 const unsigned max_segs = queue_max_segments(q);
Kent Overstreet54efd502015-04-23 22:37:18 -0700207
Ming Leidcebd752019-02-15 19:13:12 +0800208 bio_for_each_bvec(bv, bio, iter) {
Kent Overstreet54efd502015-04-23 22:37:18 -0700209 /*
210 * If the queue doesn't support SG gaps and adding this
211 * offset would create a gap, disallow it.
212 */
Jens Axboe5014c312015-09-02 16:46:02 -0600213 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
Kent Overstreet54efd502015-04-23 22:37:18 -0700214 goto split;
215
Ming Leid0e5fbb2016-01-23 08:05:33 +0800216 if (sectors + (bv.bv_len >> 9) > max_sectors) {
Keith Busche36f6202016-01-12 15:08:39 -0700217 /*
218 * Consider this a new segment if we're splitting in
219 * the middle of this vector.
220 */
Ming Lei05b700b2019-03-03 21:17:48 +0800221 if (nsegs < max_segs &&
Ming Leid0e5fbb2016-01-23 08:05:33 +0800222 sectors < max_sectors) {
Ming Leidcebd752019-02-15 19:13:12 +0800223 /* split in the middle of bvec */
224 bv.bv_len = (max_sectors - sectors) << 9;
225 bvec_split_segs(q, &bv, &nsegs,
Ming Lei05b700b2019-03-03 21:17:48 +0800226 &sectors, max_segs);
Keith Busche36f6202016-01-12 15:08:39 -0700227 }
Ming Leicf8c0c62017-12-18 20:22:16 +0800228 goto split;
Keith Busche36f6202016-01-12 15:08:39 -0700229 }
230
Ming Lei05b700b2019-03-03 21:17:48 +0800231 if (nsegs == max_segs)
Kent Overstreet54efd502015-04-23 22:37:18 -0700232 goto split;
233
Kent Overstreet54efd502015-04-23 22:37:18 -0700234 bvprv = bv;
Ming Lei578270b2015-11-24 10:35:29 +0800235 bvprvp = &bvprv;
Ming Leidcebd752019-02-15 19:13:12 +0800236
Ming Leibbcbbd52019-02-27 20:40:12 +0800237 if (bv.bv_offset + bv.bv_len <= PAGE_SIZE) {
238 nsegs++;
Ming Leibbcbbd52019-02-27 20:40:12 +0800239 sectors += bv.bv_len >> 9;
Christoph Hellwig68698752019-05-21 09:01:43 +0200240 } else if (bvec_split_segs(q, &bv, &nsegs, &sectors,
241 max_segs)) {
Ming Leidcebd752019-02-15 19:13:12 +0800242 goto split;
Ming Leibbcbbd52019-02-27 20:40:12 +0800243 }
Kent Overstreet54efd502015-04-23 22:37:18 -0700244 }
245
Christoph Hellwigd6270652019-06-06 12:29:03 +0200246 *segs = nsegs;
247 return NULL;
Kent Overstreet54efd502015-04-23 22:37:18 -0700248split:
Ming Leibdced432015-10-20 23:13:52 +0800249 *segs = nsegs;
Christoph Hellwigd6270652019-06-06 12:29:03 +0200250 return bio_split(bio, sectors, GFP_NOIO, bs);
Kent Overstreet54efd502015-04-23 22:37:18 -0700251}
252
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200253void __blk_queue_split(struct request_queue *q, struct bio **bio,
254 unsigned int *nr_segs)
Kent Overstreet54efd502015-04-23 22:37:18 -0700255{
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200256 struct bio *split;
Kent Overstreet54efd502015-04-23 22:37:18 -0700257
Adrian Hunter7afafc82016-08-16 10:59:35 +0300258 switch (bio_op(*bio)) {
259 case REQ_OP_DISCARD:
260 case REQ_OP_SECURE_ERASE:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200261 split = blk_bio_discard_split(q, *bio, &q->bio_split, nr_segs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300262 break;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800263 case REQ_OP_WRITE_ZEROES:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200264 split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
265 nr_segs);
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800266 break;
Adrian Hunter7afafc82016-08-16 10:59:35 +0300267 case REQ_OP_WRITE_SAME:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200268 split = blk_bio_write_same_split(q, *bio, &q->bio_split,
269 nr_segs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300270 break;
271 default:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200272 split = blk_bio_segment_split(q, *bio, &q->bio_split, nr_segs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300273 break;
274 }
Ming Leibdced432015-10-20 23:13:52 +0800275
Kent Overstreet54efd502015-04-23 22:37:18 -0700276 if (split) {
Ming Lei6ac45ae2015-10-20 23:13:53 +0800277 /* there isn't chance to merge the splitted bio */
Jens Axboe1eff9d32016-08-05 15:35:16 -0600278 split->bi_opf |= REQ_NOMERGE;
Ming Lei6ac45ae2015-10-20 23:13:53 +0800279
Jens Axboe947b7ac2019-01-27 06:35:28 -0700280 /*
281 * Since we're recursing into make_request here, ensure
282 * that we mark this bio as already having entered the queue.
283 * If not, and the queue is going away, we can get stuck
284 * forever on waiting for the queue reference to drop. But
285 * that will never happen, as we're already holding a
286 * reference to it.
287 */
288 bio_set_flag(*bio, BIO_QUEUE_ENTERED);
289
Kent Overstreet54efd502015-04-23 22:37:18 -0700290 bio_chain(split, *bio);
Mike Krinkincda22642015-12-03 17:32:30 +0300291 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
Kent Overstreet54efd502015-04-23 22:37:18 -0700292 generic_make_request(*bio);
293 *bio = split;
294 }
295}
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200296
297void blk_queue_split(struct request_queue *q, struct bio **bio)
298{
299 unsigned int nr_segs;
300
301 __blk_queue_split(q, bio, &nr_segs);
302}
Kent Overstreet54efd502015-04-23 22:37:18 -0700303EXPORT_SYMBOL(blk_queue_split);
304
Christoph Hellwige9cd19c2019-06-06 12:29:02 +0200305unsigned int blk_recalc_rq_segments(struct request *rq)
Jens Axboed6d48192008-01-29 14:04:06 +0100306{
Christoph Hellwig68698752019-05-21 09:01:43 +0200307 unsigned int nr_phys_segs = 0;
Christoph Hellwige9cd19c2019-06-06 12:29:02 +0200308 struct req_iterator iter;
Christoph Hellwig68698752019-05-21 09:01:43 +0200309 struct bio_vec bv;
Jens Axboed6d48192008-01-29 14:04:06 +0100310
Christoph Hellwige9cd19c2019-06-06 12:29:02 +0200311 if (!rq->bio)
Jens Axboe1e428072009-02-23 09:03:10 +0100312 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100313
Christoph Hellwige9cd19c2019-06-06 12:29:02 +0200314 switch (bio_op(rq->bio)) {
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800315 case REQ_OP_DISCARD:
316 case REQ_OP_SECURE_ERASE:
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800317 case REQ_OP_WRITE_ZEROES:
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700318 return 0;
319 case REQ_OP_WRITE_SAME:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700320 return 1;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800321 }
Kent Overstreet5cb88502014-02-07 13:53:46 -0700322
Christoph Hellwige9cd19c2019-06-06 12:29:02 +0200323 rq_for_each_bvec(bv, rq, iter)
324 bvec_split_segs(rq->q, &bv, &nr_phys_segs, NULL, UINT_MAX);
Jens Axboe1e428072009-02-23 09:03:10 +0100325 return nr_phys_segs;
326}
327
Ming Lei48d77272019-02-27 20:40:11 +0800328static inline struct scatterlist *blk_next_sg(struct scatterlist **sg,
Ming Lei862e5a52019-02-15 19:13:13 +0800329 struct scatterlist *sglist)
330{
331 if (!*sg)
332 return sglist;
333
334 /*
335 * If the driver previously mapped a shorter list, we could see a
336 * termination bit prematurely unless it fully inits the sg table
337 * on each mapping. We KNOW that there must be more entries here
338 * or the driver would be buggy, so force clear the termination bit
339 * to avoid doing a full sg_init_table() in drivers for each command.
340 */
341 sg_unmark_end(*sg);
342 return sg_next(*sg);
343}
344
345static unsigned blk_bvec_map_sg(struct request_queue *q,
346 struct bio_vec *bvec, struct scatterlist *sglist,
347 struct scatterlist **sg)
348{
349 unsigned nbytes = bvec->bv_len;
Christoph Hellwig8a96a0e2019-04-11 08:23:27 +0200350 unsigned nsegs = 0, total = 0;
Ming Lei862e5a52019-02-15 19:13:13 +0800351
352 while (nbytes > 0) {
Christoph Hellwig8a96a0e2019-04-11 08:23:27 +0200353 unsigned offset = bvec->bv_offset + total;
354 unsigned len = min(get_max_segment_size(q, offset), nbytes);
Christoph Hellwigf9f76872019-04-19 08:56:24 +0200355 struct page *page = bvec->bv_page;
356
357 /*
358 * Unfortunately a fair number of drivers barf on scatterlists
359 * that have an offset larger than PAGE_SIZE, despite other
360 * subsystems dealing with that invariant just fine. For now
361 * stick to the legacy format where we never present those from
362 * the block layer, but the code below should be removed once
363 * these offenders (mostly MMC/SD drivers) are fixed.
364 */
365 page += (offset >> PAGE_SHIFT);
366 offset &= ~PAGE_MASK;
Ming Lei862e5a52019-02-15 19:13:13 +0800367
368 *sg = blk_next_sg(sg, sglist);
Christoph Hellwigf9f76872019-04-19 08:56:24 +0200369 sg_set_page(*sg, page, len, offset);
Ming Lei862e5a52019-02-15 19:13:13 +0800370
Christoph Hellwig8a96a0e2019-04-11 08:23:27 +0200371 total += len;
372 nbytes -= len;
Ming Lei862e5a52019-02-15 19:13:13 +0800373 nsegs++;
374 }
375
376 return nsegs;
377}
378
Ming Lei16e3e412019-03-17 18:01:11 +0800379static inline int __blk_bvec_map_sg(struct bio_vec bv,
380 struct scatterlist *sglist, struct scatterlist **sg)
381{
382 *sg = blk_next_sg(sg, sglist);
383 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
384 return 1;
385}
386
Ming Leif6970f82019-03-17 18:01:12 +0800387/* only try to merge bvecs into one sg if they are from two bios */
388static inline bool
389__blk_segment_map_sg_merge(struct request_queue *q, struct bio_vec *bvec,
390 struct bio_vec *bvprv, struct scatterlist **sg)
Asias He963ab9e2012-08-02 23:42:03 +0200391{
392
393 int nbytes = bvec->bv_len;
394
Ming Leif6970f82019-03-17 18:01:12 +0800395 if (!*sg)
396 return false;
Asias He963ab9e2012-08-02 23:42:03 +0200397
Ming Leif6970f82019-03-17 18:01:12 +0800398 if ((*sg)->length + nbytes > queue_max_segment_size(q))
399 return false;
400
401 if (!biovec_phys_mergeable(q, bvprv, bvec))
402 return false;
403
404 (*sg)->length += nbytes;
405
406 return true;
Asias He963ab9e2012-08-02 23:42:03 +0200407}
408
Kent Overstreet5cb88502014-02-07 13:53:46 -0700409static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
410 struct scatterlist *sglist,
411 struct scatterlist **sg)
412{
Ming Leib21e11c2019-04-02 10:26:44 +0800413 struct bio_vec uninitialized_var(bvec), bvprv = { NULL };
Kent Overstreet5cb88502014-02-07 13:53:46 -0700414 struct bvec_iter iter;
Christoph Hellwig38417462018-12-13 16:17:10 +0100415 int nsegs = 0;
Ming Leif6970f82019-03-17 18:01:12 +0800416 bool new_bio = false;
Kent Overstreet5cb88502014-02-07 13:53:46 -0700417
Ming Leif6970f82019-03-17 18:01:12 +0800418 for_each_bio(bio) {
419 bio_for_each_bvec(bvec, bio, iter) {
420 /*
421 * Only try to merge bvecs from two bios given we
422 * have done bio internal merge when adding pages
423 * to bio
424 */
425 if (new_bio &&
426 __blk_segment_map_sg_merge(q, &bvec, &bvprv, sg))
427 goto next_bvec;
428
429 if (bvec.bv_offset + bvec.bv_len <= PAGE_SIZE)
430 nsegs += __blk_bvec_map_sg(bvec, sglist, sg);
431 else
432 nsegs += blk_bvec_map_sg(q, &bvec, sglist, sg);
433 next_bvec:
434 new_bio = false;
435 }
Ming Leib21e11c2019-04-02 10:26:44 +0800436 if (likely(bio->bi_iter.bi_size)) {
437 bvprv = bvec;
438 new_bio = true;
439 }
Ming Leif6970f82019-03-17 18:01:12 +0800440 }
Kent Overstreet5cb88502014-02-07 13:53:46 -0700441
442 return nsegs;
443}
444
Jens Axboed6d48192008-01-29 14:04:06 +0100445/*
446 * map a request to scatterlist, return number of sg entries setup. Caller
447 * must make sure sg can hold rq->nr_phys_segments entries
448 */
449int blk_rq_map_sg(struct request_queue *q, struct request *rq,
450 struct scatterlist *sglist)
451{
Kent Overstreet5cb88502014-02-07 13:53:46 -0700452 struct scatterlist *sg = NULL;
453 int nsegs = 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100454
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700455 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
Ming Leicae6c2e2019-03-17 18:01:10 +0800456 nsegs = __blk_bvec_map_sg(rq->special_vec, sglist, &sg);
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700457 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
Ming Leicae6c2e2019-03-17 18:01:10 +0800458 nsegs = __blk_bvec_map_sg(bio_iovec(rq->bio), sglist, &sg);
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700459 else if (rq->bio)
Kent Overstreet5cb88502014-02-07 13:53:46 -0700460 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200461
Christoph Hellwige8064022016-10-20 15:12:13 +0200462 if (unlikely(rq->rq_flags & RQF_COPY_USER) &&
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900463 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
464 unsigned int pad_len =
465 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200466
467 sg->length += pad_len;
468 rq->extra_len += pad_len;
469 }
470
Tejun Heo2fb98e82008-02-19 11:36:53 +0100471 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
Mike Christiea8ebb052016-06-05 14:31:45 -0500472 if (op_is_write(req_op(rq)))
Tejun Heodb0a2e02008-02-19 11:36:55 +0100473 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
474
Dan Williamsda81ed12015-08-07 18:15:14 +0200475 sg_unmark_end(sg);
Jens Axboed6d48192008-01-29 14:04:06 +0100476 sg = sg_next(sg);
477 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
478 q->dma_drain_size,
479 ((unsigned long)q->dma_drain_buffer) &
480 (PAGE_SIZE - 1));
481 nsegs++;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100482 rq->extra_len += q->dma_drain_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100483 }
484
485 if (sg)
486 sg_mark_end(sg);
487
Ming Lei12e57f52015-11-24 10:35:31 +0800488 /*
489 * Something must have been wrong if the figured number of
490 * segment is bigger than number of req's physical segments
491 */
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700492 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
Ming Lei12e57f52015-11-24 10:35:31 +0800493
Jens Axboed6d48192008-01-29 14:04:06 +0100494 return nsegs;
495}
Jens Axboed6d48192008-01-29 14:04:06 +0100496EXPORT_SYMBOL(blk_rq_map_sg);
497
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200498static inline int ll_new_hw_segment(struct request *req, struct bio *bio,
499 unsigned int nr_phys_segs)
Jens Axboed6d48192008-01-29 14:04:06 +0100500{
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200501 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(req->q))
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200502 goto no_merge;
503
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200504 if (blk_integrity_merge_bio(req->q, req, bio) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200505 goto no_merge;
Jens Axboed6d48192008-01-29 14:04:06 +0100506
507 /*
508 * This will form the start of a new hw segment. Bump both
509 * counters.
510 */
Jens Axboed6d48192008-01-29 14:04:06 +0100511 req->nr_phys_segments += nr_phys_segs;
512 return 1;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200513
514no_merge:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200515 req_set_nomerge(req->q, req);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200516 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100517}
518
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200519int ll_back_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
Jens Axboed6d48192008-01-29 14:04:06 +0100520{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300521 if (req_gap_back_merge(req, bio))
522 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600523 if (blk_integrity_rq(req) &&
524 integrity_req_gap_back_merge(req, bio))
525 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400526 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600527 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200528 req_set_nomerge(req->q, req);
Jens Axboed6d48192008-01-29 14:04:06 +0100529 return 0;
530 }
Jens Axboed6d48192008-01-29 14:04:06 +0100531
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200532 return ll_new_hw_segment(req, bio, nr_segs);
Jens Axboed6d48192008-01-29 14:04:06 +0100533}
534
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200535int ll_front_merge_fn(struct request *req, struct bio *bio, unsigned int nr_segs)
Jens Axboed6d48192008-01-29 14:04:06 +0100536{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300537 if (req_gap_front_merge(req, bio))
538 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600539 if (blk_integrity_rq(req) &&
540 integrity_req_gap_front_merge(req, bio))
541 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400542 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600543 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200544 req_set_nomerge(req->q, req);
Jens Axboed6d48192008-01-29 14:04:06 +0100545 return 0;
546 }
Jens Axboed6d48192008-01-29 14:04:06 +0100547
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200548 return ll_new_hw_segment(req, bio, nr_segs);
Jens Axboed6d48192008-01-29 14:04:06 +0100549}
550
Jens Axboe445251d2018-02-01 14:01:02 -0700551static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
552 struct request *next)
553{
554 unsigned short segments = blk_rq_nr_discard_segments(req);
555
556 if (segments >= queue_max_discard_segments(q))
557 goto no_merge;
558 if (blk_rq_sectors(req) + bio_sectors(next->bio) >
559 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
560 goto no_merge;
561
562 req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
563 return true;
564no_merge:
565 req_set_nomerge(q, req);
566 return false;
567}
568
Jens Axboed6d48192008-01-29 14:04:06 +0100569static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
570 struct request *next)
571{
572 int total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100573
Jens Axboe5e7c4272015-09-03 19:28:20 +0300574 if (req_gap_back_merge(req, next->bio))
Keith Busch854fbb92015-02-11 08:20:13 -0700575 return 0;
576
Jens Axboed6d48192008-01-29 14:04:06 +0100577 /*
578 * Will it become too large?
579 */
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400580 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600581 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
Jens Axboed6d48192008-01-29 14:04:06 +0100582 return 0;
583
584 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
Martin K. Petersen8a783622010-02-26 00:20:39 -0500585 if (total_phys_segments > queue_max_segments(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100586 return 0;
587
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400588 if (blk_integrity_merge_rq(q, req, next) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200589 return 0;
590
Jens Axboed6d48192008-01-29 14:04:06 +0100591 /* Merge is OK... */
592 req->nr_phys_segments = total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100593 return 1;
594}
595
Tejun Heo80a761f2009-07-03 17:48:17 +0900596/**
597 * blk_rq_set_mixed_merge - mark a request as mixed merge
598 * @rq: request to mark as mixed merge
599 *
600 * Description:
601 * @rq is about to be mixed merged. Make sure the attributes
602 * which can be mixed are set in each bio and mark @rq as mixed
603 * merged.
604 */
605void blk_rq_set_mixed_merge(struct request *rq)
606{
607 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
608 struct bio *bio;
609
Christoph Hellwige8064022016-10-20 15:12:13 +0200610 if (rq->rq_flags & RQF_MIXED_MERGE)
Tejun Heo80a761f2009-07-03 17:48:17 +0900611 return;
612
613 /*
614 * @rq will no longer represent mixable attributes for all the
615 * contained bios. It will just track those of the first one.
616 * Distributes the attributs to each bio.
617 */
618 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600619 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
620 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
621 bio->bi_opf |= ff;
Tejun Heo80a761f2009-07-03 17:48:17 +0900622 }
Christoph Hellwige8064022016-10-20 15:12:13 +0200623 rq->rq_flags |= RQF_MIXED_MERGE;
Tejun Heo80a761f2009-07-03 17:48:17 +0900624}
625
Jerome Marchand26308ea2009-03-27 10:31:51 +0100626static void blk_account_io_merge(struct request *req)
627{
628 if (blk_do_io_stat(req)) {
629 struct hd_struct *part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100630
Mike Snitzer112f1582018-12-06 11:41:18 -0500631 part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100632 part = req->part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100633
Jens Axboed62e26b2017-06-30 21:55:08 -0600634 part_dec_in_flight(req->q, part, rq_data_dir(req));
Jerome Marchand26308ea2009-03-27 10:31:51 +0100635
Jens Axboe6c23a962011-01-07 08:43:37 +0100636 hd_struct_put(part);
Jerome Marchand26308ea2009-03-27 10:31:51 +0100637 part_stat_unlock();
638 }
639}
Jianchao Wang698404662018-10-27 19:52:14 +0800640/*
641 * Two cases of handling DISCARD merge:
642 * If max_discard_segments > 1, the driver takes every bio
643 * as a range and send them to controller together. The ranges
644 * needn't to be contiguous.
645 * Otherwise, the bios/requests will be handled as same as
646 * others which should be contiguous.
647 */
648static inline bool blk_discard_mergable(struct request *req)
649{
650 if (req_op(req) == REQ_OP_DISCARD &&
651 queue_max_discard_segments(req->q) > 1)
652 return true;
653 return false;
654}
655
Eric Biggerse96c0d82018-11-14 17:19:46 -0800656static enum elv_merge blk_try_req_merge(struct request *req,
657 struct request *next)
Jianchao Wang698404662018-10-27 19:52:14 +0800658{
659 if (blk_discard_mergable(req))
660 return ELEVATOR_DISCARD_MERGE;
661 else if (blk_rq_pos(req) + blk_rq_sectors(req) == blk_rq_pos(next))
662 return ELEVATOR_BACK_MERGE;
663
664 return ELEVATOR_NO_MERGE;
665}
Jerome Marchand26308ea2009-03-27 10:31:51 +0100666
Jens Axboed6d48192008-01-29 14:04:06 +0100667/*
Jens Axboeb973cb72017-02-02 08:54:40 -0700668 * For non-mq, this has to be called with the request spinlock acquired.
669 * For mq with scheduling, the appropriate queue wide lock should be held.
Jens Axboed6d48192008-01-29 14:04:06 +0100670 */
Jens Axboeb973cb72017-02-02 08:54:40 -0700671static struct request *attempt_merge(struct request_queue *q,
672 struct request *req, struct request *next)
Jens Axboed6d48192008-01-29 14:04:06 +0100673{
674 if (!rq_mergeable(req) || !rq_mergeable(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700675 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100676
Christoph Hellwig288dab82016-06-09 16:00:36 +0200677 if (req_op(req) != req_op(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700678 return NULL;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400679
Jens Axboed6d48192008-01-29 14:04:06 +0100680 if (rq_data_dir(req) != rq_data_dir(next)
Jens Axboe2081a562018-10-12 12:39:10 -0600681 || req->rq_disk != next->rq_disk)
Jens Axboeb973cb72017-02-02 08:54:40 -0700682 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100683
Mike Christie8fe0d472016-06-05 14:32:15 -0500684 if (req_op(req) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400685 !blk_write_same_mergeable(req->bio, next->bio))
Jens Axboeb973cb72017-02-02 08:54:40 -0700686 return NULL;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400687
Jens Axboed6d48192008-01-29 14:04:06 +0100688 /*
Jens Axboecb6934f2017-06-27 09:22:02 -0600689 * Don't allow merge of different write hints, or for a hint with
690 * non-hint IO.
691 */
692 if (req->write_hint != next->write_hint)
693 return NULL;
694
Damien Le Moal668ffc02018-11-20 10:52:37 +0900695 if (req->ioprio != next->ioprio)
696 return NULL;
697
Jens Axboecb6934f2017-06-27 09:22:02 -0600698 /*
Jens Axboed6d48192008-01-29 14:04:06 +0100699 * If we are allowed to merge, then append bio list
700 * from next to rq and release next. merge_requests_fn
701 * will have updated segment counts, update sector
Jens Axboe445251d2018-02-01 14:01:02 -0700702 * counts here. Handle DISCARDs separately, as they
703 * have separate settings.
Jens Axboed6d48192008-01-29 14:04:06 +0100704 */
Jianchao Wang698404662018-10-27 19:52:14 +0800705
706 switch (blk_try_req_merge(req, next)) {
707 case ELEVATOR_DISCARD_MERGE:
Jens Axboe445251d2018-02-01 14:01:02 -0700708 if (!req_attempt_discard_merge(q, req, next))
709 return NULL;
Jianchao Wang698404662018-10-27 19:52:14 +0800710 break;
711 case ELEVATOR_BACK_MERGE:
712 if (!ll_merge_requests_fn(q, req, next))
713 return NULL;
714 break;
715 default:
Jens Axboeb973cb72017-02-02 08:54:40 -0700716 return NULL;
Jianchao Wang698404662018-10-27 19:52:14 +0800717 }
Jens Axboed6d48192008-01-29 14:04:06 +0100718
719 /*
Tejun Heo80a761f2009-07-03 17:48:17 +0900720 * If failfast settings disagree or any of the two is already
721 * a mixed merge, mark both as mixed before proceeding. This
722 * makes sure that all involved bios have mixable attributes
723 * set properly.
724 */
Christoph Hellwige8064022016-10-20 15:12:13 +0200725 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
Tejun Heo80a761f2009-07-03 17:48:17 +0900726 (req->cmd_flags & REQ_FAILFAST_MASK) !=
727 (next->cmd_flags & REQ_FAILFAST_MASK)) {
728 blk_rq_set_mixed_merge(req);
729 blk_rq_set_mixed_merge(next);
730 }
731
732 /*
Omar Sandoval522a7772018-05-09 02:08:53 -0700733 * At this point we have either done a back merge or front merge. We
734 * need the smaller start_time_ns of the merged requests to be the
735 * current request for accounting purposes.
Jens Axboed6d48192008-01-29 14:04:06 +0100736 */
Omar Sandoval522a7772018-05-09 02:08:53 -0700737 if (next->start_time_ns < req->start_time_ns)
738 req->start_time_ns = next->start_time_ns;
Jens Axboed6d48192008-01-29 14:04:06 +0100739
740 req->biotail->bi_next = next->bio;
741 req->biotail = next->biotail;
742
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900743 req->__data_len += blk_rq_bytes(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100744
Ming Lei2a5cf352018-12-01 00:38:18 +0800745 if (!blk_discard_mergable(req))
Jens Axboe445251d2018-02-01 14:01:02 -0700746 elv_merge_requests(q, req, next);
Jens Axboed6d48192008-01-29 14:04:06 +0100747
Jerome Marchand42dad762009-04-22 14:01:49 +0200748 /*
749 * 'next' is going away, so update stats accordingly
750 */
751 blk_account_io_merge(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100752
Jens Axboee4d750c2017-02-03 09:48:28 -0700753 /*
754 * ownership of bio passed from next to req, return 'next' for
755 * the caller to free
756 */
Boaz Harrosh1cd96c22009-03-24 12:35:07 +0100757 next->bio = NULL;
Jens Axboeb973cb72017-02-02 08:54:40 -0700758 return next;
Jens Axboed6d48192008-01-29 14:04:06 +0100759}
760
Jens Axboeb973cb72017-02-02 08:54:40 -0700761struct request *attempt_back_merge(struct request_queue *q, struct request *rq)
Jens Axboed6d48192008-01-29 14:04:06 +0100762{
763 struct request *next = elv_latter_request(q, rq);
764
765 if (next)
766 return attempt_merge(q, rq, next);
767
Jens Axboeb973cb72017-02-02 08:54:40 -0700768 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100769}
770
Jens Axboeb973cb72017-02-02 08:54:40 -0700771struct request *attempt_front_merge(struct request_queue *q, struct request *rq)
Jens Axboed6d48192008-01-29 14:04:06 +0100772{
773 struct request *prev = elv_former_request(q, rq);
774
775 if (prev)
776 return attempt_merge(q, prev, rq);
777
Jens Axboeb973cb72017-02-02 08:54:40 -0700778 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100779}
Jens Axboe5e84ea32011-03-21 10:14:27 +0100780
781int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
782 struct request *next)
783{
Jens Axboee4d750c2017-02-03 09:48:28 -0700784 struct request *free;
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700785
Jens Axboee4d750c2017-02-03 09:48:28 -0700786 free = attempt_merge(q, rq, next);
787 if (free) {
Jens Axboe92bc5a22018-10-24 13:52:28 -0600788 blk_put_request(free);
Jens Axboee4d750c2017-02-03 09:48:28 -0700789 return 1;
790 }
791
792 return 0;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100793}
Tejun Heo050c8ea2012-02-08 09:19:38 +0100794
795bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
796{
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400797 if (!rq_mergeable(rq) || !bio_mergeable(bio))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100798 return false;
799
Christoph Hellwig288dab82016-06-09 16:00:36 +0200800 if (req_op(rq) != bio_op(bio))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400801 return false;
802
Tejun Heo050c8ea2012-02-08 09:19:38 +0100803 /* different data direction or already started, don't merge */
804 if (bio_data_dir(bio) != rq_data_dir(rq))
805 return false;
806
Jens Axboe2081a562018-10-12 12:39:10 -0600807 /* must be same device */
808 if (rq->rq_disk != bio->bi_disk)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100809 return false;
810
811 /* only merge integrity protected bio into ditto rq */
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400812 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100813 return false;
814
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400815 /* must be using the same buffer */
Mike Christie8fe0d472016-06-05 14:32:15 -0500816 if (req_op(rq) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400817 !blk_write_same_mergeable(rq->bio, bio))
818 return false;
819
Jens Axboecb6934f2017-06-27 09:22:02 -0600820 /*
821 * Don't allow merge of different write hints, or for a hint with
822 * non-hint IO.
823 */
824 if (rq->write_hint != bio->bi_write_hint)
825 return false;
826
Damien Le Moal668ffc02018-11-20 10:52:37 +0900827 if (rq->ioprio != bio_prio(bio))
828 return false;
829
Tejun Heo050c8ea2012-02-08 09:19:38 +0100830 return true;
831}
832
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100833enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100834{
Jianchao Wang698404662018-10-27 19:52:14 +0800835 if (blk_discard_mergable(rq))
Christoph Hellwig1e739732017-02-08 14:46:49 +0100836 return ELEVATOR_DISCARD_MERGE;
837 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100838 return ELEVATOR_BACK_MERGE;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700839 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100840 return ELEVATOR_FRONT_MERGE;
841 return ELEVATOR_NO_MERGE;
842}