blob: 2afa262425d102e5c1500912f3b6d77525754566 [file] [log] [blame]
Jens Axboed6d48192008-01-29 14:04:06 +01001/*
2 * Functions related to segment and merge handling
3 */
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/bio.h>
7#include <linux/blkdev.h>
8#include <linux/scatterlist.h>
9
Mike Krinkincda22642015-12-03 17:32:30 +030010#include <trace/events/block.h>
11
Jens Axboed6d48192008-01-29 14:04:06 +010012#include "blk.h"
13
Kent Overstreet54efd502015-04-23 22:37:18 -070014static struct bio *blk_bio_discard_split(struct request_queue *q,
15 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +080016 struct bio_set *bs,
17 unsigned *nsegs)
Kent Overstreet54efd502015-04-23 22:37:18 -070018{
19 unsigned int max_discard_sectors, granularity;
20 int alignment;
21 sector_t tmp;
22 unsigned split_sectors;
23
Ming Leibdced432015-10-20 23:13:52 +080024 *nsegs = 1;
25
Kent Overstreet54efd502015-04-23 22:37:18 -070026 /* Zero-sector (unknown) and one-sector granularities are the same. */
27 granularity = max(q->limits.discard_granularity >> 9, 1U);
28
29 max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
30 max_discard_sectors -= max_discard_sectors % granularity;
31
32 if (unlikely(!max_discard_sectors)) {
33 /* XXX: warn */
34 return NULL;
35 }
36
37 if (bio_sectors(bio) <= max_discard_sectors)
38 return NULL;
39
40 split_sectors = max_discard_sectors;
41
42 /*
43 * If the next starting sector would be misaligned, stop the discard at
44 * the previous aligned sector.
45 */
46 alignment = (q->limits.discard_alignment >> 9) % granularity;
47
48 tmp = bio->bi_iter.bi_sector + split_sectors - alignment;
49 tmp = sector_div(tmp, granularity);
50
51 if (split_sectors > tmp)
52 split_sectors -= tmp;
53
54 return bio_split(bio, split_sectors, GFP_NOIO, bs);
55}
56
57static struct bio *blk_bio_write_same_split(struct request_queue *q,
58 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +080059 struct bio_set *bs,
60 unsigned *nsegs)
Kent Overstreet54efd502015-04-23 22:37:18 -070061{
Ming Leibdced432015-10-20 23:13:52 +080062 *nsegs = 1;
63
Kent Overstreet54efd502015-04-23 22:37:18 -070064 if (!q->limits.max_write_same_sectors)
65 return NULL;
66
67 if (bio_sectors(bio) <= q->limits.max_write_same_sectors)
68 return NULL;
69
70 return bio_split(bio, q->limits.max_write_same_sectors, GFP_NOIO, bs);
71}
72
Ming Leid0e5fbb2016-01-23 08:05:33 +080073static inline unsigned get_max_io_size(struct request_queue *q,
74 struct bio *bio)
75{
76 unsigned sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
77 unsigned mask = queue_logical_block_size(q) - 1;
78
79 /* aligned to logical block size */
80 sectors &= ~(mask >> 9);
81
82 return sectors;
83}
84
Kent Overstreet54efd502015-04-23 22:37:18 -070085static struct bio *blk_bio_segment_split(struct request_queue *q,
86 struct bio *bio,
Ming Leibdced432015-10-20 23:13:52 +080087 struct bio_set *bs,
88 unsigned *segs)
Kent Overstreet54efd502015-04-23 22:37:18 -070089{
Jens Axboe5014c312015-09-02 16:46:02 -060090 struct bio_vec bv, bvprv, *bvprvp = NULL;
Kent Overstreet54efd502015-04-23 22:37:18 -070091 struct bvec_iter iter;
Kent Overstreet8ae12662015-04-27 23:48:34 -070092 unsigned seg_size = 0, nsegs = 0, sectors = 0;
Ming Lei02e70742015-11-24 10:35:30 +080093 unsigned front_seg_size = bio->bi_seg_front_size;
94 bool do_split = true;
95 struct bio *new = NULL;
Ming Leid0e5fbb2016-01-23 08:05:33 +080096 const unsigned max_sectors = get_max_io_size(q, bio);
Ming Lei4d70dca2016-08-23 21:49:45 +080097 unsigned bvecs = 0;
Kent Overstreet54efd502015-04-23 22:37:18 -070098
Kent Overstreet54efd502015-04-23 22:37:18 -070099 bio_for_each_segment(bv, bio, iter) {
Kent Overstreet54efd502015-04-23 22:37:18 -0700100 /*
Ming Lei4d70dca2016-08-23 21:49:45 +0800101 * With arbitrary bio size, the incoming bio may be very
102 * big. We have to split the bio into small bios so that
103 * each holds at most BIO_MAX_PAGES bvecs because
104 * bio_clone() can fail to allocate big bvecs.
105 *
106 * It should have been better to apply the limit per
107 * request queue in which bio_clone() is involved,
108 * instead of globally. The biggest blocker is the
109 * bio_clone() in bio bounce.
110 *
111 * If bio is splitted by this reason, we should have
112 * allowed to continue bios merging, but don't do
113 * that now for making the change simple.
114 *
115 * TODO: deal with bio bounce's bio_clone() gracefully
116 * and convert the global limit into per-queue limit.
117 */
118 if (bvecs++ >= BIO_MAX_PAGES)
119 goto split;
120
121 /*
Kent Overstreet54efd502015-04-23 22:37:18 -0700122 * If the queue doesn't support SG gaps and adding this
123 * offset would create a gap, disallow it.
124 */
Jens Axboe5014c312015-09-02 16:46:02 -0600125 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
Kent Overstreet54efd502015-04-23 22:37:18 -0700126 goto split;
127
Ming Leid0e5fbb2016-01-23 08:05:33 +0800128 if (sectors + (bv.bv_len >> 9) > max_sectors) {
Keith Busche36f6202016-01-12 15:08:39 -0700129 /*
130 * Consider this a new segment if we're splitting in
131 * the middle of this vector.
132 */
133 if (nsegs < queue_max_segments(q) &&
Ming Leid0e5fbb2016-01-23 08:05:33 +0800134 sectors < max_sectors) {
Keith Busche36f6202016-01-12 15:08:39 -0700135 nsegs++;
Ming Leid0e5fbb2016-01-23 08:05:33 +0800136 sectors = max_sectors;
Keith Busche36f6202016-01-12 15:08:39 -0700137 }
Ming Leid0e5fbb2016-01-23 08:05:33 +0800138 if (sectors)
139 goto split;
140 /* Make this single bvec as the 1st segment */
Keith Busche36f6202016-01-12 15:08:39 -0700141 }
142
Jens Axboe5014c312015-09-02 16:46:02 -0600143 if (bvprvp && blk_queue_cluster(q)) {
Kent Overstreet54efd502015-04-23 22:37:18 -0700144 if (seg_size + bv.bv_len > queue_max_segment_size(q))
145 goto new_segment;
Jens Axboe5014c312015-09-02 16:46:02 -0600146 if (!BIOVEC_PHYS_MERGEABLE(bvprvp, &bv))
Kent Overstreet54efd502015-04-23 22:37:18 -0700147 goto new_segment;
Jens Axboe5014c312015-09-02 16:46:02 -0600148 if (!BIOVEC_SEG_BOUNDARY(q, bvprvp, &bv))
Kent Overstreet54efd502015-04-23 22:37:18 -0700149 goto new_segment;
150
151 seg_size += bv.bv_len;
152 bvprv = bv;
Ming Lei578270b2015-11-24 10:35:29 +0800153 bvprvp = &bvprv;
Ming Lei52cc6ee2015-09-17 09:58:38 -0600154 sectors += bv.bv_len >> 9;
Ming Leia88d32a2015-11-30 16:05:49 +0800155
156 if (nsegs == 1 && seg_size > front_seg_size)
157 front_seg_size = seg_size;
Kent Overstreet54efd502015-04-23 22:37:18 -0700158 continue;
159 }
160new_segment:
161 if (nsegs == queue_max_segments(q))
162 goto split;
163
164 nsegs++;
165 bvprv = bv;
Ming Lei578270b2015-11-24 10:35:29 +0800166 bvprvp = &bvprv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700167 seg_size = bv.bv_len;
Ming Lei52cc6ee2015-09-17 09:58:38 -0600168 sectors += bv.bv_len >> 9;
Ming Lei02e70742015-11-24 10:35:30 +0800169
170 if (nsegs == 1 && seg_size > front_seg_size)
171 front_seg_size = seg_size;
Kent Overstreet54efd502015-04-23 22:37:18 -0700172 }
173
Ming Lei02e70742015-11-24 10:35:30 +0800174 do_split = false;
Kent Overstreet54efd502015-04-23 22:37:18 -0700175split:
Ming Leibdced432015-10-20 23:13:52 +0800176 *segs = nsegs;
Ming Lei02e70742015-11-24 10:35:30 +0800177
178 if (do_split) {
179 new = bio_split(bio, sectors, GFP_NOIO, bs);
180 if (new)
181 bio = new;
182 }
183
184 bio->bi_seg_front_size = front_seg_size;
185 if (seg_size > bio->bi_seg_back_size)
186 bio->bi_seg_back_size = seg_size;
187
188 return do_split ? new : NULL;
Kent Overstreet54efd502015-04-23 22:37:18 -0700189}
190
191void blk_queue_split(struct request_queue *q, struct bio **bio,
192 struct bio_set *bs)
193{
Ming Leibdced432015-10-20 23:13:52 +0800194 struct bio *split, *res;
195 unsigned nsegs;
Kent Overstreet54efd502015-04-23 22:37:18 -0700196
Adrian Hunter7afafc82016-08-16 10:59:35 +0300197 switch (bio_op(*bio)) {
198 case REQ_OP_DISCARD:
199 case REQ_OP_SECURE_ERASE:
Ming Leibdced432015-10-20 23:13:52 +0800200 split = blk_bio_discard_split(q, *bio, bs, &nsegs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300201 break;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800202 case REQ_OP_WRITE_ZEROES:
203 split = NULL;
204 nsegs = (*bio)->bi_phys_segments;
205 break;
Adrian Hunter7afafc82016-08-16 10:59:35 +0300206 case REQ_OP_WRITE_SAME:
Ming Leibdced432015-10-20 23:13:52 +0800207 split = blk_bio_write_same_split(q, *bio, bs, &nsegs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300208 break;
209 default:
Ming Leibdced432015-10-20 23:13:52 +0800210 split = blk_bio_segment_split(q, *bio, q->bio_split, &nsegs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300211 break;
212 }
Ming Leibdced432015-10-20 23:13:52 +0800213
214 /* physical segments can be figured out during splitting */
215 res = split ? split : *bio;
216 res->bi_phys_segments = nsegs;
217 bio_set_flag(res, BIO_SEG_VALID);
Kent Overstreet54efd502015-04-23 22:37:18 -0700218
219 if (split) {
Ming Lei6ac45ae2015-10-20 23:13:53 +0800220 /* there isn't chance to merge the splitted bio */
Jens Axboe1eff9d32016-08-05 15:35:16 -0600221 split->bi_opf |= REQ_NOMERGE;
Ming Lei6ac45ae2015-10-20 23:13:53 +0800222
Kent Overstreet54efd502015-04-23 22:37:18 -0700223 bio_chain(split, *bio);
Mike Krinkincda22642015-12-03 17:32:30 +0300224 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
Kent Overstreet54efd502015-04-23 22:37:18 -0700225 generic_make_request(*bio);
226 *bio = split;
227 }
228}
229EXPORT_SYMBOL(blk_queue_split);
230
Jens Axboe1e428072009-02-23 09:03:10 +0100231static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
Ming Lei07388542014-09-02 23:02:59 +0800232 struct bio *bio,
233 bool no_sg_merge)
Jens Axboed6d48192008-01-29 14:04:06 +0100234{
Kent Overstreet79886132013-11-23 17:19:00 -0800235 struct bio_vec bv, bvprv = { NULL };
Kent Overstreet54efd502015-04-23 22:37:18 -0700236 int cluster, prev = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100237 unsigned int seg_size, nr_phys_segs;
Jens Axboe59247ea2009-03-06 08:55:24 +0100238 struct bio *fbio, *bbio;
Kent Overstreet79886132013-11-23 17:19:00 -0800239 struct bvec_iter iter;
Jens Axboed6d48192008-01-29 14:04:06 +0100240
Jens Axboe1e428072009-02-23 09:03:10 +0100241 if (!bio)
242 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100243
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800244 switch (bio_op(bio)) {
245 case REQ_OP_DISCARD:
246 case REQ_OP_SECURE_ERASE:
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800247 case REQ_OP_WRITE_ZEROES:
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700248 return 0;
249 case REQ_OP_WRITE_SAME:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700250 return 1;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800251 }
Kent Overstreet5cb88502014-02-07 13:53:46 -0700252
Jens Axboe1e428072009-02-23 09:03:10 +0100253 fbio = bio;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100254 cluster = blk_queue_cluster(q);
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200255 seg_size = 0;
Andi Kleen2c8919d2010-06-21 11:02:47 +0200256 nr_phys_segs = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100257 for_each_bio(bio) {
Kent Overstreet79886132013-11-23 17:19:00 -0800258 bio_for_each_segment(bv, bio, iter) {
Jens Axboe1e428072009-02-23 09:03:10 +0100259 /*
Jens Axboe05f1dd52014-05-29 09:53:32 -0600260 * If SG merging is disabled, each bio vector is
261 * a segment
262 */
263 if (no_sg_merge)
264 goto new_segment;
265
Kent Overstreet54efd502015-04-23 22:37:18 -0700266 if (prev && cluster) {
Kent Overstreet79886132013-11-23 17:19:00 -0800267 if (seg_size + bv.bv_len
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400268 > queue_max_segment_size(q))
Jens Axboe1e428072009-02-23 09:03:10 +0100269 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800270 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100271 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800272 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100273 goto new_segment;
Jens Axboed6d48192008-01-29 14:04:06 +0100274
Kent Overstreet79886132013-11-23 17:19:00 -0800275 seg_size += bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100276 bvprv = bv;
277 continue;
278 }
Jens Axboed6d48192008-01-29 14:04:06 +0100279new_segment:
Jens Axboe1e428072009-02-23 09:03:10 +0100280 if (nr_phys_segs == 1 && seg_size >
281 fbio->bi_seg_front_size)
282 fbio->bi_seg_front_size = seg_size;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200283
Jens Axboe1e428072009-02-23 09:03:10 +0100284 nr_phys_segs++;
285 bvprv = bv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700286 prev = 1;
Kent Overstreet79886132013-11-23 17:19:00 -0800287 seg_size = bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100288 }
Jens Axboe59247ea2009-03-06 08:55:24 +0100289 bbio = bio;
Jens Axboed6d48192008-01-29 14:04:06 +0100290 }
291
Jens Axboe59247ea2009-03-06 08:55:24 +0100292 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
293 fbio->bi_seg_front_size = seg_size;
294 if (seg_size > bbio->bi_seg_back_size)
295 bbio->bi_seg_back_size = seg_size;
Jens Axboe1e428072009-02-23 09:03:10 +0100296
297 return nr_phys_segs;
298}
299
300void blk_recalc_rq_segments(struct request *rq)
301{
Ming Lei07388542014-09-02 23:02:59 +0800302 bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE,
303 &rq->q->queue_flags);
304
305 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio,
306 no_sg_merge);
Jens Axboed6d48192008-01-29 14:04:06 +0100307}
308
309void blk_recount_segments(struct request_queue *q, struct bio *bio)
310{
Ming Lei7f60dca2014-11-12 00:15:41 +0800311 unsigned short seg_cnt;
Ming Lei764f6122014-10-09 23:17:35 +0800312
Ming Lei7f60dca2014-11-12 00:15:41 +0800313 /* estimate segment number by bi_vcnt for non-cloned bio */
314 if (bio_flagged(bio, BIO_CLONED))
315 seg_cnt = bio_segments(bio);
316 else
317 seg_cnt = bio->bi_vcnt;
318
319 if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) &&
320 (seg_cnt < queue_max_segments(q)))
321 bio->bi_phys_segments = seg_cnt;
Jens Axboe05f1dd52014-05-29 09:53:32 -0600322 else {
323 struct bio *nxt = bio->bi_next;
Jens Axboe1e428072009-02-23 09:03:10 +0100324
Jens Axboe05f1dd52014-05-29 09:53:32 -0600325 bio->bi_next = NULL;
Ming Lei7f60dca2014-11-12 00:15:41 +0800326 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false);
Jens Axboe05f1dd52014-05-29 09:53:32 -0600327 bio->bi_next = nxt;
328 }
329
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600330 bio_set_flag(bio, BIO_SEG_VALID);
Jens Axboed6d48192008-01-29 14:04:06 +0100331}
332EXPORT_SYMBOL(blk_recount_segments);
333
334static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
335 struct bio *nxt)
336{
Kent Overstreet2b8221e2013-12-03 14:29:09 -0700337 struct bio_vec end_bv = { NULL }, nxt_bv;
Kent Overstreetf619d252013-08-07 14:30:33 -0700338
Martin K. Petersene692cb62010-12-01 19:41:49 +0100339 if (!blk_queue_cluster(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100340 return 0;
341
FUJITA Tomonori86771422008-10-13 14:19:05 +0200342 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400343 queue_max_segment_size(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100344 return 0;
345
David Woodhousee17fc0a2008-08-09 16:42:20 +0100346 if (!bio_has_data(bio))
347 return 1;
348
Ming Leie8270912016-02-26 23:40:53 +0800349 bio_get_last_bvec(bio, &end_bv);
350 bio_get_first_bvec(nxt, &nxt_bv);
Kent Overstreetf619d252013-08-07 14:30:33 -0700351
352 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100353 return 0;
354
Jens Axboed6d48192008-01-29 14:04:06 +0100355 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +0100356 * bio and nxt are contiguous in memory; check if the queue allows
Jens Axboed6d48192008-01-29 14:04:06 +0100357 * these two to be merged into one
358 */
Kent Overstreetf619d252013-08-07 14:30:33 -0700359 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
Jens Axboed6d48192008-01-29 14:04:06 +0100360 return 1;
361
362 return 0;
363}
364
Kent Overstreet79886132013-11-23 17:19:00 -0800365static inline void
Asias He963ab9e2012-08-02 23:42:03 +0200366__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
Kent Overstreet79886132013-11-23 17:19:00 -0800367 struct scatterlist *sglist, struct bio_vec *bvprv,
Asias He963ab9e2012-08-02 23:42:03 +0200368 struct scatterlist **sg, int *nsegs, int *cluster)
369{
370
371 int nbytes = bvec->bv_len;
372
Kent Overstreet79886132013-11-23 17:19:00 -0800373 if (*sg && *cluster) {
Asias He963ab9e2012-08-02 23:42:03 +0200374 if ((*sg)->length + nbytes > queue_max_segment_size(q))
375 goto new_segment;
376
Kent Overstreet79886132013-11-23 17:19:00 -0800377 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200378 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800379 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200380 goto new_segment;
381
382 (*sg)->length += nbytes;
383 } else {
384new_segment:
385 if (!*sg)
386 *sg = sglist;
387 else {
388 /*
389 * If the driver previously mapped a shorter
390 * list, we could see a termination bit
391 * prematurely unless it fully inits the sg
392 * table on each mapping. We KNOW that there
393 * must be more entries here or the driver
394 * would be buggy, so force clear the
395 * termination bit to avoid doing a full
396 * sg_init_table() in drivers for each command.
397 */
Paolo Bonzinic8164d82013-03-20 15:37:08 +1030398 sg_unmark_end(*sg);
Asias He963ab9e2012-08-02 23:42:03 +0200399 *sg = sg_next(*sg);
400 }
401
402 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
403 (*nsegs)++;
404 }
Kent Overstreet79886132013-11-23 17:19:00 -0800405 *bvprv = *bvec;
Asias He963ab9e2012-08-02 23:42:03 +0200406}
407
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700408static inline int __blk_bvec_map_sg(struct request_queue *q, struct bio_vec bv,
409 struct scatterlist *sglist, struct scatterlist **sg)
410{
411 *sg = sglist;
412 sg_set_page(*sg, bv.bv_page, bv.bv_len, bv.bv_offset);
413 return 1;
414}
415
Kent Overstreet5cb88502014-02-07 13:53:46 -0700416static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
417 struct scatterlist *sglist,
418 struct scatterlist **sg)
419{
420 struct bio_vec bvec, bvprv = { NULL };
421 struct bvec_iter iter;
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700422 int cluster = blk_queue_cluster(q), nsegs = 0;
Kent Overstreet5cb88502014-02-07 13:53:46 -0700423
424 for_each_bio(bio)
425 bio_for_each_segment(bvec, bio, iter)
426 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
427 &nsegs, &cluster);
428
429 return nsegs;
430}
431
Jens Axboed6d48192008-01-29 14:04:06 +0100432/*
433 * map a request to scatterlist, return number of sg entries setup. Caller
434 * must make sure sg can hold rq->nr_phys_segments entries
435 */
436int blk_rq_map_sg(struct request_queue *q, struct request *rq,
437 struct scatterlist *sglist)
438{
Kent Overstreet5cb88502014-02-07 13:53:46 -0700439 struct scatterlist *sg = NULL;
440 int nsegs = 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100441
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700442 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
443 nsegs = __blk_bvec_map_sg(q, rq->special_vec, sglist, &sg);
444 else if (rq->bio && bio_op(rq->bio) == REQ_OP_WRITE_SAME)
445 nsegs = __blk_bvec_map_sg(q, bio_iovec(rq->bio), sglist, &sg);
446 else if (rq->bio)
Kent Overstreet5cb88502014-02-07 13:53:46 -0700447 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200448
Christoph Hellwige8064022016-10-20 15:12:13 +0200449 if (unlikely(rq->rq_flags & RQF_COPY_USER) &&
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900450 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
451 unsigned int pad_len =
452 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200453
454 sg->length += pad_len;
455 rq->extra_len += pad_len;
456 }
457
Tejun Heo2fb98e82008-02-19 11:36:53 +0100458 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
Mike Christiea8ebb052016-06-05 14:31:45 -0500459 if (op_is_write(req_op(rq)))
Tejun Heodb0a2e02008-02-19 11:36:55 +0100460 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
461
Dan Williamsda81ed12015-08-07 18:15:14 +0200462 sg_unmark_end(sg);
Jens Axboed6d48192008-01-29 14:04:06 +0100463 sg = sg_next(sg);
464 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
465 q->dma_drain_size,
466 ((unsigned long)q->dma_drain_buffer) &
467 (PAGE_SIZE - 1));
468 nsegs++;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100469 rq->extra_len += q->dma_drain_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100470 }
471
472 if (sg)
473 sg_mark_end(sg);
474
Ming Lei12e57f52015-11-24 10:35:31 +0800475 /*
476 * Something must have been wrong if the figured number of
477 * segment is bigger than number of req's physical segments
478 */
Christoph Hellwigf9d03f92016-12-08 15:20:32 -0700479 WARN_ON(nsegs > blk_rq_nr_phys_segments(rq));
Ming Lei12e57f52015-11-24 10:35:31 +0800480
Jens Axboed6d48192008-01-29 14:04:06 +0100481 return nsegs;
482}
Jens Axboed6d48192008-01-29 14:04:06 +0100483EXPORT_SYMBOL(blk_rq_map_sg);
484
Jens Axboed6d48192008-01-29 14:04:06 +0100485static inline int ll_new_hw_segment(struct request_queue *q,
486 struct request *req,
487 struct bio *bio)
488{
Jens Axboed6d48192008-01-29 14:04:06 +0100489 int nr_phys_segs = bio_phys_segments(q, bio);
490
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200491 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
492 goto no_merge;
493
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400494 if (blk_integrity_merge_bio(q, req, bio) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200495 goto no_merge;
Jens Axboed6d48192008-01-29 14:04:06 +0100496
497 /*
498 * This will form the start of a new hw segment. Bump both
499 * counters.
500 */
Jens Axboed6d48192008-01-29 14:04:06 +0100501 req->nr_phys_segments += nr_phys_segs;
502 return 1;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200503
504no_merge:
Ritesh Harjanie0c72302016-12-01 08:36:16 -0700505 req_set_nomerge(q, req);
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200506 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100507}
508
509int ll_back_merge_fn(struct request_queue *q, struct request *req,
510 struct bio *bio)
511{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300512 if (req_gap_back_merge(req, bio))
513 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600514 if (blk_integrity_rq(req) &&
515 integrity_req_gap_back_merge(req, bio))
516 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400517 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600518 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
Ritesh Harjanie0c72302016-12-01 08:36:16 -0700519 req_set_nomerge(q, req);
Jens Axboed6d48192008-01-29 14:04:06 +0100520 return 0;
521 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200522 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100523 blk_recount_segments(q, req->biotail);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200524 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100525 blk_recount_segments(q, bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100526
527 return ll_new_hw_segment(q, req, bio);
528}
529
Jens Axboe6728cb02008-01-31 13:03:55 +0100530int ll_front_merge_fn(struct request_queue *q, struct request *req,
Jens Axboed6d48192008-01-29 14:04:06 +0100531 struct bio *bio)
532{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300533
534 if (req_gap_front_merge(req, bio))
535 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600536 if (blk_integrity_rq(req) &&
537 integrity_req_gap_front_merge(req, bio))
538 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400539 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600540 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
Ritesh Harjanie0c72302016-12-01 08:36:16 -0700541 req_set_nomerge(q, req);
Jens Axboed6d48192008-01-29 14:04:06 +0100542 return 0;
543 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200544 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100545 blk_recount_segments(q, bio);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200546 if (!bio_flagged(req->bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100547 blk_recount_segments(q, req->bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100548
549 return ll_new_hw_segment(q, req, bio);
550}
551
Jens Axboee7e24502013-10-29 12:11:47 -0600552/*
553 * blk-mq uses req->special to carry normal driver per-request payload, it
554 * does not indicate a prepared command that we cannot merge with.
555 */
556static bool req_no_special_merge(struct request *req)
557{
558 struct request_queue *q = req->q;
559
560 return !q->mq_ops && req->special;
561}
562
Jens Axboed6d48192008-01-29 14:04:06 +0100563static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
564 struct request *next)
565{
566 int total_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200567 unsigned int seg_size =
568 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100569
570 /*
571 * First check if the either of the requests are re-queued
572 * requests. Can't merge them if they are.
573 */
Jens Axboee7e24502013-10-29 12:11:47 -0600574 if (req_no_special_merge(req) || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100575 return 0;
576
Jens Axboe5e7c4272015-09-03 19:28:20 +0300577 if (req_gap_back_merge(req, next->bio))
Keith Busch854fbb92015-02-11 08:20:13 -0700578 return 0;
579
Jens Axboed6d48192008-01-29 14:04:06 +0100580 /*
581 * Will it become too large?
582 */
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400583 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600584 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
Jens Axboed6d48192008-01-29 14:04:06 +0100585 return 0;
586
587 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200588 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
589 if (req->nr_phys_segments == 1)
590 req->bio->bi_seg_front_size = seg_size;
591 if (next->nr_phys_segments == 1)
592 next->biotail->bi_seg_back_size = seg_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100593 total_phys_segments--;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200594 }
Jens Axboed6d48192008-01-29 14:04:06 +0100595
Martin K. Petersen8a783622010-02-26 00:20:39 -0500596 if (total_phys_segments > queue_max_segments(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100597 return 0;
598
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400599 if (blk_integrity_merge_rq(q, req, next) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200600 return 0;
601
Jens Axboed6d48192008-01-29 14:04:06 +0100602 /* Merge is OK... */
603 req->nr_phys_segments = total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100604 return 1;
605}
606
Tejun Heo80a761f2009-07-03 17:48:17 +0900607/**
608 * blk_rq_set_mixed_merge - mark a request as mixed merge
609 * @rq: request to mark as mixed merge
610 *
611 * Description:
612 * @rq is about to be mixed merged. Make sure the attributes
613 * which can be mixed are set in each bio and mark @rq as mixed
614 * merged.
615 */
616void blk_rq_set_mixed_merge(struct request *rq)
617{
618 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
619 struct bio *bio;
620
Christoph Hellwige8064022016-10-20 15:12:13 +0200621 if (rq->rq_flags & RQF_MIXED_MERGE)
Tejun Heo80a761f2009-07-03 17:48:17 +0900622 return;
623
624 /*
625 * @rq will no longer represent mixable attributes for all the
626 * contained bios. It will just track those of the first one.
627 * Distributes the attributs to each bio.
628 */
629 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600630 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
631 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
632 bio->bi_opf |= ff;
Tejun Heo80a761f2009-07-03 17:48:17 +0900633 }
Christoph Hellwige8064022016-10-20 15:12:13 +0200634 rq->rq_flags |= RQF_MIXED_MERGE;
Tejun Heo80a761f2009-07-03 17:48:17 +0900635}
636
Jerome Marchand26308ea2009-03-27 10:31:51 +0100637static void blk_account_io_merge(struct request *req)
638{
639 if (blk_do_io_stat(req)) {
640 struct hd_struct *part;
641 int cpu;
642
643 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100644 part = req->part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100645
646 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200647 part_dec_in_flight(part, rq_data_dir(req));
Jerome Marchand26308ea2009-03-27 10:31:51 +0100648
Jens Axboe6c23a962011-01-07 08:43:37 +0100649 hd_struct_put(part);
Jerome Marchand26308ea2009-03-27 10:31:51 +0100650 part_stat_unlock();
651 }
652}
653
Jens Axboed6d48192008-01-29 14:04:06 +0100654/*
Jens Axboeb973cb72017-02-02 08:54:40 -0700655 * For non-mq, this has to be called with the request spinlock acquired.
656 * For mq with scheduling, the appropriate queue wide lock should be held.
Jens Axboed6d48192008-01-29 14:04:06 +0100657 */
Jens Axboeb973cb72017-02-02 08:54:40 -0700658static struct request *attempt_merge(struct request_queue *q,
659 struct request *req, struct request *next)
Jens Axboed6d48192008-01-29 14:04:06 +0100660{
661 if (!rq_mergeable(req) || !rq_mergeable(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700662 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100663
Christoph Hellwig288dab82016-06-09 16:00:36 +0200664 if (req_op(req) != req_op(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700665 return NULL;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400666
Jens Axboed6d48192008-01-29 14:04:06 +0100667 /*
668 * not contiguous
669 */
Tejun Heo83096eb2009-05-07 22:24:39 +0900670 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700671 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100672
673 if (rq_data_dir(req) != rq_data_dir(next)
674 || req->rq_disk != next->rq_disk
Jens Axboee7e24502013-10-29 12:11:47 -0600675 || req_no_special_merge(next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700676 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100677
Mike Christie8fe0d472016-06-05 14:32:15 -0500678 if (req_op(req) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400679 !blk_write_same_mergeable(req->bio, next->bio))
Jens Axboeb973cb72017-02-02 08:54:40 -0700680 return NULL;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400681
Jens Axboed6d48192008-01-29 14:04:06 +0100682 /*
683 * If we are allowed to merge, then append bio list
684 * from next to rq and release next. merge_requests_fn
685 * will have updated segment counts, update sector
686 * counts here.
687 */
688 if (!ll_merge_requests_fn(q, req, next))
Jens Axboeb973cb72017-02-02 08:54:40 -0700689 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100690
691 /*
Tejun Heo80a761f2009-07-03 17:48:17 +0900692 * If failfast settings disagree or any of the two is already
693 * a mixed merge, mark both as mixed before proceeding. This
694 * makes sure that all involved bios have mixable attributes
695 * set properly.
696 */
Christoph Hellwige8064022016-10-20 15:12:13 +0200697 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
Tejun Heo80a761f2009-07-03 17:48:17 +0900698 (req->cmd_flags & REQ_FAILFAST_MASK) !=
699 (next->cmd_flags & REQ_FAILFAST_MASK)) {
700 blk_rq_set_mixed_merge(req);
701 blk_rq_set_mixed_merge(next);
702 }
703
704 /*
Jens Axboed6d48192008-01-29 14:04:06 +0100705 * At this point we have either done a back merge
706 * or front merge. We need the smaller start_time of
707 * the merged requests to be the current request
708 * for accounting purposes.
709 */
710 if (time_after(req->start_time, next->start_time))
711 req->start_time = next->start_time;
712
713 req->biotail->bi_next = next->bio;
714 req->biotail = next->biotail;
715
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900716 req->__data_len += blk_rq_bytes(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100717
718 elv_merge_requests(q, req, next);
719
Jerome Marchand42dad762009-04-22 14:01:49 +0200720 /*
721 * 'next' is going away, so update stats accordingly
722 */
723 blk_account_io_merge(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100724
725 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
Jens Axboeab780f12008-08-26 10:25:02 +0200726 if (blk_rq_cpu_valid(next))
727 req->cpu = next->cpu;
Jens Axboed6d48192008-01-29 14:04:06 +0100728
Jens Axboee4d750c2017-02-03 09:48:28 -0700729 /*
730 * ownership of bio passed from next to req, return 'next' for
731 * the caller to free
732 */
Boaz Harrosh1cd96c22009-03-24 12:35:07 +0100733 next->bio = NULL;
Jens Axboeb973cb72017-02-02 08:54:40 -0700734 return next;
Jens Axboed6d48192008-01-29 14:04:06 +0100735}
736
Jens Axboeb973cb72017-02-02 08:54:40 -0700737struct request *attempt_back_merge(struct request_queue *q, struct request *rq)
Jens Axboed6d48192008-01-29 14:04:06 +0100738{
739 struct request *next = elv_latter_request(q, rq);
740
741 if (next)
742 return attempt_merge(q, rq, next);
743
Jens Axboeb973cb72017-02-02 08:54:40 -0700744 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100745}
746
Jens Axboeb973cb72017-02-02 08:54:40 -0700747struct request *attempt_front_merge(struct request_queue *q, struct request *rq)
Jens Axboed6d48192008-01-29 14:04:06 +0100748{
749 struct request *prev = elv_former_request(q, rq);
750
751 if (prev)
752 return attempt_merge(q, prev, rq);
753
Jens Axboeb973cb72017-02-02 08:54:40 -0700754 return NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100755}
Jens Axboe5e84ea32011-03-21 10:14:27 +0100756
757int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
758 struct request *next)
759{
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700760 struct elevator_queue *e = q->elevator;
Jens Axboee4d750c2017-02-03 09:48:28 -0700761 struct request *free;
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700762
Jens Axboebd166ef2017-01-17 06:03:22 -0700763 if (!e->uses_mq && e->type->ops.sq.elevator_allow_rq_merge_fn)
Jens Axboec51ca6c2016-12-10 15:13:59 -0700764 if (!e->type->ops.sq.elevator_allow_rq_merge_fn(q, rq, next))
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700765 return 0;
766
Jens Axboee4d750c2017-02-03 09:48:28 -0700767 free = attempt_merge(q, rq, next);
768 if (free) {
769 __blk_put_request(q, free);
770 return 1;
771 }
772
773 return 0;
Jens Axboe5e84ea32011-03-21 10:14:27 +0100774}
Tejun Heo050c8ea2012-02-08 09:19:38 +0100775
776bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
777{
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400778 if (!rq_mergeable(rq) || !bio_mergeable(bio))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100779 return false;
780
Christoph Hellwig288dab82016-06-09 16:00:36 +0200781 if (req_op(rq) != bio_op(bio))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400782 return false;
783
Tejun Heo050c8ea2012-02-08 09:19:38 +0100784 /* different data direction or already started, don't merge */
785 if (bio_data_dir(bio) != rq_data_dir(rq))
786 return false;
787
788 /* must be same device and not a special request */
Jens Axboee7e24502013-10-29 12:11:47 -0600789 if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100790 return false;
791
792 /* only merge integrity protected bio into ditto rq */
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400793 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100794 return false;
795
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400796 /* must be using the same buffer */
Mike Christie8fe0d472016-06-05 14:32:15 -0500797 if (req_op(rq) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400798 !blk_write_same_mergeable(rq->bio, bio))
799 return false;
800
Tejun Heo050c8ea2012-02-08 09:19:38 +0100801 return true;
802}
803
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100804enum elv_merge blk_try_merge(struct request *rq, struct bio *bio)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100805{
Christoph Hellwig1e739732017-02-08 14:46:49 +0100806 if (req_op(rq) == REQ_OP_DISCARD &&
807 queue_max_discard_segments(rq->q) > 1)
808 return ELEVATOR_DISCARD_MERGE;
809 else if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100810 return ELEVATOR_BACK_MERGE;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700811 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100812 return ELEVATOR_FRONT_MERGE;
813 return ELEVATOR_NO_MERGE;
814}