blob: abde3707438d80062aaf5b56c220609bce4feb63 [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;
202 case REQ_OP_WRITE_SAME:
Ming Leibdced432015-10-20 23:13:52 +0800203 split = blk_bio_write_same_split(q, *bio, bs, &nsegs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300204 break;
205 default:
Ming Leibdced432015-10-20 23:13:52 +0800206 split = blk_bio_segment_split(q, *bio, q->bio_split, &nsegs);
Adrian Hunter7afafc82016-08-16 10:59:35 +0300207 break;
208 }
Ming Leibdced432015-10-20 23:13:52 +0800209
210 /* physical segments can be figured out during splitting */
211 res = split ? split : *bio;
212 res->bi_phys_segments = nsegs;
213 bio_set_flag(res, BIO_SEG_VALID);
Kent Overstreet54efd502015-04-23 22:37:18 -0700214
215 if (split) {
Ming Lei6ac45ae2015-10-20 23:13:53 +0800216 /* there isn't chance to merge the splitted bio */
Jens Axboe1eff9d32016-08-05 15:35:16 -0600217 split->bi_opf |= REQ_NOMERGE;
Ming Lei6ac45ae2015-10-20 23:13:53 +0800218
Kent Overstreet54efd502015-04-23 22:37:18 -0700219 bio_chain(split, *bio);
Mike Krinkincda22642015-12-03 17:32:30 +0300220 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
Kent Overstreet54efd502015-04-23 22:37:18 -0700221 generic_make_request(*bio);
222 *bio = split;
223 }
224}
225EXPORT_SYMBOL(blk_queue_split);
226
Jens Axboe1e428072009-02-23 09:03:10 +0100227static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
Ming Lei07388542014-09-02 23:02:59 +0800228 struct bio *bio,
229 bool no_sg_merge)
Jens Axboed6d48192008-01-29 14:04:06 +0100230{
Kent Overstreet79886132013-11-23 17:19:00 -0800231 struct bio_vec bv, bvprv = { NULL };
Kent Overstreet54efd502015-04-23 22:37:18 -0700232 int cluster, prev = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100233 unsigned int seg_size, nr_phys_segs;
Jens Axboe59247ea2009-03-06 08:55:24 +0100234 struct bio *fbio, *bbio;
Kent Overstreet79886132013-11-23 17:19:00 -0800235 struct bvec_iter iter;
Jens Axboed6d48192008-01-29 14:04:06 +0100236
Jens Axboe1e428072009-02-23 09:03:10 +0100237 if (!bio)
238 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100239
Kent Overstreet5cb88502014-02-07 13:53:46 -0700240 /*
241 * This should probably be returning 0, but blk_add_request_payload()
242 * (Christoph!!!!)
243 */
Adrian Hunter7afafc82016-08-16 10:59:35 +0300244 if (bio_op(bio) == REQ_OP_DISCARD || bio_op(bio) == REQ_OP_SECURE_ERASE)
Kent Overstreet5cb88502014-02-07 13:53:46 -0700245 return 1;
246
Mike Christie95fe6c12016-06-05 14:31:48 -0500247 if (bio_op(bio) == REQ_OP_WRITE_SAME)
Kent Overstreet5cb88502014-02-07 13:53:46 -0700248 return 1;
249
Jens Axboe1e428072009-02-23 09:03:10 +0100250 fbio = bio;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100251 cluster = blk_queue_cluster(q);
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200252 seg_size = 0;
Andi Kleen2c8919d2010-06-21 11:02:47 +0200253 nr_phys_segs = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100254 for_each_bio(bio) {
Kent Overstreet79886132013-11-23 17:19:00 -0800255 bio_for_each_segment(bv, bio, iter) {
Jens Axboe1e428072009-02-23 09:03:10 +0100256 /*
Jens Axboe05f1dd52014-05-29 09:53:32 -0600257 * If SG merging is disabled, each bio vector is
258 * a segment
259 */
260 if (no_sg_merge)
261 goto new_segment;
262
Kent Overstreet54efd502015-04-23 22:37:18 -0700263 if (prev && cluster) {
Kent Overstreet79886132013-11-23 17:19:00 -0800264 if (seg_size + bv.bv_len
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400265 > queue_max_segment_size(q))
Jens Axboe1e428072009-02-23 09:03:10 +0100266 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800267 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100268 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800269 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100270 goto new_segment;
Jens Axboed6d48192008-01-29 14:04:06 +0100271
Kent Overstreet79886132013-11-23 17:19:00 -0800272 seg_size += bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100273 bvprv = bv;
274 continue;
275 }
Jens Axboed6d48192008-01-29 14:04:06 +0100276new_segment:
Jens Axboe1e428072009-02-23 09:03:10 +0100277 if (nr_phys_segs == 1 && seg_size >
278 fbio->bi_seg_front_size)
279 fbio->bi_seg_front_size = seg_size;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200280
Jens Axboe1e428072009-02-23 09:03:10 +0100281 nr_phys_segs++;
282 bvprv = bv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700283 prev = 1;
Kent Overstreet79886132013-11-23 17:19:00 -0800284 seg_size = bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100285 }
Jens Axboe59247ea2009-03-06 08:55:24 +0100286 bbio = bio;
Jens Axboed6d48192008-01-29 14:04:06 +0100287 }
288
Jens Axboe59247ea2009-03-06 08:55:24 +0100289 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
290 fbio->bi_seg_front_size = seg_size;
291 if (seg_size > bbio->bi_seg_back_size)
292 bbio->bi_seg_back_size = seg_size;
Jens Axboe1e428072009-02-23 09:03:10 +0100293
294 return nr_phys_segs;
295}
296
297void blk_recalc_rq_segments(struct request *rq)
298{
Ming Lei07388542014-09-02 23:02:59 +0800299 bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE,
300 &rq->q->queue_flags);
301
302 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio,
303 no_sg_merge);
Jens Axboed6d48192008-01-29 14:04:06 +0100304}
305
306void blk_recount_segments(struct request_queue *q, struct bio *bio)
307{
Ming Lei7f60dca2014-11-12 00:15:41 +0800308 unsigned short seg_cnt;
Ming Lei764f6122014-10-09 23:17:35 +0800309
Ming Lei7f60dca2014-11-12 00:15:41 +0800310 /* estimate segment number by bi_vcnt for non-cloned bio */
311 if (bio_flagged(bio, BIO_CLONED))
312 seg_cnt = bio_segments(bio);
313 else
314 seg_cnt = bio->bi_vcnt;
315
316 if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) &&
317 (seg_cnt < queue_max_segments(q)))
318 bio->bi_phys_segments = seg_cnt;
Jens Axboe05f1dd52014-05-29 09:53:32 -0600319 else {
320 struct bio *nxt = bio->bi_next;
Jens Axboe1e428072009-02-23 09:03:10 +0100321
Jens Axboe05f1dd52014-05-29 09:53:32 -0600322 bio->bi_next = NULL;
Ming Lei7f60dca2014-11-12 00:15:41 +0800323 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false);
Jens Axboe05f1dd52014-05-29 09:53:32 -0600324 bio->bi_next = nxt;
325 }
326
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600327 bio_set_flag(bio, BIO_SEG_VALID);
Jens Axboed6d48192008-01-29 14:04:06 +0100328}
329EXPORT_SYMBOL(blk_recount_segments);
330
331static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
332 struct bio *nxt)
333{
Kent Overstreet2b8221e2013-12-03 14:29:09 -0700334 struct bio_vec end_bv = { NULL }, nxt_bv;
Kent Overstreetf619d252013-08-07 14:30:33 -0700335
Martin K. Petersene692cb62010-12-01 19:41:49 +0100336 if (!blk_queue_cluster(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100337 return 0;
338
FUJITA Tomonori86771422008-10-13 14:19:05 +0200339 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400340 queue_max_segment_size(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100341 return 0;
342
David Woodhousee17fc0a2008-08-09 16:42:20 +0100343 if (!bio_has_data(bio))
344 return 1;
345
Ming Leie8270912016-02-26 23:40:53 +0800346 bio_get_last_bvec(bio, &end_bv);
347 bio_get_first_bvec(nxt, &nxt_bv);
Kent Overstreetf619d252013-08-07 14:30:33 -0700348
349 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100350 return 0;
351
Jens Axboed6d48192008-01-29 14:04:06 +0100352 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +0100353 * bio and nxt are contiguous in memory; check if the queue allows
Jens Axboed6d48192008-01-29 14:04:06 +0100354 * these two to be merged into one
355 */
Kent Overstreetf619d252013-08-07 14:30:33 -0700356 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
Jens Axboed6d48192008-01-29 14:04:06 +0100357 return 1;
358
359 return 0;
360}
361
Kent Overstreet79886132013-11-23 17:19:00 -0800362static inline void
Asias He963ab9e2012-08-02 23:42:03 +0200363__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
Kent Overstreet79886132013-11-23 17:19:00 -0800364 struct scatterlist *sglist, struct bio_vec *bvprv,
Asias He963ab9e2012-08-02 23:42:03 +0200365 struct scatterlist **sg, int *nsegs, int *cluster)
366{
367
368 int nbytes = bvec->bv_len;
369
Kent Overstreet79886132013-11-23 17:19:00 -0800370 if (*sg && *cluster) {
Asias He963ab9e2012-08-02 23:42:03 +0200371 if ((*sg)->length + nbytes > queue_max_segment_size(q))
372 goto new_segment;
373
Kent Overstreet79886132013-11-23 17:19:00 -0800374 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200375 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800376 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200377 goto new_segment;
378
379 (*sg)->length += nbytes;
380 } else {
381new_segment:
382 if (!*sg)
383 *sg = sglist;
384 else {
385 /*
386 * If the driver previously mapped a shorter
387 * list, we could see a termination bit
388 * prematurely unless it fully inits the sg
389 * table on each mapping. We KNOW that there
390 * must be more entries here or the driver
391 * would be buggy, so force clear the
392 * termination bit to avoid doing a full
393 * sg_init_table() in drivers for each command.
394 */
Paolo Bonzinic8164d82013-03-20 15:37:08 +1030395 sg_unmark_end(*sg);
Asias He963ab9e2012-08-02 23:42:03 +0200396 *sg = sg_next(*sg);
397 }
398
399 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
400 (*nsegs)++;
401 }
Kent Overstreet79886132013-11-23 17:19:00 -0800402 *bvprv = *bvec;
Asias He963ab9e2012-08-02 23:42:03 +0200403}
404
Kent Overstreet5cb88502014-02-07 13:53:46 -0700405static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
406 struct scatterlist *sglist,
407 struct scatterlist **sg)
408{
409 struct bio_vec bvec, bvprv = { NULL };
410 struct bvec_iter iter;
411 int nsegs, cluster;
412
413 nsegs = 0;
414 cluster = blk_queue_cluster(q);
415
Adrian Hunter7afafc82016-08-16 10:59:35 +0300416 switch (bio_op(bio)) {
417 case REQ_OP_DISCARD:
418 case REQ_OP_SECURE_ERASE:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700419 /*
420 * This is a hack - drivers should be neither modifying the
421 * biovec, nor relying on bi_vcnt - but because of
422 * blk_add_request_payload(), a discard bio may or may not have
423 * a payload we need to set up here (thank you Christoph) and
424 * bi_vcnt is really the only way of telling if we need to.
425 */
Adrian Hunter7afafc82016-08-16 10:59:35 +0300426 if (!bio->bi_vcnt)
427 return 0;
428 /* Fall through */
429 case REQ_OP_WRITE_SAME:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700430 *sg = sglist;
431 bvec = bio_iovec(bio);
432 sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset);
433 return 1;
Adrian Hunter7afafc82016-08-16 10:59:35 +0300434 default:
435 break;
Kent Overstreet5cb88502014-02-07 13:53:46 -0700436 }
437
438 for_each_bio(bio)
439 bio_for_each_segment(bvec, bio, iter)
440 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
441 &nsegs, &cluster);
442
443 return nsegs;
444}
445
Jens Axboed6d48192008-01-29 14:04:06 +0100446/*
447 * map a request to scatterlist, return number of sg entries setup. Caller
448 * must make sure sg can hold rq->nr_phys_segments entries
449 */
450int blk_rq_map_sg(struct request_queue *q, struct request *rq,
451 struct scatterlist *sglist)
452{
Kent Overstreet5cb88502014-02-07 13:53:46 -0700453 struct scatterlist *sg = NULL;
454 int nsegs = 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100455
Kent Overstreet5cb88502014-02-07 13:53:46 -0700456 if (rq->bio)
457 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200458
459 if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900460 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
461 unsigned int pad_len =
462 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200463
464 sg->length += pad_len;
465 rq->extra_len += pad_len;
466 }
467
Tejun Heo2fb98e82008-02-19 11:36:53 +0100468 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
Mike Christiea8ebb052016-06-05 14:31:45 -0500469 if (op_is_write(req_op(rq)))
Tejun Heodb0a2e02008-02-19 11:36:55 +0100470 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
471
Dan Williamsda81ed12015-08-07 18:15:14 +0200472 sg_unmark_end(sg);
Jens Axboed6d48192008-01-29 14:04:06 +0100473 sg = sg_next(sg);
474 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
475 q->dma_drain_size,
476 ((unsigned long)q->dma_drain_buffer) &
477 (PAGE_SIZE - 1));
478 nsegs++;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100479 rq->extra_len += q->dma_drain_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100480 }
481
482 if (sg)
483 sg_mark_end(sg);
484
Ming Lei12e57f52015-11-24 10:35:31 +0800485 /*
486 * Something must have been wrong if the figured number of
487 * segment is bigger than number of req's physical segments
488 */
489 WARN_ON(nsegs > rq->nr_phys_segments);
490
Jens Axboed6d48192008-01-29 14:04:06 +0100491 return nsegs;
492}
Jens Axboed6d48192008-01-29 14:04:06 +0100493EXPORT_SYMBOL(blk_rq_map_sg);
494
AnilKumar Chimata7214d7e2017-06-23 03:09:59 -0700495/*
496 * map a request to scatterlist without combining PHY CONT
497 * blocks, return number of sg entries setup. Caller
498 * must make sure sg can hold rq->nr_phys_segments entries
499 */
500int blk_rq_map_sg_no_cluster(struct request_queue *q, struct request *rq,
501 struct scatterlist *sglist)
502{
503 struct bio_vec bvec, bvprv = { NULL };
504 struct req_iterator iter;
505 struct scatterlist *sg;
506 int nsegs, cluster = 0;
507
508 nsegs = 0;
509
510 /*
511 * for each bio in rq
512 */
513 sg = NULL;
514 rq_for_each_segment(bvec, rq, iter) {
515 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, &sg,
516 &nsegs, &cluster);
517 } /* segments in rq */
518
519
520 if (!sg)
521 return nsegs;
522
523 if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
524 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
525 unsigned int pad_len =
526 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
527
528 sg->length += pad_len;
529 rq->extra_len += pad_len;
530 }
531
532 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
533 if (rq->cmd_flags & REQ_OP_WRITE)
534 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
535
536 sg->page_link &= ~0x02;
537 sg = sg_next(sg);
538 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
539 q->dma_drain_size,
540 ((unsigned long)q->dma_drain_buffer) &
541 (PAGE_SIZE - 1));
542 nsegs++;
543 rq->extra_len += q->dma_drain_size;
544 }
545
546 if (sg)
547 sg_mark_end(sg);
548
549 return nsegs;
550}
551EXPORT_SYMBOL(blk_rq_map_sg_no_cluster);
552
Jens Axboed6d48192008-01-29 14:04:06 +0100553static inline int ll_new_hw_segment(struct request_queue *q,
554 struct request *req,
555 struct bio *bio)
556{
Jens Axboed6d48192008-01-29 14:04:06 +0100557 int nr_phys_segs = bio_phys_segments(q, bio);
558
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200559 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
560 goto no_merge;
561
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400562 if (blk_integrity_merge_bio(q, req, bio) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200563 goto no_merge;
Jens Axboed6d48192008-01-29 14:04:06 +0100564
565 /*
566 * This will form the start of a new hw segment. Bump both
567 * counters.
568 */
Jens Axboed6d48192008-01-29 14:04:06 +0100569 req->nr_phys_segments += nr_phys_segs;
570 return 1;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200571
572no_merge:
573 req->cmd_flags |= REQ_NOMERGE;
574 if (req == q->last_merge)
575 q->last_merge = NULL;
576 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100577}
578
579int ll_back_merge_fn(struct request_queue *q, struct request *req,
580 struct bio *bio)
581{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300582 if (req_gap_back_merge(req, bio))
583 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600584 if (blk_integrity_rq(req) &&
585 integrity_req_gap_back_merge(req, bio))
586 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400587 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600588 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
Jens Axboed6d48192008-01-29 14:04:06 +0100589 req->cmd_flags |= REQ_NOMERGE;
590 if (req == q->last_merge)
591 q->last_merge = NULL;
592 return 0;
593 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200594 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100595 blk_recount_segments(q, req->biotail);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200596 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100597 blk_recount_segments(q, bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100598
599 return ll_new_hw_segment(q, req, bio);
600}
601
Jens Axboe6728cb02008-01-31 13:03:55 +0100602int ll_front_merge_fn(struct request_queue *q, struct request *req,
Jens Axboed6d48192008-01-29 14:04:06 +0100603 struct bio *bio)
604{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300605
606 if (req_gap_front_merge(req, bio))
607 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600608 if (blk_integrity_rq(req) &&
609 integrity_req_gap_front_merge(req, bio))
610 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400611 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600612 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
Jens Axboed6d48192008-01-29 14:04:06 +0100613 req->cmd_flags |= REQ_NOMERGE;
614 if (req == q->last_merge)
615 q->last_merge = NULL;
616 return 0;
617 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200618 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100619 blk_recount_segments(q, bio);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200620 if (!bio_flagged(req->bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100621 blk_recount_segments(q, req->bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100622
623 return ll_new_hw_segment(q, req, bio);
624}
625
Jens Axboee7e24502013-10-29 12:11:47 -0600626/*
627 * blk-mq uses req->special to carry normal driver per-request payload, it
628 * does not indicate a prepared command that we cannot merge with.
629 */
630static bool req_no_special_merge(struct request *req)
631{
632 struct request_queue *q = req->q;
633
634 return !q->mq_ops && req->special;
635}
636
Jens Axboed6d48192008-01-29 14:04:06 +0100637static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
638 struct request *next)
639{
640 int total_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200641 unsigned int seg_size =
642 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100643
644 /*
645 * First check if the either of the requests are re-queued
646 * requests. Can't merge them if they are.
647 */
Jens Axboee7e24502013-10-29 12:11:47 -0600648 if (req_no_special_merge(req) || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100649 return 0;
650
Jens Axboe5e7c4272015-09-03 19:28:20 +0300651 if (req_gap_back_merge(req, next->bio))
Keith Busch854fbb92015-02-11 08:20:13 -0700652 return 0;
653
Jens Axboed6d48192008-01-29 14:04:06 +0100654 /*
655 * Will it become too large?
656 */
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400657 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600658 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
Jens Axboed6d48192008-01-29 14:04:06 +0100659 return 0;
660
661 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200662 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
663 if (req->nr_phys_segments == 1)
664 req->bio->bi_seg_front_size = seg_size;
665 if (next->nr_phys_segments == 1)
666 next->biotail->bi_seg_back_size = seg_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100667 total_phys_segments--;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200668 }
Jens Axboed6d48192008-01-29 14:04:06 +0100669
Martin K. Petersen8a783622010-02-26 00:20:39 -0500670 if (total_phys_segments > queue_max_segments(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100671 return 0;
672
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400673 if (blk_integrity_merge_rq(q, req, next) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200674 return 0;
675
Jens Axboed6d48192008-01-29 14:04:06 +0100676 /* Merge is OK... */
677 req->nr_phys_segments = total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100678 return 1;
679}
680
Tejun Heo80a761f2009-07-03 17:48:17 +0900681/**
682 * blk_rq_set_mixed_merge - mark a request as mixed merge
683 * @rq: request to mark as mixed merge
684 *
685 * Description:
686 * @rq is about to be mixed merged. Make sure the attributes
687 * which can be mixed are set in each bio and mark @rq as mixed
688 * merged.
689 */
690void blk_rq_set_mixed_merge(struct request *rq)
691{
692 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
693 struct bio *bio;
694
695 if (rq->cmd_flags & REQ_MIXED_MERGE)
696 return;
697
698 /*
699 * @rq will no longer represent mixable attributes for all the
700 * contained bios. It will just track those of the first one.
701 * Distributes the attributs to each bio.
702 */
703 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600704 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
705 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
706 bio->bi_opf |= ff;
Tejun Heo80a761f2009-07-03 17:48:17 +0900707 }
708 rq->cmd_flags |= REQ_MIXED_MERGE;
709}
710
Jerome Marchand26308ea2009-03-27 10:31:51 +0100711static void blk_account_io_merge(struct request *req)
712{
713 if (blk_do_io_stat(req)) {
714 struct hd_struct *part;
715 int cpu;
716
717 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100718 part = req->part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100719
720 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200721 part_dec_in_flight(part, rq_data_dir(req));
Jerome Marchand26308ea2009-03-27 10:31:51 +0100722
Jens Axboe6c23a962011-01-07 08:43:37 +0100723 hd_struct_put(part);
Jerome Marchand26308ea2009-03-27 10:31:51 +0100724 part_stat_unlock();
725 }
726}
727
Jens Axboed6d48192008-01-29 14:04:06 +0100728/*
729 * Has to be called with the request spinlock acquired
730 */
731static int attempt_merge(struct request_queue *q, struct request *req,
732 struct request *next)
733{
734 if (!rq_mergeable(req) || !rq_mergeable(next))
735 return 0;
736
Christoph Hellwig288dab82016-06-09 16:00:36 +0200737 if (req_op(req) != req_op(next))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400738 return 0;
739
Jens Axboed6d48192008-01-29 14:04:06 +0100740 /*
741 * not contiguous
742 */
Tejun Heo83096eb2009-05-07 22:24:39 +0900743 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100744 return 0;
745
746 if (rq_data_dir(req) != rq_data_dir(next)
747 || req->rq_disk != next->rq_disk
Jens Axboee7e24502013-10-29 12:11:47 -0600748 || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100749 return 0;
750
Mike Christie8fe0d472016-06-05 14:32:15 -0500751 if (req_op(req) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400752 !blk_write_same_mergeable(req->bio, next->bio))
753 return 0;
754
Jens Axboed6d48192008-01-29 14:04:06 +0100755 /*
756 * If we are allowed to merge, then append bio list
757 * from next to rq and release next. merge_requests_fn
758 * will have updated segment counts, update sector
759 * counts here.
760 */
761 if (!ll_merge_requests_fn(q, req, next))
762 return 0;
763
764 /*
Tejun Heo80a761f2009-07-03 17:48:17 +0900765 * If failfast settings disagree or any of the two is already
766 * a mixed merge, mark both as mixed before proceeding. This
767 * makes sure that all involved bios have mixable attributes
768 * set properly.
769 */
770 if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
771 (req->cmd_flags & REQ_FAILFAST_MASK) !=
772 (next->cmd_flags & REQ_FAILFAST_MASK)) {
773 blk_rq_set_mixed_merge(req);
774 blk_rq_set_mixed_merge(next);
775 }
776
777 /*
Jens Axboed6d48192008-01-29 14:04:06 +0100778 * At this point we have either done a back merge
779 * or front merge. We need the smaller start_time of
780 * the merged requests to be the current request
781 * for accounting purposes.
782 */
783 if (time_after(req->start_time, next->start_time))
784 req->start_time = next->start_time;
785
786 req->biotail->bi_next = next->bio;
787 req->biotail = next->biotail;
788
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900789 req->__data_len += blk_rq_bytes(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100790
791 elv_merge_requests(q, req, next);
792
Jerome Marchand42dad762009-04-22 14:01:49 +0200793 /*
794 * 'next' is going away, so update stats accordingly
795 */
796 blk_account_io_merge(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100797
798 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
Jens Axboeab780f12008-08-26 10:25:02 +0200799 if (blk_rq_cpu_valid(next))
800 req->cpu = next->cpu;
Jens Axboed6d48192008-01-29 14:04:06 +0100801
Boaz Harrosh1cd96c22009-03-24 12:35:07 +0100802 /* owner-ship of bio passed from next to req */
803 next->bio = NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100804 __blk_put_request(q, next);
805 return 1;
806}
807
808int attempt_back_merge(struct request_queue *q, struct request *rq)
809{
810 struct request *next = elv_latter_request(q, rq);
811
812 if (next)
813 return attempt_merge(q, rq, next);
814
815 return 0;
816}
817
818int attempt_front_merge(struct request_queue *q, struct request *rq)
819{
820 struct request *prev = elv_former_request(q, rq);
821
822 if (prev)
823 return attempt_merge(q, prev, rq);
824
825 return 0;
826}
Jens Axboe5e84ea32011-03-21 10:14:27 +0100827
828int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
829 struct request *next)
830{
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700831 struct elevator_queue *e = q->elevator;
832
833 if (e->type->ops.elevator_allow_rq_merge_fn)
834 if (!e->type->ops.elevator_allow_rq_merge_fn(q, rq, next))
835 return 0;
836
Jens Axboe5e84ea32011-03-21 10:14:27 +0100837 return attempt_merge(q, rq, next);
838}
Tejun Heo050c8ea2012-02-08 09:19:38 +0100839
840bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
841{
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400842 if (!rq_mergeable(rq) || !bio_mergeable(bio))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100843 return false;
844
Christoph Hellwig288dab82016-06-09 16:00:36 +0200845 if (req_op(rq) != bio_op(bio))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400846 return false;
847
Tejun Heo050c8ea2012-02-08 09:19:38 +0100848 /* different data direction or already started, don't merge */
849 if (bio_data_dir(bio) != rq_data_dir(rq))
850 return false;
851
852 /* must be same device and not a special request */
Jens Axboee7e24502013-10-29 12:11:47 -0600853 if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100854 return false;
855
856 /* only merge integrity protected bio into ditto rq */
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400857 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100858 return false;
859
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400860 /* must be using the same buffer */
Mike Christie8fe0d472016-06-05 14:32:15 -0500861 if (req_op(rq) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400862 !blk_write_same_mergeable(rq->bio, bio))
863 return false;
864
Tejun Heo050c8ea2012-02-08 09:19:38 +0100865 return true;
866}
867
868int blk_try_merge(struct request *rq, struct bio *bio)
869{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700870 if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100871 return ELEVATOR_BACK_MERGE;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700872 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100873 return ELEVATOR_FRONT_MERGE;
874 return ELEVATOR_NO_MERGE;
875}