blob: cf2848cb91d8037d9eccdb4d0fdf84b2563975e9 [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
Kent Overstreet5cb88502014-02-07 13:53:46 -0700244 /*
245 * This should probably be returning 0, but blk_add_request_payload()
246 * (Christoph!!!!)
247 */
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800248 switch (bio_op(bio)) {
249 case REQ_OP_DISCARD:
250 case REQ_OP_SECURE_ERASE:
251 case REQ_OP_WRITE_SAME:
252 case REQ_OP_WRITE_ZEROES:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700253 return 1;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800254 default:
255 break;
256 }
Kent Overstreet5cb88502014-02-07 13:53:46 -0700257
Jens Axboe1e428072009-02-23 09:03:10 +0100258 fbio = bio;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100259 cluster = blk_queue_cluster(q);
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200260 seg_size = 0;
Andi Kleen2c8919d2010-06-21 11:02:47 +0200261 nr_phys_segs = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100262 for_each_bio(bio) {
Kent Overstreet79886132013-11-23 17:19:00 -0800263 bio_for_each_segment(bv, bio, iter) {
Jens Axboe1e428072009-02-23 09:03:10 +0100264 /*
Jens Axboe05f1dd52014-05-29 09:53:32 -0600265 * If SG merging is disabled, each bio vector is
266 * a segment
267 */
268 if (no_sg_merge)
269 goto new_segment;
270
Kent Overstreet54efd502015-04-23 22:37:18 -0700271 if (prev && cluster) {
Kent Overstreet79886132013-11-23 17:19:00 -0800272 if (seg_size + bv.bv_len
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400273 > queue_max_segment_size(q))
Jens Axboe1e428072009-02-23 09:03:10 +0100274 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800275 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100276 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800277 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100278 goto new_segment;
Jens Axboed6d48192008-01-29 14:04:06 +0100279
Kent Overstreet79886132013-11-23 17:19:00 -0800280 seg_size += bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100281 bvprv = bv;
282 continue;
283 }
Jens Axboed6d48192008-01-29 14:04:06 +0100284new_segment:
Jens Axboe1e428072009-02-23 09:03:10 +0100285 if (nr_phys_segs == 1 && seg_size >
286 fbio->bi_seg_front_size)
287 fbio->bi_seg_front_size = seg_size;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200288
Jens Axboe1e428072009-02-23 09:03:10 +0100289 nr_phys_segs++;
290 bvprv = bv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700291 prev = 1;
Kent Overstreet79886132013-11-23 17:19:00 -0800292 seg_size = bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100293 }
Jens Axboe59247ea2009-03-06 08:55:24 +0100294 bbio = bio;
Jens Axboed6d48192008-01-29 14:04:06 +0100295 }
296
Jens Axboe59247ea2009-03-06 08:55:24 +0100297 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
298 fbio->bi_seg_front_size = seg_size;
299 if (seg_size > bbio->bi_seg_back_size)
300 bbio->bi_seg_back_size = seg_size;
Jens Axboe1e428072009-02-23 09:03:10 +0100301
302 return nr_phys_segs;
303}
304
305void blk_recalc_rq_segments(struct request *rq)
306{
Ming Lei07388542014-09-02 23:02:59 +0800307 bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE,
308 &rq->q->queue_flags);
309
310 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio,
311 no_sg_merge);
Jens Axboed6d48192008-01-29 14:04:06 +0100312}
313
314void blk_recount_segments(struct request_queue *q, struct bio *bio)
315{
Ming Lei7f60dca2014-11-12 00:15:41 +0800316 unsigned short seg_cnt;
Ming Lei764f6122014-10-09 23:17:35 +0800317
Ming Lei7f60dca2014-11-12 00:15:41 +0800318 /* estimate segment number by bi_vcnt for non-cloned bio */
319 if (bio_flagged(bio, BIO_CLONED))
320 seg_cnt = bio_segments(bio);
321 else
322 seg_cnt = bio->bi_vcnt;
323
324 if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) &&
325 (seg_cnt < queue_max_segments(q)))
326 bio->bi_phys_segments = seg_cnt;
Jens Axboe05f1dd52014-05-29 09:53:32 -0600327 else {
328 struct bio *nxt = bio->bi_next;
Jens Axboe1e428072009-02-23 09:03:10 +0100329
Jens Axboe05f1dd52014-05-29 09:53:32 -0600330 bio->bi_next = NULL;
Ming Lei7f60dca2014-11-12 00:15:41 +0800331 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false);
Jens Axboe05f1dd52014-05-29 09:53:32 -0600332 bio->bi_next = nxt;
333 }
334
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600335 bio_set_flag(bio, BIO_SEG_VALID);
Jens Axboed6d48192008-01-29 14:04:06 +0100336}
337EXPORT_SYMBOL(blk_recount_segments);
338
339static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
340 struct bio *nxt)
341{
Kent Overstreet2b8221e2013-12-03 14:29:09 -0700342 struct bio_vec end_bv = { NULL }, nxt_bv;
Kent Overstreetf619d252013-08-07 14:30:33 -0700343
Martin K. Petersene692cb62010-12-01 19:41:49 +0100344 if (!blk_queue_cluster(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100345 return 0;
346
FUJITA Tomonori86771422008-10-13 14:19:05 +0200347 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400348 queue_max_segment_size(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100349 return 0;
350
David Woodhousee17fc0a2008-08-09 16:42:20 +0100351 if (!bio_has_data(bio))
352 return 1;
353
Ming Leie8270912016-02-26 23:40:53 +0800354 bio_get_last_bvec(bio, &end_bv);
355 bio_get_first_bvec(nxt, &nxt_bv);
Kent Overstreetf619d252013-08-07 14:30:33 -0700356
357 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100358 return 0;
359
Jens Axboed6d48192008-01-29 14:04:06 +0100360 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +0100361 * bio and nxt are contiguous in memory; check if the queue allows
Jens Axboed6d48192008-01-29 14:04:06 +0100362 * these two to be merged into one
363 */
Kent Overstreetf619d252013-08-07 14:30:33 -0700364 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
Jens Axboed6d48192008-01-29 14:04:06 +0100365 return 1;
366
367 return 0;
368}
369
Kent Overstreet79886132013-11-23 17:19:00 -0800370static inline void
Asias He963ab9e2012-08-02 23:42:03 +0200371__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
Kent Overstreet79886132013-11-23 17:19:00 -0800372 struct scatterlist *sglist, struct bio_vec *bvprv,
Asias He963ab9e2012-08-02 23:42:03 +0200373 struct scatterlist **sg, int *nsegs, int *cluster)
374{
375
376 int nbytes = bvec->bv_len;
377
Kent Overstreet79886132013-11-23 17:19:00 -0800378 if (*sg && *cluster) {
Asias He963ab9e2012-08-02 23:42:03 +0200379 if ((*sg)->length + nbytes > queue_max_segment_size(q))
380 goto new_segment;
381
Kent Overstreet79886132013-11-23 17:19:00 -0800382 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200383 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800384 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200385 goto new_segment;
386
387 (*sg)->length += nbytes;
388 } else {
389new_segment:
390 if (!*sg)
391 *sg = sglist;
392 else {
393 /*
394 * If the driver previously mapped a shorter
395 * list, we could see a termination bit
396 * prematurely unless it fully inits the sg
397 * table on each mapping. We KNOW that there
398 * must be more entries here or the driver
399 * would be buggy, so force clear the
400 * termination bit to avoid doing a full
401 * sg_init_table() in drivers for each command.
402 */
Paolo Bonzinic8164d82013-03-20 15:37:08 +1030403 sg_unmark_end(*sg);
Asias He963ab9e2012-08-02 23:42:03 +0200404 *sg = sg_next(*sg);
405 }
406
407 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
408 (*nsegs)++;
409 }
Kent Overstreet79886132013-11-23 17:19:00 -0800410 *bvprv = *bvec;
Asias He963ab9e2012-08-02 23:42:03 +0200411}
412
Kent Overstreet5cb88502014-02-07 13:53:46 -0700413static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
414 struct scatterlist *sglist,
415 struct scatterlist **sg)
416{
417 struct bio_vec bvec, bvprv = { NULL };
418 struct bvec_iter iter;
419 int nsegs, cluster;
420
421 nsegs = 0;
422 cluster = blk_queue_cluster(q);
423
Adrian Hunter7afafc82016-08-16 10:59:35 +0300424 switch (bio_op(bio)) {
425 case REQ_OP_DISCARD:
426 case REQ_OP_SECURE_ERASE:
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800427 case REQ_OP_WRITE_ZEROES:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700428 /*
429 * This is a hack - drivers should be neither modifying the
430 * biovec, nor relying on bi_vcnt - but because of
431 * blk_add_request_payload(), a discard bio may or may not have
432 * a payload we need to set up here (thank you Christoph) and
433 * bi_vcnt is really the only way of telling if we need to.
434 */
Adrian Hunter7afafc82016-08-16 10:59:35 +0300435 if (!bio->bi_vcnt)
436 return 0;
437 /* Fall through */
438 case REQ_OP_WRITE_SAME:
Kent Overstreet5cb88502014-02-07 13:53:46 -0700439 *sg = sglist;
440 bvec = bio_iovec(bio);
441 sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset);
442 return 1;
Adrian Hunter7afafc82016-08-16 10:59:35 +0300443 default:
444 break;
Kent Overstreet5cb88502014-02-07 13:53:46 -0700445 }
446
447 for_each_bio(bio)
448 bio_for_each_segment(bvec, bio, iter)
449 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
450 &nsegs, &cluster);
451
452 return nsegs;
453}
454
Jens Axboed6d48192008-01-29 14:04:06 +0100455/*
456 * map a request to scatterlist, return number of sg entries setup. Caller
457 * must make sure sg can hold rq->nr_phys_segments entries
458 */
459int blk_rq_map_sg(struct request_queue *q, struct request *rq,
460 struct scatterlist *sglist)
461{
Kent Overstreet5cb88502014-02-07 13:53:46 -0700462 struct scatterlist *sg = NULL;
463 int nsegs = 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100464
Kent Overstreet5cb88502014-02-07 13:53:46 -0700465 if (rq->bio)
466 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200467
Christoph Hellwige8064022016-10-20 15:12:13 +0200468 if (unlikely(rq->rq_flags & RQF_COPY_USER) &&
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900469 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
470 unsigned int pad_len =
471 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200472
473 sg->length += pad_len;
474 rq->extra_len += pad_len;
475 }
476
Tejun Heo2fb98e82008-02-19 11:36:53 +0100477 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
Mike Christiea8ebb052016-06-05 14:31:45 -0500478 if (op_is_write(req_op(rq)))
Tejun Heodb0a2e02008-02-19 11:36:55 +0100479 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
480
Dan Williamsda81ed12015-08-07 18:15:14 +0200481 sg_unmark_end(sg);
Jens Axboed6d48192008-01-29 14:04:06 +0100482 sg = sg_next(sg);
483 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
484 q->dma_drain_size,
485 ((unsigned long)q->dma_drain_buffer) &
486 (PAGE_SIZE - 1));
487 nsegs++;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100488 rq->extra_len += q->dma_drain_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100489 }
490
491 if (sg)
492 sg_mark_end(sg);
493
Ming Lei12e57f52015-11-24 10:35:31 +0800494 /*
495 * Something must have been wrong if the figured number of
496 * segment is bigger than number of req's physical segments
497 */
498 WARN_ON(nsegs > rq->nr_phys_segments);
499
Jens Axboed6d48192008-01-29 14:04:06 +0100500 return nsegs;
501}
Jens Axboed6d48192008-01-29 14:04:06 +0100502EXPORT_SYMBOL(blk_rq_map_sg);
503
Jens Axboed6d48192008-01-29 14:04:06 +0100504static inline int ll_new_hw_segment(struct request_queue *q,
505 struct request *req,
506 struct bio *bio)
507{
Jens Axboed6d48192008-01-29 14:04:06 +0100508 int nr_phys_segs = bio_phys_segments(q, bio);
509
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200510 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
511 goto no_merge;
512
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400513 if (blk_integrity_merge_bio(q, req, bio) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200514 goto no_merge;
Jens Axboed6d48192008-01-29 14:04:06 +0100515
516 /*
517 * This will form the start of a new hw segment. Bump both
518 * counters.
519 */
Jens Axboed6d48192008-01-29 14:04:06 +0100520 req->nr_phys_segments += nr_phys_segs;
521 return 1;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200522
523no_merge:
524 req->cmd_flags |= REQ_NOMERGE;
525 if (req == q->last_merge)
526 q->last_merge = NULL;
527 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100528}
529
530int ll_back_merge_fn(struct request_queue *q, struct request *req,
531 struct bio *bio)
532{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300533 if (req_gap_back_merge(req, bio))
534 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600535 if (blk_integrity_rq(req) &&
536 integrity_req_gap_back_merge(req, bio))
537 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400538 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600539 blk_rq_get_max_sectors(req, blk_rq_pos(req))) {
Jens Axboed6d48192008-01-29 14:04:06 +0100540 req->cmd_flags |= REQ_NOMERGE;
541 if (req == q->last_merge)
542 q->last_merge = NULL;
543 return 0;
544 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200545 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100546 blk_recount_segments(q, req->biotail);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200547 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100548 blk_recount_segments(q, bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100549
550 return ll_new_hw_segment(q, req, bio);
551}
552
Jens Axboe6728cb02008-01-31 13:03:55 +0100553int ll_front_merge_fn(struct request_queue *q, struct request *req,
Jens Axboed6d48192008-01-29 14:04:06 +0100554 struct bio *bio)
555{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300556
557 if (req_gap_front_merge(req, bio))
558 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600559 if (blk_integrity_rq(req) &&
560 integrity_req_gap_front_merge(req, bio))
561 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400562 if (blk_rq_sectors(req) + bio_sectors(bio) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600563 blk_rq_get_max_sectors(req, bio->bi_iter.bi_sector)) {
Jens Axboed6d48192008-01-29 14:04:06 +0100564 req->cmd_flags |= REQ_NOMERGE;
565 if (req == q->last_merge)
566 q->last_merge = NULL;
567 return 0;
568 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200569 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100570 blk_recount_segments(q, bio);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200571 if (!bio_flagged(req->bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100572 blk_recount_segments(q, req->bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100573
574 return ll_new_hw_segment(q, req, bio);
575}
576
Jens Axboee7e24502013-10-29 12:11:47 -0600577/*
578 * blk-mq uses req->special to carry normal driver per-request payload, it
579 * does not indicate a prepared command that we cannot merge with.
580 */
581static bool req_no_special_merge(struct request *req)
582{
583 struct request_queue *q = req->q;
584
585 return !q->mq_ops && req->special;
586}
587
Jens Axboed6d48192008-01-29 14:04:06 +0100588static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
589 struct request *next)
590{
591 int total_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200592 unsigned int seg_size =
593 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100594
595 /*
596 * First check if the either of the requests are re-queued
597 * requests. Can't merge them if they are.
598 */
Jens Axboee7e24502013-10-29 12:11:47 -0600599 if (req_no_special_merge(req) || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100600 return 0;
601
Jens Axboe5e7c4272015-09-03 19:28:20 +0300602 if (req_gap_back_merge(req, next->bio))
Keith Busch854fbb92015-02-11 08:20:13 -0700603 return 0;
604
Jens Axboed6d48192008-01-29 14:04:06 +0100605 /*
606 * Will it become too large?
607 */
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400608 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
Damien Le Moal17007f32016-07-20 21:40:47 -0600609 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
Jens Axboed6d48192008-01-29 14:04:06 +0100610 return 0;
611
612 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200613 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
614 if (req->nr_phys_segments == 1)
615 req->bio->bi_seg_front_size = seg_size;
616 if (next->nr_phys_segments == 1)
617 next->biotail->bi_seg_back_size = seg_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100618 total_phys_segments--;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200619 }
Jens Axboed6d48192008-01-29 14:04:06 +0100620
Martin K. Petersen8a783622010-02-26 00:20:39 -0500621 if (total_phys_segments > queue_max_segments(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100622 return 0;
623
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400624 if (blk_integrity_merge_rq(q, req, next) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200625 return 0;
626
Jens Axboed6d48192008-01-29 14:04:06 +0100627 /* Merge is OK... */
628 req->nr_phys_segments = total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100629 return 1;
630}
631
Tejun Heo80a761f2009-07-03 17:48:17 +0900632/**
633 * blk_rq_set_mixed_merge - mark a request as mixed merge
634 * @rq: request to mark as mixed merge
635 *
636 * Description:
637 * @rq is about to be mixed merged. Make sure the attributes
638 * which can be mixed are set in each bio and mark @rq as mixed
639 * merged.
640 */
641void blk_rq_set_mixed_merge(struct request *rq)
642{
643 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
644 struct bio *bio;
645
Christoph Hellwige8064022016-10-20 15:12:13 +0200646 if (rq->rq_flags & RQF_MIXED_MERGE)
Tejun Heo80a761f2009-07-03 17:48:17 +0900647 return;
648
649 /*
650 * @rq will no longer represent mixable attributes for all the
651 * contained bios. It will just track those of the first one.
652 * Distributes the attributs to each bio.
653 */
654 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600655 WARN_ON_ONCE((bio->bi_opf & REQ_FAILFAST_MASK) &&
656 (bio->bi_opf & REQ_FAILFAST_MASK) != ff);
657 bio->bi_opf |= ff;
Tejun Heo80a761f2009-07-03 17:48:17 +0900658 }
Christoph Hellwige8064022016-10-20 15:12:13 +0200659 rq->rq_flags |= RQF_MIXED_MERGE;
Tejun Heo80a761f2009-07-03 17:48:17 +0900660}
661
Jerome Marchand26308ea2009-03-27 10:31:51 +0100662static void blk_account_io_merge(struct request *req)
663{
664 if (blk_do_io_stat(req)) {
665 struct hd_struct *part;
666 int cpu;
667
668 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100669 part = req->part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100670
671 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200672 part_dec_in_flight(part, rq_data_dir(req));
Jerome Marchand26308ea2009-03-27 10:31:51 +0100673
Jens Axboe6c23a962011-01-07 08:43:37 +0100674 hd_struct_put(part);
Jerome Marchand26308ea2009-03-27 10:31:51 +0100675 part_stat_unlock();
676 }
677}
678
Jens Axboed6d48192008-01-29 14:04:06 +0100679/*
680 * Has to be called with the request spinlock acquired
681 */
682static int attempt_merge(struct request_queue *q, struct request *req,
683 struct request *next)
684{
685 if (!rq_mergeable(req) || !rq_mergeable(next))
686 return 0;
687
Christoph Hellwig288dab82016-06-09 16:00:36 +0200688 if (req_op(req) != req_op(next))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400689 return 0;
690
Jens Axboed6d48192008-01-29 14:04:06 +0100691 /*
692 * not contiguous
693 */
Tejun Heo83096eb2009-05-07 22:24:39 +0900694 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100695 return 0;
696
697 if (rq_data_dir(req) != rq_data_dir(next)
698 || req->rq_disk != next->rq_disk
Jens Axboee7e24502013-10-29 12:11:47 -0600699 || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100700 return 0;
701
Mike Christie8fe0d472016-06-05 14:32:15 -0500702 if (req_op(req) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400703 !blk_write_same_mergeable(req->bio, next->bio))
704 return 0;
705
Jens Axboed6d48192008-01-29 14:04:06 +0100706 /*
707 * If we are allowed to merge, then append bio list
708 * from next to rq and release next. merge_requests_fn
709 * will have updated segment counts, update sector
710 * counts here.
711 */
712 if (!ll_merge_requests_fn(q, req, next))
713 return 0;
714
715 /*
Tejun Heo80a761f2009-07-03 17:48:17 +0900716 * If failfast settings disagree or any of the two is already
717 * a mixed merge, mark both as mixed before proceeding. This
718 * makes sure that all involved bios have mixable attributes
719 * set properly.
720 */
Christoph Hellwige8064022016-10-20 15:12:13 +0200721 if (((req->rq_flags | next->rq_flags) & RQF_MIXED_MERGE) ||
Tejun Heo80a761f2009-07-03 17:48:17 +0900722 (req->cmd_flags & REQ_FAILFAST_MASK) !=
723 (next->cmd_flags & REQ_FAILFAST_MASK)) {
724 blk_rq_set_mixed_merge(req);
725 blk_rq_set_mixed_merge(next);
726 }
727
728 /*
Jens Axboed6d48192008-01-29 14:04:06 +0100729 * At this point we have either done a back merge
730 * or front merge. We need the smaller start_time of
731 * the merged requests to be the current request
732 * for accounting purposes.
733 */
734 if (time_after(req->start_time, next->start_time))
735 req->start_time = next->start_time;
736
737 req->biotail->bi_next = next->bio;
738 req->biotail = next->biotail;
739
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900740 req->__data_len += blk_rq_bytes(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100741
742 elv_merge_requests(q, req, next);
743
Jerome Marchand42dad762009-04-22 14:01:49 +0200744 /*
745 * 'next' is going away, so update stats accordingly
746 */
747 blk_account_io_merge(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100748
749 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
Jens Axboeab780f12008-08-26 10:25:02 +0200750 if (blk_rq_cpu_valid(next))
751 req->cpu = next->cpu;
Jens Axboed6d48192008-01-29 14:04:06 +0100752
Boaz Harrosh1cd96c22009-03-24 12:35:07 +0100753 /* owner-ship of bio passed from next to req */
754 next->bio = NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100755 __blk_put_request(q, next);
756 return 1;
757}
758
759int attempt_back_merge(struct request_queue *q, struct request *rq)
760{
761 struct request *next = elv_latter_request(q, rq);
762
763 if (next)
764 return attempt_merge(q, rq, next);
765
766 return 0;
767}
768
769int attempt_front_merge(struct request_queue *q, struct request *rq)
770{
771 struct request *prev = elv_former_request(q, rq);
772
773 if (prev)
774 return attempt_merge(q, prev, rq);
775
776 return 0;
777}
Jens Axboe5e84ea32011-03-21 10:14:27 +0100778
779int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
780 struct request *next)
781{
Tahsin Erdogan72ef7992016-07-07 11:48:22 -0700782 struct elevator_queue *e = q->elevator;
783
784 if (e->type->ops.elevator_allow_rq_merge_fn)
785 if (!e->type->ops.elevator_allow_rq_merge_fn(q, rq, next))
786 return 0;
787
Jens Axboe5e84ea32011-03-21 10:14:27 +0100788 return attempt_merge(q, rq, next);
789}
Tejun Heo050c8ea2012-02-08 09:19:38 +0100790
791bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
792{
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400793 if (!rq_mergeable(rq) || !bio_mergeable(bio))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100794 return false;
795
Christoph Hellwig288dab82016-06-09 16:00:36 +0200796 if (req_op(rq) != bio_op(bio))
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400797 return false;
798
Tejun Heo050c8ea2012-02-08 09:19:38 +0100799 /* different data direction or already started, don't merge */
800 if (bio_data_dir(bio) != rq_data_dir(rq))
801 return false;
802
803 /* must be same device and not a special request */
Jens Axboee7e24502013-10-29 12:11:47 -0600804 if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100805 return false;
806
807 /* only merge integrity protected bio into ditto rq */
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400808 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100809 return false;
810
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400811 /* must be using the same buffer */
Mike Christie8fe0d472016-06-05 14:32:15 -0500812 if (req_op(rq) == REQ_OP_WRITE_SAME &&
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400813 !blk_write_same_mergeable(rq->bio, bio))
814 return false;
815
Tejun Heo050c8ea2012-02-08 09:19:38 +0100816 return true;
817}
818
819int blk_try_merge(struct request *rq, struct bio *bio)
820{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700821 if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100822 return ELEVATOR_BACK_MERGE;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700823 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100824 return ELEVATOR_FRONT_MERGE;
825 return ELEVATOR_NO_MERGE;
826}