blob: 888a7fec81f715199c3050576516e55b692f2b9b [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);
Kent Overstreet54efd502015-04-23 22:37:18 -070097
Kent Overstreet54efd502015-04-23 22:37:18 -070098 bio_for_each_segment(bv, bio, iter) {
Kent Overstreet54efd502015-04-23 22:37:18 -070099 /*
100 * If the queue doesn't support SG gaps and adding this
101 * offset would create a gap, disallow it.
102 */
Jens Axboe5014c312015-09-02 16:46:02 -0600103 if (bvprvp && bvec_gap_to_prev(q, bvprvp, bv.bv_offset))
Kent Overstreet54efd502015-04-23 22:37:18 -0700104 goto split;
105
Ming Leid0e5fbb2016-01-23 08:05:33 +0800106 if (sectors + (bv.bv_len >> 9) > max_sectors) {
Keith Busche36f6202016-01-12 15:08:39 -0700107 /*
108 * Consider this a new segment if we're splitting in
109 * the middle of this vector.
110 */
111 if (nsegs < queue_max_segments(q) &&
Ming Leid0e5fbb2016-01-23 08:05:33 +0800112 sectors < max_sectors) {
Keith Busche36f6202016-01-12 15:08:39 -0700113 nsegs++;
Ming Leid0e5fbb2016-01-23 08:05:33 +0800114 sectors = max_sectors;
Keith Busche36f6202016-01-12 15:08:39 -0700115 }
Ming Leid0e5fbb2016-01-23 08:05:33 +0800116 if (sectors)
117 goto split;
118 /* Make this single bvec as the 1st segment */
Keith Busche36f6202016-01-12 15:08:39 -0700119 }
120
Jens Axboe5014c312015-09-02 16:46:02 -0600121 if (bvprvp && blk_queue_cluster(q)) {
Kent Overstreet54efd502015-04-23 22:37:18 -0700122 if (seg_size + bv.bv_len > queue_max_segment_size(q))
123 goto new_segment;
Jens Axboe5014c312015-09-02 16:46:02 -0600124 if (!BIOVEC_PHYS_MERGEABLE(bvprvp, &bv))
Kent Overstreet54efd502015-04-23 22:37:18 -0700125 goto new_segment;
Jens Axboe5014c312015-09-02 16:46:02 -0600126 if (!BIOVEC_SEG_BOUNDARY(q, bvprvp, &bv))
Kent Overstreet54efd502015-04-23 22:37:18 -0700127 goto new_segment;
128
129 seg_size += bv.bv_len;
130 bvprv = bv;
Ming Lei578270b2015-11-24 10:35:29 +0800131 bvprvp = &bvprv;
Ming Lei52cc6ee2015-09-17 09:58:38 -0600132 sectors += bv.bv_len >> 9;
Ming Leia88d32a2015-11-30 16:05:49 +0800133
134 if (nsegs == 1 && seg_size > front_seg_size)
135 front_seg_size = seg_size;
Kent Overstreet54efd502015-04-23 22:37:18 -0700136 continue;
137 }
138new_segment:
139 if (nsegs == queue_max_segments(q))
140 goto split;
141
142 nsegs++;
143 bvprv = bv;
Ming Lei578270b2015-11-24 10:35:29 +0800144 bvprvp = &bvprv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700145 seg_size = bv.bv_len;
Ming Lei52cc6ee2015-09-17 09:58:38 -0600146 sectors += bv.bv_len >> 9;
Ming Lei02e70742015-11-24 10:35:30 +0800147
148 if (nsegs == 1 && seg_size > front_seg_size)
149 front_seg_size = seg_size;
Kent Overstreet54efd502015-04-23 22:37:18 -0700150 }
151
Ming Lei02e70742015-11-24 10:35:30 +0800152 do_split = false;
Kent Overstreet54efd502015-04-23 22:37:18 -0700153split:
Ming Leibdced432015-10-20 23:13:52 +0800154 *segs = nsegs;
Ming Lei02e70742015-11-24 10:35:30 +0800155
156 if (do_split) {
157 new = bio_split(bio, sectors, GFP_NOIO, bs);
158 if (new)
159 bio = new;
160 }
161
162 bio->bi_seg_front_size = front_seg_size;
163 if (seg_size > bio->bi_seg_back_size)
164 bio->bi_seg_back_size = seg_size;
165
166 return do_split ? new : NULL;
Kent Overstreet54efd502015-04-23 22:37:18 -0700167}
168
169void blk_queue_split(struct request_queue *q, struct bio **bio,
170 struct bio_set *bs)
171{
Ming Leibdced432015-10-20 23:13:52 +0800172 struct bio *split, *res;
173 unsigned nsegs;
Kent Overstreet54efd502015-04-23 22:37:18 -0700174
175 if ((*bio)->bi_rw & REQ_DISCARD)
Ming Leibdced432015-10-20 23:13:52 +0800176 split = blk_bio_discard_split(q, *bio, bs, &nsegs);
Kent Overstreet54efd502015-04-23 22:37:18 -0700177 else if ((*bio)->bi_rw & REQ_WRITE_SAME)
Ming Leibdced432015-10-20 23:13:52 +0800178 split = blk_bio_write_same_split(q, *bio, bs, &nsegs);
Kent Overstreet54efd502015-04-23 22:37:18 -0700179 else
Ming Leibdced432015-10-20 23:13:52 +0800180 split = blk_bio_segment_split(q, *bio, q->bio_split, &nsegs);
181
182 /* physical segments can be figured out during splitting */
183 res = split ? split : *bio;
184 res->bi_phys_segments = nsegs;
185 bio_set_flag(res, BIO_SEG_VALID);
Kent Overstreet54efd502015-04-23 22:37:18 -0700186
187 if (split) {
Ming Lei6ac45ae2015-10-20 23:13:53 +0800188 /* there isn't chance to merge the splitted bio */
189 split->bi_rw |= REQ_NOMERGE;
190
Kent Overstreet54efd502015-04-23 22:37:18 -0700191 bio_chain(split, *bio);
Mike Krinkincda22642015-12-03 17:32:30 +0300192 trace_block_split(q, split, (*bio)->bi_iter.bi_sector);
Kent Overstreet54efd502015-04-23 22:37:18 -0700193 generic_make_request(*bio);
194 *bio = split;
195 }
196}
197EXPORT_SYMBOL(blk_queue_split);
198
Jens Axboe1e428072009-02-23 09:03:10 +0100199static unsigned int __blk_recalc_rq_segments(struct request_queue *q,
Ming Lei07388542014-09-02 23:02:59 +0800200 struct bio *bio,
201 bool no_sg_merge)
Jens Axboed6d48192008-01-29 14:04:06 +0100202{
Kent Overstreet79886132013-11-23 17:19:00 -0800203 struct bio_vec bv, bvprv = { NULL };
Kent Overstreet54efd502015-04-23 22:37:18 -0700204 int cluster, prev = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100205 unsigned int seg_size, nr_phys_segs;
Jens Axboe59247ea2009-03-06 08:55:24 +0100206 struct bio *fbio, *bbio;
Kent Overstreet79886132013-11-23 17:19:00 -0800207 struct bvec_iter iter;
Jens Axboed6d48192008-01-29 14:04:06 +0100208
Jens Axboe1e428072009-02-23 09:03:10 +0100209 if (!bio)
210 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100211
Kent Overstreet5cb88502014-02-07 13:53:46 -0700212 /*
213 * This should probably be returning 0, but blk_add_request_payload()
214 * (Christoph!!!!)
215 */
216 if (bio->bi_rw & REQ_DISCARD)
217 return 1;
218
219 if (bio->bi_rw & REQ_WRITE_SAME)
220 return 1;
221
Jens Axboe1e428072009-02-23 09:03:10 +0100222 fbio = bio;
Martin K. Petersene692cb62010-12-01 19:41:49 +0100223 cluster = blk_queue_cluster(q);
Mikulas Patocka5df97b92008-08-15 10:20:02 +0200224 seg_size = 0;
Andi Kleen2c8919d2010-06-21 11:02:47 +0200225 nr_phys_segs = 0;
Jens Axboe1e428072009-02-23 09:03:10 +0100226 for_each_bio(bio) {
Kent Overstreet79886132013-11-23 17:19:00 -0800227 bio_for_each_segment(bv, bio, iter) {
Jens Axboe1e428072009-02-23 09:03:10 +0100228 /*
Jens Axboe05f1dd52014-05-29 09:53:32 -0600229 * If SG merging is disabled, each bio vector is
230 * a segment
231 */
232 if (no_sg_merge)
233 goto new_segment;
234
Kent Overstreet54efd502015-04-23 22:37:18 -0700235 if (prev && cluster) {
Kent Overstreet79886132013-11-23 17:19:00 -0800236 if (seg_size + bv.bv_len
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400237 > queue_max_segment_size(q))
Jens Axboe1e428072009-02-23 09:03:10 +0100238 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800239 if (!BIOVEC_PHYS_MERGEABLE(&bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100240 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800241 if (!BIOVEC_SEG_BOUNDARY(q, &bvprv, &bv))
Jens Axboe1e428072009-02-23 09:03:10 +0100242 goto new_segment;
Jens Axboed6d48192008-01-29 14:04:06 +0100243
Kent Overstreet79886132013-11-23 17:19:00 -0800244 seg_size += bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100245 bvprv = bv;
246 continue;
247 }
Jens Axboed6d48192008-01-29 14:04:06 +0100248new_segment:
Jens Axboe1e428072009-02-23 09:03:10 +0100249 if (nr_phys_segs == 1 && seg_size >
250 fbio->bi_seg_front_size)
251 fbio->bi_seg_front_size = seg_size;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200252
Jens Axboe1e428072009-02-23 09:03:10 +0100253 nr_phys_segs++;
254 bvprv = bv;
Kent Overstreet54efd502015-04-23 22:37:18 -0700255 prev = 1;
Kent Overstreet79886132013-11-23 17:19:00 -0800256 seg_size = bv.bv_len;
Jens Axboe1e428072009-02-23 09:03:10 +0100257 }
Jens Axboe59247ea2009-03-06 08:55:24 +0100258 bbio = bio;
Jens Axboed6d48192008-01-29 14:04:06 +0100259 }
260
Jens Axboe59247ea2009-03-06 08:55:24 +0100261 if (nr_phys_segs == 1 && seg_size > fbio->bi_seg_front_size)
262 fbio->bi_seg_front_size = seg_size;
263 if (seg_size > bbio->bi_seg_back_size)
264 bbio->bi_seg_back_size = seg_size;
Jens Axboe1e428072009-02-23 09:03:10 +0100265
266 return nr_phys_segs;
267}
268
269void blk_recalc_rq_segments(struct request *rq)
270{
Ming Lei07388542014-09-02 23:02:59 +0800271 bool no_sg_merge = !!test_bit(QUEUE_FLAG_NO_SG_MERGE,
272 &rq->q->queue_flags);
273
274 rq->nr_phys_segments = __blk_recalc_rq_segments(rq->q, rq->bio,
275 no_sg_merge);
Jens Axboed6d48192008-01-29 14:04:06 +0100276}
277
278void blk_recount_segments(struct request_queue *q, struct bio *bio)
279{
Ming Lei7f60dca2014-11-12 00:15:41 +0800280 unsigned short seg_cnt;
Ming Lei764f6122014-10-09 23:17:35 +0800281
Ming Lei7f60dca2014-11-12 00:15:41 +0800282 /* estimate segment number by bi_vcnt for non-cloned bio */
283 if (bio_flagged(bio, BIO_CLONED))
284 seg_cnt = bio_segments(bio);
285 else
286 seg_cnt = bio->bi_vcnt;
287
288 if (test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags) &&
289 (seg_cnt < queue_max_segments(q)))
290 bio->bi_phys_segments = seg_cnt;
Jens Axboe05f1dd52014-05-29 09:53:32 -0600291 else {
292 struct bio *nxt = bio->bi_next;
Jens Axboe1e428072009-02-23 09:03:10 +0100293
Jens Axboe05f1dd52014-05-29 09:53:32 -0600294 bio->bi_next = NULL;
Ming Lei7f60dca2014-11-12 00:15:41 +0800295 bio->bi_phys_segments = __blk_recalc_rq_segments(q, bio, false);
Jens Axboe05f1dd52014-05-29 09:53:32 -0600296 bio->bi_next = nxt;
297 }
298
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600299 bio_set_flag(bio, BIO_SEG_VALID);
Jens Axboed6d48192008-01-29 14:04:06 +0100300}
301EXPORT_SYMBOL(blk_recount_segments);
302
303static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio,
304 struct bio *nxt)
305{
Kent Overstreet2b8221e2013-12-03 14:29:09 -0700306 struct bio_vec end_bv = { NULL }, nxt_bv;
Kent Overstreetf619d252013-08-07 14:30:33 -0700307 struct bvec_iter iter;
308
Martin K. Petersene692cb62010-12-01 19:41:49 +0100309 if (!blk_queue_cluster(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100310 return 0;
311
FUJITA Tomonori86771422008-10-13 14:19:05 +0200312 if (bio->bi_seg_back_size + nxt->bi_seg_front_size >
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400313 queue_max_segment_size(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100314 return 0;
315
David Woodhousee17fc0a2008-08-09 16:42:20 +0100316 if (!bio_has_data(bio))
317 return 1;
318
Kent Overstreetf619d252013-08-07 14:30:33 -0700319 bio_for_each_segment(end_bv, bio, iter)
320 if (end_bv.bv_len == iter.bi_size)
321 break;
322
323 nxt_bv = bio_iovec(nxt);
324
325 if (!BIOVEC_PHYS_MERGEABLE(&end_bv, &nxt_bv))
David Woodhousee17fc0a2008-08-09 16:42:20 +0100326 return 0;
327
Jens Axboed6d48192008-01-29 14:04:06 +0100328 /*
David Woodhousee17fc0a2008-08-09 16:42:20 +0100329 * bio and nxt are contiguous in memory; check if the queue allows
Jens Axboed6d48192008-01-29 14:04:06 +0100330 * these two to be merged into one
331 */
Kent Overstreetf619d252013-08-07 14:30:33 -0700332 if (BIOVEC_SEG_BOUNDARY(q, &end_bv, &nxt_bv))
Jens Axboed6d48192008-01-29 14:04:06 +0100333 return 1;
334
335 return 0;
336}
337
Kent Overstreet79886132013-11-23 17:19:00 -0800338static inline void
Asias He963ab9e2012-08-02 23:42:03 +0200339__blk_segment_map_sg(struct request_queue *q, struct bio_vec *bvec,
Kent Overstreet79886132013-11-23 17:19:00 -0800340 struct scatterlist *sglist, struct bio_vec *bvprv,
Asias He963ab9e2012-08-02 23:42:03 +0200341 struct scatterlist **sg, int *nsegs, int *cluster)
342{
343
344 int nbytes = bvec->bv_len;
345
Kent Overstreet79886132013-11-23 17:19:00 -0800346 if (*sg && *cluster) {
Asias He963ab9e2012-08-02 23:42:03 +0200347 if ((*sg)->length + nbytes > queue_max_segment_size(q))
348 goto new_segment;
349
Kent Overstreet79886132013-11-23 17:19:00 -0800350 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200351 goto new_segment;
Kent Overstreet79886132013-11-23 17:19:00 -0800352 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
Asias He963ab9e2012-08-02 23:42:03 +0200353 goto new_segment;
354
355 (*sg)->length += nbytes;
356 } else {
357new_segment:
358 if (!*sg)
359 *sg = sglist;
360 else {
361 /*
362 * If the driver previously mapped a shorter
363 * list, we could see a termination bit
364 * prematurely unless it fully inits the sg
365 * table on each mapping. We KNOW that there
366 * must be more entries here or the driver
367 * would be buggy, so force clear the
368 * termination bit to avoid doing a full
369 * sg_init_table() in drivers for each command.
370 */
Paolo Bonzinic8164d82013-03-20 15:37:08 +1030371 sg_unmark_end(*sg);
Asias He963ab9e2012-08-02 23:42:03 +0200372 *sg = sg_next(*sg);
373 }
374
375 sg_set_page(*sg, bvec->bv_page, nbytes, bvec->bv_offset);
376 (*nsegs)++;
377 }
Kent Overstreet79886132013-11-23 17:19:00 -0800378 *bvprv = *bvec;
Asias He963ab9e2012-08-02 23:42:03 +0200379}
380
Kent Overstreet5cb88502014-02-07 13:53:46 -0700381static int __blk_bios_map_sg(struct request_queue *q, struct bio *bio,
382 struct scatterlist *sglist,
383 struct scatterlist **sg)
384{
385 struct bio_vec bvec, bvprv = { NULL };
386 struct bvec_iter iter;
387 int nsegs, cluster;
388
389 nsegs = 0;
390 cluster = blk_queue_cluster(q);
391
392 if (bio->bi_rw & REQ_DISCARD) {
393 /*
394 * This is a hack - drivers should be neither modifying the
395 * biovec, nor relying on bi_vcnt - but because of
396 * blk_add_request_payload(), a discard bio may or may not have
397 * a payload we need to set up here (thank you Christoph) and
398 * bi_vcnt is really the only way of telling if we need to.
399 */
400
401 if (bio->bi_vcnt)
402 goto single_segment;
403
404 return 0;
405 }
406
407 if (bio->bi_rw & REQ_WRITE_SAME) {
408single_segment:
409 *sg = sglist;
410 bvec = bio_iovec(bio);
411 sg_set_page(*sg, bvec.bv_page, bvec.bv_len, bvec.bv_offset);
412 return 1;
413 }
414
415 for_each_bio(bio)
416 bio_for_each_segment(bvec, bio, iter)
417 __blk_segment_map_sg(q, &bvec, sglist, &bvprv, sg,
418 &nsegs, &cluster);
419
420 return nsegs;
421}
422
Jens Axboed6d48192008-01-29 14:04:06 +0100423/*
424 * map a request to scatterlist, return number of sg entries setup. Caller
425 * must make sure sg can hold rq->nr_phys_segments entries
426 */
427int blk_rq_map_sg(struct request_queue *q, struct request *rq,
428 struct scatterlist *sglist)
429{
Kent Overstreet5cb88502014-02-07 13:53:46 -0700430 struct scatterlist *sg = NULL;
431 int nsegs = 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100432
Kent Overstreet5cb88502014-02-07 13:53:46 -0700433 if (rq->bio)
434 nsegs = __blk_bios_map_sg(q, rq->bio, sglist, &sg);
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200435
436 if (unlikely(rq->cmd_flags & REQ_COPY_USER) &&
Tejun Heo2e46e8b2009-05-07 22:24:41 +0900437 (blk_rq_bytes(rq) & q->dma_pad_mask)) {
438 unsigned int pad_len =
439 (q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
FUJITA Tomonorif18573a2008-04-11 12:56:52 +0200440
441 sg->length += pad_len;
442 rq->extra_len += pad_len;
443 }
444
Tejun Heo2fb98e82008-02-19 11:36:53 +0100445 if (q->dma_drain_size && q->dma_drain_needed(rq)) {
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200446 if (rq->cmd_flags & REQ_WRITE)
Tejun Heodb0a2e02008-02-19 11:36:55 +0100447 memset(q->dma_drain_buffer, 0, q->dma_drain_size);
448
Dan Williamsda81ed12015-08-07 18:15:14 +0200449 sg_unmark_end(sg);
Jens Axboed6d48192008-01-29 14:04:06 +0100450 sg = sg_next(sg);
451 sg_set_page(sg, virt_to_page(q->dma_drain_buffer),
452 q->dma_drain_size,
453 ((unsigned long)q->dma_drain_buffer) &
454 (PAGE_SIZE - 1));
455 nsegs++;
FUJITA Tomonori7a85f882008-03-04 11:17:11 +0100456 rq->extra_len += q->dma_drain_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100457 }
458
459 if (sg)
460 sg_mark_end(sg);
461
Ming Lei12e57f52015-11-24 10:35:31 +0800462 /*
463 * Something must have been wrong if the figured number of
464 * segment is bigger than number of req's physical segments
465 */
466 WARN_ON(nsegs > rq->nr_phys_segments);
467
Jens Axboed6d48192008-01-29 14:04:06 +0100468 return nsegs;
469}
Jens Axboed6d48192008-01-29 14:04:06 +0100470EXPORT_SYMBOL(blk_rq_map_sg);
471
Jens Axboed6d48192008-01-29 14:04:06 +0100472static inline int ll_new_hw_segment(struct request_queue *q,
473 struct request *req,
474 struct bio *bio)
475{
Jens Axboed6d48192008-01-29 14:04:06 +0100476 int nr_phys_segs = bio_phys_segments(q, bio);
477
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200478 if (req->nr_phys_segments + nr_phys_segs > queue_max_segments(q))
479 goto no_merge;
480
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400481 if (blk_integrity_merge_bio(q, req, bio) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200482 goto no_merge;
Jens Axboed6d48192008-01-29 14:04:06 +0100483
484 /*
485 * This will form the start of a new hw segment. Bump both
486 * counters.
487 */
Jens Axboed6d48192008-01-29 14:04:06 +0100488 req->nr_phys_segments += nr_phys_segs;
489 return 1;
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200490
491no_merge:
492 req->cmd_flags |= REQ_NOMERGE;
493 if (req == q->last_merge)
494 q->last_merge = NULL;
495 return 0;
Jens Axboed6d48192008-01-29 14:04:06 +0100496}
497
498int ll_back_merge_fn(struct request_queue *q, struct request *req,
499 struct bio *bio)
500{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300501 if (req_gap_back_merge(req, bio))
502 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600503 if (blk_integrity_rq(req) &&
504 integrity_req_gap_back_merge(req, bio))
505 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400506 if (blk_rq_sectors(req) + bio_sectors(bio) >
507 blk_rq_get_max_sectors(req)) {
Jens Axboed6d48192008-01-29 14:04:06 +0100508 req->cmd_flags |= REQ_NOMERGE;
509 if (req == q->last_merge)
510 q->last_merge = NULL;
511 return 0;
512 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200513 if (!bio_flagged(req->biotail, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100514 blk_recount_segments(q, req->biotail);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200515 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100516 blk_recount_segments(q, bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100517
518 return ll_new_hw_segment(q, req, bio);
519}
520
Jens Axboe6728cb02008-01-31 13:03:55 +0100521int ll_front_merge_fn(struct request_queue *q, struct request *req,
Jens Axboed6d48192008-01-29 14:04:06 +0100522 struct bio *bio)
523{
Jens Axboe5e7c4272015-09-03 19:28:20 +0300524
525 if (req_gap_front_merge(req, bio))
526 return 0;
Sagi Grimberg7f39add2015-09-11 09:03:04 -0600527 if (blk_integrity_rq(req) &&
528 integrity_req_gap_front_merge(req, bio))
529 return 0;
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400530 if (blk_rq_sectors(req) + bio_sectors(bio) >
531 blk_rq_get_max_sectors(req)) {
Jens Axboed6d48192008-01-29 14:04:06 +0100532 req->cmd_flags |= REQ_NOMERGE;
533 if (req == q->last_merge)
534 q->last_merge = NULL;
535 return 0;
536 }
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200537 if (!bio_flagged(bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100538 blk_recount_segments(q, bio);
Jens Axboe2cdf79c2008-05-07 09:33:55 +0200539 if (!bio_flagged(req->bio, BIO_SEG_VALID))
Jens Axboed6d48192008-01-29 14:04:06 +0100540 blk_recount_segments(q, req->bio);
Jens Axboed6d48192008-01-29 14:04:06 +0100541
542 return ll_new_hw_segment(q, req, bio);
543}
544
Jens Axboee7e24502013-10-29 12:11:47 -0600545/*
546 * blk-mq uses req->special to carry normal driver per-request payload, it
547 * does not indicate a prepared command that we cannot merge with.
548 */
549static bool req_no_special_merge(struct request *req)
550{
551 struct request_queue *q = req->q;
552
553 return !q->mq_ops && req->special;
554}
555
Jens Axboed6d48192008-01-29 14:04:06 +0100556static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
557 struct request *next)
558{
559 int total_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200560 unsigned int seg_size =
561 req->biotail->bi_seg_back_size + next->bio->bi_seg_front_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100562
563 /*
564 * First check if the either of the requests are re-queued
565 * requests. Can't merge them if they are.
566 */
Jens Axboee7e24502013-10-29 12:11:47 -0600567 if (req_no_special_merge(req) || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100568 return 0;
569
Jens Axboe5e7c4272015-09-03 19:28:20 +0300570 if (req_gap_back_merge(req, next->bio))
Keith Busch854fbb92015-02-11 08:20:13 -0700571 return 0;
572
Jens Axboed6d48192008-01-29 14:04:06 +0100573 /*
574 * Will it become too large?
575 */
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400576 if ((blk_rq_sectors(req) + blk_rq_sectors(next)) >
577 blk_rq_get_max_sectors(req))
Jens Axboed6d48192008-01-29 14:04:06 +0100578 return 0;
579
580 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200581 if (blk_phys_contig_segment(q, req->biotail, next->bio)) {
582 if (req->nr_phys_segments == 1)
583 req->bio->bi_seg_front_size = seg_size;
584 if (next->nr_phys_segments == 1)
585 next->biotail->bi_seg_back_size = seg_size;
Jens Axboed6d48192008-01-29 14:04:06 +0100586 total_phys_segments--;
FUJITA Tomonori86771422008-10-13 14:19:05 +0200587 }
Jens Axboed6d48192008-01-29 14:04:06 +0100588
Martin K. Petersen8a783622010-02-26 00:20:39 -0500589 if (total_phys_segments > queue_max_segments(q))
Jens Axboed6d48192008-01-29 14:04:06 +0100590 return 0;
591
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400592 if (blk_integrity_merge_rq(q, req, next) == false)
Martin K. Petersen13f05c82010-09-10 20:50:10 +0200593 return 0;
594
Jens Axboed6d48192008-01-29 14:04:06 +0100595 /* Merge is OK... */
596 req->nr_phys_segments = total_phys_segments;
Jens Axboed6d48192008-01-29 14:04:06 +0100597 return 1;
598}
599
Tejun Heo80a761f2009-07-03 17:48:17 +0900600/**
601 * blk_rq_set_mixed_merge - mark a request as mixed merge
602 * @rq: request to mark as mixed merge
603 *
604 * Description:
605 * @rq is about to be mixed merged. Make sure the attributes
606 * which can be mixed are set in each bio and mark @rq as mixed
607 * merged.
608 */
609void blk_rq_set_mixed_merge(struct request *rq)
610{
611 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
612 struct bio *bio;
613
614 if (rq->cmd_flags & REQ_MIXED_MERGE)
615 return;
616
617 /*
618 * @rq will no longer represent mixable attributes for all the
619 * contained bios. It will just track those of the first one.
620 * Distributes the attributs to each bio.
621 */
622 for (bio = rq->bio; bio; bio = bio->bi_next) {
623 WARN_ON_ONCE((bio->bi_rw & REQ_FAILFAST_MASK) &&
624 (bio->bi_rw & REQ_FAILFAST_MASK) != ff);
625 bio->bi_rw |= ff;
626 }
627 rq->cmd_flags |= REQ_MIXED_MERGE;
628}
629
Jerome Marchand26308ea2009-03-27 10:31:51 +0100630static void blk_account_io_merge(struct request *req)
631{
632 if (blk_do_io_stat(req)) {
633 struct hd_struct *part;
634 int cpu;
635
636 cpu = part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100637 part = req->part;
Jerome Marchand26308ea2009-03-27 10:31:51 +0100638
639 part_round_stats(cpu, part);
Nikanth Karthikesan316d3152009-10-06 20:16:55 +0200640 part_dec_in_flight(part, rq_data_dir(req));
Jerome Marchand26308ea2009-03-27 10:31:51 +0100641
Jens Axboe6c23a962011-01-07 08:43:37 +0100642 hd_struct_put(part);
Jerome Marchand26308ea2009-03-27 10:31:51 +0100643 part_stat_unlock();
644 }
645}
646
Jens Axboed6d48192008-01-29 14:04:06 +0100647/*
648 * Has to be called with the request spinlock acquired
649 */
650static int attempt_merge(struct request_queue *q, struct request *req,
651 struct request *next)
652{
653 if (!rq_mergeable(req) || !rq_mergeable(next))
654 return 0;
655
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400656 if (!blk_check_merge_flags(req->cmd_flags, next->cmd_flags))
657 return 0;
658
Jens Axboed6d48192008-01-29 14:04:06 +0100659 /*
660 * not contiguous
661 */
Tejun Heo83096eb2009-05-07 22:24:39 +0900662 if (blk_rq_pos(req) + blk_rq_sectors(req) != blk_rq_pos(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100663 return 0;
664
665 if (rq_data_dir(req) != rq_data_dir(next)
666 || req->rq_disk != next->rq_disk
Jens Axboee7e24502013-10-29 12:11:47 -0600667 || req_no_special_merge(next))
Jens Axboed6d48192008-01-29 14:04:06 +0100668 return 0;
669
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400670 if (req->cmd_flags & REQ_WRITE_SAME &&
671 !blk_write_same_mergeable(req->bio, next->bio))
672 return 0;
673
Jens Axboed6d48192008-01-29 14:04:06 +0100674 /*
675 * If we are allowed to merge, then append bio list
676 * from next to rq and release next. merge_requests_fn
677 * will have updated segment counts, update sector
678 * counts here.
679 */
680 if (!ll_merge_requests_fn(q, req, next))
681 return 0;
682
683 /*
Tejun Heo80a761f2009-07-03 17:48:17 +0900684 * If failfast settings disagree or any of the two is already
685 * a mixed merge, mark both as mixed before proceeding. This
686 * makes sure that all involved bios have mixable attributes
687 * set properly.
688 */
689 if ((req->cmd_flags | next->cmd_flags) & REQ_MIXED_MERGE ||
690 (req->cmd_flags & REQ_FAILFAST_MASK) !=
691 (next->cmd_flags & REQ_FAILFAST_MASK)) {
692 blk_rq_set_mixed_merge(req);
693 blk_rq_set_mixed_merge(next);
694 }
695
696 /*
Jens Axboed6d48192008-01-29 14:04:06 +0100697 * At this point we have either done a back merge
698 * or front merge. We need the smaller start_time of
699 * the merged requests to be the current request
700 * for accounting purposes.
701 */
702 if (time_after(req->start_time, next->start_time))
703 req->start_time = next->start_time;
704
705 req->biotail->bi_next = next->bio;
706 req->biotail = next->biotail;
707
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900708 req->__data_len += blk_rq_bytes(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100709
710 elv_merge_requests(q, req, next);
711
Jerome Marchand42dad762009-04-22 14:01:49 +0200712 /*
713 * 'next' is going away, so update stats accordingly
714 */
715 blk_account_io_merge(next);
Jens Axboed6d48192008-01-29 14:04:06 +0100716
717 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
Jens Axboeab780f12008-08-26 10:25:02 +0200718 if (blk_rq_cpu_valid(next))
719 req->cpu = next->cpu;
Jens Axboed6d48192008-01-29 14:04:06 +0100720
Boaz Harrosh1cd96c22009-03-24 12:35:07 +0100721 /* owner-ship of bio passed from next to req */
722 next->bio = NULL;
Jens Axboed6d48192008-01-29 14:04:06 +0100723 __blk_put_request(q, next);
724 return 1;
725}
726
727int attempt_back_merge(struct request_queue *q, struct request *rq)
728{
729 struct request *next = elv_latter_request(q, rq);
730
731 if (next)
732 return attempt_merge(q, rq, next);
733
734 return 0;
735}
736
737int attempt_front_merge(struct request_queue *q, struct request *rq)
738{
739 struct request *prev = elv_former_request(q, rq);
740
741 if (prev)
742 return attempt_merge(q, prev, rq);
743
744 return 0;
745}
Jens Axboe5e84ea32011-03-21 10:14:27 +0100746
747int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
748 struct request *next)
749{
750 return attempt_merge(q, rq, next);
751}
Tejun Heo050c8ea2012-02-08 09:19:38 +0100752
753bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
754{
Martin K. Petersene2a60da2012-09-18 12:19:25 -0400755 if (!rq_mergeable(rq) || !bio_mergeable(bio))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100756 return false;
757
Martin K. Petersenf31dc1c2012-09-18 12:19:26 -0400758 if (!blk_check_merge_flags(rq->cmd_flags, bio->bi_rw))
759 return false;
760
Tejun Heo050c8ea2012-02-08 09:19:38 +0100761 /* different data direction or already started, don't merge */
762 if (bio_data_dir(bio) != rq_data_dir(rq))
763 return false;
764
765 /* must be same device and not a special request */
Jens Axboee7e24502013-10-29 12:11:47 -0600766 if (rq->rq_disk != bio->bi_bdev->bd_disk || req_no_special_merge(rq))
Tejun Heo050c8ea2012-02-08 09:19:38 +0100767 return false;
768
769 /* only merge integrity protected bio into ditto rq */
Martin K. Petersen4eaf99b2014-09-26 19:20:06 -0400770 if (blk_integrity_merge_bio(rq->q, rq, bio) == false)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100771 return false;
772
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400773 /* must be using the same buffer */
774 if (rq->cmd_flags & REQ_WRITE_SAME &&
775 !blk_write_same_mergeable(rq->bio, bio))
776 return false;
777
Tejun Heo050c8ea2012-02-08 09:19:38 +0100778 return true;
779}
780
781int blk_try_merge(struct request *rq, struct bio *bio)
782{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700783 if (blk_rq_pos(rq) + blk_rq_sectors(rq) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100784 return ELEVATOR_BACK_MERGE;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700785 else if (blk_rq_pos(rq) - bio_sectors(bio) == bio->bi_iter.bi_sector)
Tejun Heo050c8ea2012-02-08 09:19:38 +0100786 return ELEVATOR_FRONT_MERGE;
787 return ELEVATOR_NO_MERGE;
788}