blob: 7688ee3f5d729ea3210bb2d421c607bf063d4119 [file] [log] [blame]
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +04001/*
2 * Functions related to generic helpers functions
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
10#include "blk.h"
11
Lukas Czerner5dba3082011-05-06 19:26:27 -060012struct bio_batch {
13 atomic_t done;
14 unsigned long flags;
15 struct completion *wait;
16};
17
18static void bio_batch_end_io(struct bio *bio, int err)
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040019{
Lukas Czerner5dba3082011-05-06 19:26:27 -060020 struct bio_batch *bb = bio->bi_private;
21
Lukas Czerner8af19542011-05-06 19:30:01 -060022 if (err && (err != -EOPNOTSUPP))
Lukas Czerner5dba3082011-05-06 19:26:27 -060023 clear_bit(BIO_UPTODATE, &bb->flags);
Lukas Czerner5dba3082011-05-06 19:26:27 -060024 if (atomic_dec_and_test(&bb->done))
25 complete(bb->wait);
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040026 bio_put(bio);
27}
28
29/**
30 * blkdev_issue_discard - queue a discard
31 * @bdev: blockdev to issue discard for
32 * @sector: start sector
33 * @nr_sects: number of sectors to discard
34 * @gfp_mask: memory allocation flags (for bio_alloc)
35 * @flags: BLKDEV_IFL_* flags to control behaviour
36 *
37 * Description:
38 * Issue a discard request for the sectors in question.
39 */
40int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
41 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
42{
43 DECLARE_COMPLETION_ONSTACK(wait);
44 struct request_queue *q = bdev_get_queue(bdev);
Christoph Hellwig8c555362010-08-18 05:29:22 -040045 int type = REQ_WRITE | REQ_DISCARD;
Geert Uytterhoeven97597dc2013-11-04 14:00:06 +010046 unsigned int max_discard_sectors, granularity;
47 int alignment;
Lukas Czerner5dba3082011-05-06 19:26:27 -060048 struct bio_batch bb;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040049 struct bio *bio;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040050 int ret = 0;
Shaohua Li0cfbcaf2012-12-14 11:15:51 +080051 struct blk_plug plug;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040052
53 if (!q)
54 return -ENXIO;
55
56 if (!blk_queue_discard(q))
57 return -EOPNOTSUPP;
58
Paolo Bonzinif6ff53d2012-08-02 09:48:49 +020059 /* Zero-sector (unknown) and one-sector granularities are the same. */
60 granularity = max(q->limits.discard_granularity >> 9, 1U);
Geert Uytterhoeven97597dc2013-11-04 14:00:06 +010061 alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
Paolo Bonzinif6ff53d2012-08-02 09:48:49 +020062
Jens Axboe10d1f9e2010-07-15 10:49:31 -060063 /*
64 * Ensure that max_discard_sectors is of the proper
Paolo Bonzinic6e66632012-08-02 09:48:50 +020065 * granularity, so that requests stay aligned after a split.
Jens Axboe10d1f9e2010-07-15 10:49:31 -060066 */
67 max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
Geert Uytterhoeven97597dc2013-11-04 14:00:06 +010068 max_discard_sectors -= max_discard_sectors % granularity;
Jens Axboe4c645002011-07-23 20:34:59 +020069 if (unlikely(!max_discard_sectors)) {
Mike Snitzer0f799602011-07-06 21:30:50 +020070 /* Avoid infinite loop below. Being cautious never hurts. */
71 return -EOPNOTSUPP;
Jens Axboe10d1f9e2010-07-15 10:49:31 -060072 }
73
Christoph Hellwigdd3932e2010-09-16 20:51:46 +020074 if (flags & BLKDEV_DISCARD_SECURE) {
Adrian Hunter8d57a982010-08-11 14:17:49 -070075 if (!blk_queue_secdiscard(q))
76 return -EOPNOTSUPP;
Christoph Hellwig8c555362010-08-18 05:29:22 -040077 type |= REQ_SECURE;
Adrian Hunter8d57a982010-08-11 14:17:49 -070078 }
79
Lukas Czerner5dba3082011-05-06 19:26:27 -060080 atomic_set(&bb.done, 1);
81 bb.flags = 1 << BIO_UPTODATE;
82 bb.wait = &wait;
83
Shaohua Li0cfbcaf2012-12-14 11:15:51 +080084 blk_start_plug(&plug);
Lukas Czerner5dba3082011-05-06 19:26:27 -060085 while (nr_sects) {
Paolo Bonzinic6e66632012-08-02 09:48:50 +020086 unsigned int req_sects;
Shaohua Li8dd2cb72012-12-14 11:15:36 +080087 sector_t end_sect, tmp;
Paolo Bonzinic6e66632012-08-02 09:48:50 +020088
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +040089 bio = bio_alloc(gfp_mask, 1);
Christoph Hellwig66ac0282010-06-18 16:59:42 +020090 if (!bio) {
91 ret = -ENOMEM;
92 break;
93 }
94
Paolo Bonzinic6e66632012-08-02 09:48:50 +020095 req_sects = min_t(sector_t, nr_sects, max_discard_sectors);
96
97 /*
98 * If splitting a request, and the next starting sector would be
99 * misaligned, stop the discard at the previous aligned sector.
100 */
101 end_sect = sector + req_sects;
Shaohua Li8dd2cb72012-12-14 11:15:36 +0800102 tmp = end_sect;
103 if (req_sects < nr_sects &&
104 sector_div(tmp, granularity) != alignment) {
105 end_sect = end_sect - alignment;
106 sector_div(end_sect, granularity);
107 end_sect = end_sect * granularity + alignment;
Paolo Bonzinic6e66632012-08-02 09:48:50 +0200108 req_sects = end_sect - sector;
109 }
110
Kent Overstreet4f024f32013-10-11 15:44:27 -0700111 bio->bi_iter.bi_sector = sector;
Lukas Czerner5dba3082011-05-06 19:26:27 -0600112 bio->bi_end_io = bio_batch_end_io;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400113 bio->bi_bdev = bdev;
Lukas Czerner5dba3082011-05-06 19:26:27 -0600114 bio->bi_private = &bb;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400115
Kent Overstreet4f024f32013-10-11 15:44:27 -0700116 bio->bi_iter.bi_size = req_sects << 9;
Paolo Bonzinic6e66632012-08-02 09:48:50 +0200117 nr_sects -= req_sects;
118 sector = end_sect;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400119
Lukas Czerner5dba3082011-05-06 19:26:27 -0600120 atomic_inc(&bb.done);
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400121 submit_bio(type, bio);
Jens Axboec8123f82014-02-12 09:34:01 -0700122
123 /*
124 * We can loop for a long time in here, if someone does
125 * full device discards (like mkfs). Be nice and allow
126 * us to schedule out to avoid softlocking if preempt
127 * is disabled.
128 */
129 cond_resched();
Lukas Czerner5dba3082011-05-06 19:26:27 -0600130 }
Shaohua Li0cfbcaf2012-12-14 11:15:51 +0800131 blk_finish_plug(&plug);
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400132
Lukas Czerner5dba3082011-05-06 19:26:27 -0600133 /* Wait for bios in-flight */
134 if (!atomic_dec_and_test(&bb.done))
Vladimir Davydov55770222013-02-14 18:19:59 +0400135 wait_for_completion_io(&wait);
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400136
Lukas Czerner8af19542011-05-06 19:30:01 -0600137 if (!test_bit(BIO_UPTODATE, &bb.flags))
Lukas Czerner5dba3082011-05-06 19:26:27 -0600138 ret = -EIO;
Christoph Hellwig66ac0282010-06-18 16:59:42 +0200139
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400140 return ret;
Dmitry Monakhovf31e7e42010-04-28 17:55:08 +0400141}
142EXPORT_SYMBOL(blkdev_issue_discard);
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400143
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400144/**
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400145 * blkdev_issue_write_same - queue a write same operation
146 * @bdev: target blockdev
147 * @sector: start sector
148 * @nr_sects: number of sectors to write
149 * @gfp_mask: memory allocation flags (for bio_alloc)
150 * @page: page containing data to write
151 *
152 * Description:
153 * Issue a write same request for the sectors in question.
154 */
155int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
156 sector_t nr_sects, gfp_t gfp_mask,
157 struct page *page)
158{
159 DECLARE_COMPLETION_ONSTACK(wait);
160 struct request_queue *q = bdev_get_queue(bdev);
161 unsigned int max_write_same_sectors;
162 struct bio_batch bb;
163 struct bio *bio;
164 int ret = 0;
165
166 if (!q)
167 return -ENXIO;
168
169 max_write_same_sectors = q->limits.max_write_same_sectors;
170
171 if (max_write_same_sectors == 0)
172 return -EOPNOTSUPP;
173
174 atomic_set(&bb.done, 1);
175 bb.flags = 1 << BIO_UPTODATE;
176 bb.wait = &wait;
177
178 while (nr_sects) {
179 bio = bio_alloc(gfp_mask, 1);
180 if (!bio) {
181 ret = -ENOMEM;
182 break;
183 }
184
Kent Overstreet4f024f32013-10-11 15:44:27 -0700185 bio->bi_iter.bi_sector = sector;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400186 bio->bi_end_io = bio_batch_end_io;
187 bio->bi_bdev = bdev;
188 bio->bi_private = &bb;
189 bio->bi_vcnt = 1;
190 bio->bi_io_vec->bv_page = page;
191 bio->bi_io_vec->bv_offset = 0;
192 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
193
194 if (nr_sects > max_write_same_sectors) {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700195 bio->bi_iter.bi_size = max_write_same_sectors << 9;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400196 nr_sects -= max_write_same_sectors;
197 sector += max_write_same_sectors;
198 } else {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700199 bio->bi_iter.bi_size = nr_sects << 9;
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400200 nr_sects = 0;
201 }
202
203 atomic_inc(&bb.done);
204 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
205 }
206
207 /* Wait for bios in-flight */
208 if (!atomic_dec_and_test(&bb.done))
Vladimir Davydov55770222013-02-14 18:19:59 +0400209 wait_for_completion_io(&wait);
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400210
211 if (!test_bit(BIO_UPTODATE, &bb.flags))
212 ret = -ENOTSUPP;
213
214 return ret;
215}
216EXPORT_SYMBOL(blkdev_issue_write_same);
217
218/**
Ben Hutchings291d24f2011-03-01 13:45:24 -0500219 * blkdev_issue_zeroout - generate number of zero filed write bios
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400220 * @bdev: blockdev to issue
221 * @sector: start sector
222 * @nr_sects: number of sectors to write
223 * @gfp_mask: memory allocation flags (for bio_alloc)
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400224 *
225 * Description:
226 * Generate and issue number of bios with zerofiled pages.
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400227 */
228
Fabian Frederick35086782014-05-26 22:19:14 +0200229static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
230 sector_t nr_sects, gfp_t gfp_mask)
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400231{
Dmitry Monakhov18edc8e2010-08-06 13:23:25 +0200232 int ret;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400233 struct bio *bio;
234 struct bio_batch bb;
Lukas Czerner0aeea182011-03-11 10:23:53 +0100235 unsigned int sz;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400236 DECLARE_COMPLETION_ONSTACK(wait);
237
Lukas Czerner0aeea182011-03-11 10:23:53 +0100238 atomic_set(&bb.done, 1);
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400239 bb.flags = 1 << BIO_UPTODATE;
240 bb.wait = &wait;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400241
Dmitry Monakhov18edc8e2010-08-06 13:23:25 +0200242 ret = 0;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400243 while (nr_sects != 0) {
244 bio = bio_alloc(gfp_mask,
245 min(nr_sects, (sector_t)BIO_MAX_PAGES));
Dmitry Monakhov18edc8e2010-08-06 13:23:25 +0200246 if (!bio) {
247 ret = -ENOMEM;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400248 break;
Dmitry Monakhov18edc8e2010-08-06 13:23:25 +0200249 }
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400250
Kent Overstreet4f024f32013-10-11 15:44:27 -0700251 bio->bi_iter.bi_sector = sector;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400252 bio->bi_bdev = bdev;
253 bio->bi_end_io = bio_batch_end_io;
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200254 bio->bi_private = &bb;
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400255
Jens Axboe0341aaf2010-04-29 09:28:21 +0200256 while (nr_sects != 0) {
257 sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400258 ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
259 nr_sects -= ret >> 9;
260 sector += ret >> 9;
261 if (ret < (sz << 9))
262 break;
263 }
Dmitry Monakhov18edc8e2010-08-06 13:23:25 +0200264 ret = 0;
Lukas Czerner0aeea182011-03-11 10:23:53 +0100265 atomic_inc(&bb.done);
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400266 submit_bio(WRITE, bio);
267 }
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400268
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200269 /* Wait for bios in-flight */
Lukas Czerner0aeea182011-03-11 10:23:53 +0100270 if (!atomic_dec_and_test(&bb.done))
Vladimir Davydov55770222013-02-14 18:19:59 +0400271 wait_for_completion_io(&wait);
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400272
273 if (!test_bit(BIO_UPTODATE, &bb.flags))
274 /* One of bios in the batch was completed with error.*/
275 ret = -EIO;
276
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400277 return ret;
278}
Martin K. Petersen579e8f32012-09-18 12:19:28 -0400279
280/**
281 * blkdev_issue_zeroout - zero-fill a block range
282 * @bdev: blockdev to write
283 * @sector: start sector
284 * @nr_sects: number of sectors to write
285 * @gfp_mask: memory allocation flags (for bio_alloc)
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500286 * @discard: whether to discard the block range
Martin K. Petersen579e8f32012-09-18 12:19:28 -0400287 *
288 * Description:
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500289 * Zero-fill a block range. If the discard flag is set and the block
290 * device guarantees that subsequent READ operations to the block range
291 * in question will return zeroes, the blocks will be discarded. Should
292 * the discard request fail, if the discard flag is not set, or if
293 * discard_zeroes_data is not supported, this function will resort to
294 * zeroing the blocks manually, thus provisioning (allocating,
295 * anchoring) them. If the block device supports the WRITE SAME command
296 * blkdev_issue_zeroout() will use it to optimize the process of
297 * clearing the block range. Otherwise the zeroing will be performed
298 * using regular WRITE calls.
Martin K. Petersen579e8f32012-09-18 12:19:28 -0400299 */
300
301int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500302 sector_t nr_sects, gfp_t gfp_mask, bool discard)
Martin K. Petersen579e8f32012-09-18 12:19:28 -0400303{
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500304 struct request_queue *q = bdev_get_queue(bdev);
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500305
Martin K. Petersen9f9ee1f2015-02-05 10:14:54 -0700306 if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
307 blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
308 return 0;
Martin K. Petersend93ba7a2015-01-20 20:06:30 -0500309
Martin K. Petersen9f9ee1f2015-02-05 10:14:54 -0700310 if (bdev_write_same(bdev) &&
311 blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
312 ZERO_PAGE(0)) == 0)
313 return 0;
Martin K. Petersen579e8f32012-09-18 12:19:28 -0400314
315 return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
316}
Dmitry Monakhov3f14d792010-04-28 17:55:09 +0400317EXPORT_SYMBOL(blkdev_issue_zeroout);