block: discard granularity might not be power of 2
[linux-3.10.git] / block / blk-lib.c
1 /*
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
12 struct bio_batch {
13         atomic_t                done;
14         unsigned long           flags;
15         struct completion       *wait;
16 };
17
18 static void bio_batch_end_io(struct bio *bio, int err)
19 {
20         struct bio_batch *bb = bio->bi_private;
21
22         if (err && (err != -EOPNOTSUPP))
23                 clear_bit(BIO_UPTODATE, &bb->flags);
24         if (atomic_dec_and_test(&bb->done))
25                 complete(bb->wait);
26         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  */
40 int 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);
45         int type = REQ_WRITE | REQ_DISCARD;
46         sector_t max_discard_sectors;
47         sector_t granularity, alignment;
48         struct bio_batch bb;
49         struct bio *bio;
50         int ret = 0;
51
52         if (!q)
53                 return -ENXIO;
54
55         if (!blk_queue_discard(q))
56                 return -EOPNOTSUPP;
57
58         /* Zero-sector (unknown) and one-sector granularities are the same.  */
59         granularity = max(q->limits.discard_granularity >> 9, 1U);
60         alignment = bdev_discard_alignment(bdev) >> 9;
61         alignment = sector_div(alignment, granularity);
62
63         /*
64          * Ensure that max_discard_sectors is of the proper
65          * granularity, so that requests stay aligned after a split.
66          */
67         max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
68         sector_div(max_discard_sectors, granularity);
69         max_discard_sectors *= granularity;
70         if (unlikely(!max_discard_sectors)) {
71                 /* Avoid infinite loop below. Being cautious never hurts. */
72                 return -EOPNOTSUPP;
73         }
74
75         if (flags & BLKDEV_DISCARD_SECURE) {
76                 if (!blk_queue_secdiscard(q))
77                         return -EOPNOTSUPP;
78                 type |= REQ_SECURE;
79         }
80
81         atomic_set(&bb.done, 1);
82         bb.flags = 1 << BIO_UPTODATE;
83         bb.wait = &wait;
84
85         while (nr_sects) {
86                 unsigned int req_sects;
87                 sector_t end_sect, tmp;
88
89                 bio = bio_alloc(gfp_mask, 1);
90                 if (!bio) {
91                         ret = -ENOMEM;
92                         break;
93                 }
94
95                 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;
102                 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;
108                         req_sects = end_sect - sector;
109                 }
110
111                 bio->bi_sector = sector;
112                 bio->bi_end_io = bio_batch_end_io;
113                 bio->bi_bdev = bdev;
114                 bio->bi_private = &bb;
115
116                 bio->bi_size = req_sects << 9;
117                 nr_sects -= req_sects;
118                 sector = end_sect;
119
120                 atomic_inc(&bb.done);
121                 submit_bio(type, bio);
122         }
123
124         /* Wait for bios in-flight */
125         if (!atomic_dec_and_test(&bb.done))
126                 wait_for_completion(&wait);
127
128         if (!test_bit(BIO_UPTODATE, &bb.flags))
129                 ret = -EIO;
130
131         return ret;
132 }
133 EXPORT_SYMBOL(blkdev_issue_discard);
134
135 /**
136  * blkdev_issue_write_same - queue a write same operation
137  * @bdev:       target blockdev
138  * @sector:     start sector
139  * @nr_sects:   number of sectors to write
140  * @gfp_mask:   memory allocation flags (for bio_alloc)
141  * @page:       page containing data to write
142  *
143  * Description:
144  *    Issue a write same request for the sectors in question.
145  */
146 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
147                             sector_t nr_sects, gfp_t gfp_mask,
148                             struct page *page)
149 {
150         DECLARE_COMPLETION_ONSTACK(wait);
151         struct request_queue *q = bdev_get_queue(bdev);
152         unsigned int max_write_same_sectors;
153         struct bio_batch bb;
154         struct bio *bio;
155         int ret = 0;
156
157         if (!q)
158                 return -ENXIO;
159
160         max_write_same_sectors = q->limits.max_write_same_sectors;
161
162         if (max_write_same_sectors == 0)
163                 return -EOPNOTSUPP;
164
165         atomic_set(&bb.done, 1);
166         bb.flags = 1 << BIO_UPTODATE;
167         bb.wait = &wait;
168
169         while (nr_sects) {
170                 bio = bio_alloc(gfp_mask, 1);
171                 if (!bio) {
172                         ret = -ENOMEM;
173                         break;
174                 }
175
176                 bio->bi_sector = sector;
177                 bio->bi_end_io = bio_batch_end_io;
178                 bio->bi_bdev = bdev;
179                 bio->bi_private = &bb;
180                 bio->bi_vcnt = 1;
181                 bio->bi_io_vec->bv_page = page;
182                 bio->bi_io_vec->bv_offset = 0;
183                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
184
185                 if (nr_sects > max_write_same_sectors) {
186                         bio->bi_size = max_write_same_sectors << 9;
187                         nr_sects -= max_write_same_sectors;
188                         sector += max_write_same_sectors;
189                 } else {
190                         bio->bi_size = nr_sects << 9;
191                         nr_sects = 0;
192                 }
193
194                 atomic_inc(&bb.done);
195                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
196         }
197
198         /* Wait for bios in-flight */
199         if (!atomic_dec_and_test(&bb.done))
200                 wait_for_completion(&wait);
201
202         if (!test_bit(BIO_UPTODATE, &bb.flags))
203                 ret = -ENOTSUPP;
204
205         return ret;
206 }
207 EXPORT_SYMBOL(blkdev_issue_write_same);
208
209 /**
210  * blkdev_issue_zeroout - generate number of zero filed write bios
211  * @bdev:       blockdev to issue
212  * @sector:     start sector
213  * @nr_sects:   number of sectors to write
214  * @gfp_mask:   memory allocation flags (for bio_alloc)
215  *
216  * Description:
217  *  Generate and issue number of bios with zerofiled pages.
218  */
219
220 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
221                         sector_t nr_sects, gfp_t gfp_mask)
222 {
223         int ret;
224         struct bio *bio;
225         struct bio_batch bb;
226         unsigned int sz;
227         DECLARE_COMPLETION_ONSTACK(wait);
228
229         atomic_set(&bb.done, 1);
230         bb.flags = 1 << BIO_UPTODATE;
231         bb.wait = &wait;
232
233         ret = 0;
234         while (nr_sects != 0) {
235                 bio = bio_alloc(gfp_mask,
236                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
237                 if (!bio) {
238                         ret = -ENOMEM;
239                         break;
240                 }
241
242                 bio->bi_sector = sector;
243                 bio->bi_bdev   = bdev;
244                 bio->bi_end_io = bio_batch_end_io;
245                 bio->bi_private = &bb;
246
247                 while (nr_sects != 0) {
248                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
249                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
250                         nr_sects -= ret >> 9;
251                         sector += ret >> 9;
252                         if (ret < (sz << 9))
253                                 break;
254                 }
255                 ret = 0;
256                 atomic_inc(&bb.done);
257                 submit_bio(WRITE, bio);
258         }
259
260         /* Wait for bios in-flight */
261         if (!atomic_dec_and_test(&bb.done))
262                 wait_for_completion(&wait);
263
264         if (!test_bit(BIO_UPTODATE, &bb.flags))
265                 /* One of bios in the batch was completed with error.*/
266                 ret = -EIO;
267
268         return ret;
269 }
270
271 /**
272  * blkdev_issue_zeroout - zero-fill a block range
273  * @bdev:       blockdev to write
274  * @sector:     start sector
275  * @nr_sects:   number of sectors to write
276  * @gfp_mask:   memory allocation flags (for bio_alloc)
277  *
278  * Description:
279  *  Generate and issue number of bios with zerofiled pages.
280  */
281
282 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
283                          sector_t nr_sects, gfp_t gfp_mask)
284 {
285         if (bdev_write_same(bdev)) {
286                 unsigned char bdn[BDEVNAME_SIZE];
287
288                 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
289                                              ZERO_PAGE(0)))
290                         return 0;
291
292                 bdevname(bdev, bdn);
293                 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
294         }
295
296         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
297 }
298 EXPORT_SYMBOL(blkdev_issue_zeroout);