Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[sfrench/cifs-2.6.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         unsigned int max_discard_sectors;
47         unsigned int granularity, alignment, mask;
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         mask = granularity - 1;
61         alignment = (bdev_discard_alignment(bdev) >> 9) & mask;
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         max_discard_sectors = round_down(max_discard_sectors, granularity);
69         if (unlikely(!max_discard_sectors)) {
70                 /* Avoid infinite loop below. Being cautious never hurts. */
71                 return -EOPNOTSUPP;
72         }
73
74         if (flags & BLKDEV_DISCARD_SECURE) {
75                 if (!blk_queue_secdiscard(q))
76                         return -EOPNOTSUPP;
77                 type |= REQ_SECURE;
78         }
79
80         atomic_set(&bb.done, 1);
81         bb.flags = 1 << BIO_UPTODATE;
82         bb.wait = &wait;
83
84         while (nr_sects) {
85                 unsigned int req_sects;
86                 sector_t end_sect;
87
88                 bio = bio_alloc(gfp_mask, 1);
89                 if (!bio) {
90                         ret = -ENOMEM;
91                         break;
92                 }
93
94                 req_sects = min_t(sector_t, nr_sects, max_discard_sectors);
95
96                 /*
97                  * If splitting a request, and the next starting sector would be
98                  * misaligned, stop the discard at the previous aligned sector.
99                  */
100                 end_sect = sector + req_sects;
101                 if (req_sects < nr_sects && (end_sect & mask) != alignment) {
102                         end_sect =
103                                 round_down(end_sect - alignment, granularity)
104                                 + alignment;
105                         req_sects = end_sect - sector;
106                 }
107
108                 bio->bi_sector = sector;
109                 bio->bi_end_io = bio_batch_end_io;
110                 bio->bi_bdev = bdev;
111                 bio->bi_private = &bb;
112
113                 bio->bi_size = req_sects << 9;
114                 nr_sects -= req_sects;
115                 sector = end_sect;
116
117                 atomic_inc(&bb.done);
118                 submit_bio(type, bio);
119         }
120
121         /* Wait for bios in-flight */
122         if (!atomic_dec_and_test(&bb.done))
123                 wait_for_completion(&wait);
124
125         if (!test_bit(BIO_UPTODATE, &bb.flags))
126                 ret = -EIO;
127
128         return ret;
129 }
130 EXPORT_SYMBOL(blkdev_issue_discard);
131
132 /**
133  * blkdev_issue_write_same - queue a write same operation
134  * @bdev:       target blockdev
135  * @sector:     start sector
136  * @nr_sects:   number of sectors to write
137  * @gfp_mask:   memory allocation flags (for bio_alloc)
138  * @page:       page containing data to write
139  *
140  * Description:
141  *    Issue a write same request for the sectors in question.
142  */
143 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
144                             sector_t nr_sects, gfp_t gfp_mask,
145                             struct page *page)
146 {
147         DECLARE_COMPLETION_ONSTACK(wait);
148         struct request_queue *q = bdev_get_queue(bdev);
149         unsigned int max_write_same_sectors;
150         struct bio_batch bb;
151         struct bio *bio;
152         int ret = 0;
153
154         if (!q)
155                 return -ENXIO;
156
157         max_write_same_sectors = q->limits.max_write_same_sectors;
158
159         if (max_write_same_sectors == 0)
160                 return -EOPNOTSUPP;
161
162         atomic_set(&bb.done, 1);
163         bb.flags = 1 << BIO_UPTODATE;
164         bb.wait = &wait;
165
166         while (nr_sects) {
167                 bio = bio_alloc(gfp_mask, 1);
168                 if (!bio) {
169                         ret = -ENOMEM;
170                         break;
171                 }
172
173                 bio->bi_sector = sector;
174                 bio->bi_end_io = bio_batch_end_io;
175                 bio->bi_bdev = bdev;
176                 bio->bi_private = &bb;
177                 bio->bi_vcnt = 1;
178                 bio->bi_io_vec->bv_page = page;
179                 bio->bi_io_vec->bv_offset = 0;
180                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
181
182                 if (nr_sects > max_write_same_sectors) {
183                         bio->bi_size = max_write_same_sectors << 9;
184                         nr_sects -= max_write_same_sectors;
185                         sector += max_write_same_sectors;
186                 } else {
187                         bio->bi_size = nr_sects << 9;
188                         nr_sects = 0;
189                 }
190
191                 atomic_inc(&bb.done);
192                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
193         }
194
195         /* Wait for bios in-flight */
196         if (!atomic_dec_and_test(&bb.done))
197                 wait_for_completion(&wait);
198
199         if (!test_bit(BIO_UPTODATE, &bb.flags))
200                 ret = -ENOTSUPP;
201
202         return ret;
203 }
204 EXPORT_SYMBOL(blkdev_issue_write_same);
205
206 /**
207  * blkdev_issue_zeroout - generate number of zero filed write bios
208  * @bdev:       blockdev to issue
209  * @sector:     start sector
210  * @nr_sects:   number of sectors to write
211  * @gfp_mask:   memory allocation flags (for bio_alloc)
212  *
213  * Description:
214  *  Generate and issue number of bios with zerofiled pages.
215  */
216
217 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
218                         sector_t nr_sects, gfp_t gfp_mask)
219 {
220         int ret;
221         struct bio *bio;
222         struct bio_batch bb;
223         unsigned int sz;
224         DECLARE_COMPLETION_ONSTACK(wait);
225
226         atomic_set(&bb.done, 1);
227         bb.flags = 1 << BIO_UPTODATE;
228         bb.wait = &wait;
229
230         ret = 0;
231         while (nr_sects != 0) {
232                 bio = bio_alloc(gfp_mask,
233                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
234                 if (!bio) {
235                         ret = -ENOMEM;
236                         break;
237                 }
238
239                 bio->bi_sector = sector;
240                 bio->bi_bdev   = bdev;
241                 bio->bi_end_io = bio_batch_end_io;
242                 bio->bi_private = &bb;
243
244                 while (nr_sects != 0) {
245                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
246                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
247                         nr_sects -= ret >> 9;
248                         sector += ret >> 9;
249                         if (ret < (sz << 9))
250                                 break;
251                 }
252                 ret = 0;
253                 atomic_inc(&bb.done);
254                 submit_bio(WRITE, bio);
255         }
256
257         /* Wait for bios in-flight */
258         if (!atomic_dec_and_test(&bb.done))
259                 wait_for_completion(&wait);
260
261         if (!test_bit(BIO_UPTODATE, &bb.flags))
262                 /* One of bios in the batch was completed with error.*/
263                 ret = -EIO;
264
265         return ret;
266 }
267
268 /**
269  * blkdev_issue_zeroout - zero-fill a block range
270  * @bdev:       blockdev to write
271  * @sector:     start sector
272  * @nr_sects:   number of sectors to write
273  * @gfp_mask:   memory allocation flags (for bio_alloc)
274  *
275  * Description:
276  *  Generate and issue number of bios with zerofiled pages.
277  */
278
279 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
280                          sector_t nr_sects, gfp_t gfp_mask)
281 {
282         if (bdev_write_same(bdev)) {
283                 unsigned char bdn[BDEVNAME_SIZE];
284
285                 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
286                                              ZERO_PAGE(0)))
287                         return 0;
288
289                 bdevname(bdev, bdn);
290                 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
291         }
292
293         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
294 }
295 EXPORT_SYMBOL(blkdev_issue_zeroout);