cgroup: fix cgroup_path() vs rename() race
[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         struct blk_plug plug;
52
53         if (!q)
54                 return -ENXIO;
55
56         if (!blk_queue_discard(q))
57                 return -EOPNOTSUPP;
58
59         /* Zero-sector (unknown) and one-sector granularities are the same.  */
60         granularity = max(q->limits.discard_granularity >> 9, 1U);
61         alignment = bdev_discard_alignment(bdev) >> 9;
62         alignment = sector_div(alignment, granularity);
63
64         /*
65          * Ensure that max_discard_sectors is of the proper
66          * granularity, so that requests stay aligned after a split.
67          */
68         max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
69         sector_div(max_discard_sectors, granularity);
70         max_discard_sectors *= granularity;
71         if (unlikely(!max_discard_sectors)) {
72                 /* Avoid infinite loop below. Being cautious never hurts. */
73                 return -EOPNOTSUPP;
74         }
75
76         if (flags & BLKDEV_DISCARD_SECURE) {
77                 if (!blk_queue_secdiscard(q))
78                         return -EOPNOTSUPP;
79                 type |= REQ_SECURE;
80         }
81
82         atomic_set(&bb.done, 1);
83         bb.flags = 1 << BIO_UPTODATE;
84         bb.wait = &wait;
85
86         blk_start_plug(&plug);
87         while (nr_sects) {
88                 unsigned int req_sects;
89                 sector_t end_sect, tmp;
90
91                 bio = bio_alloc(gfp_mask, 1);
92                 if (!bio) {
93                         ret = -ENOMEM;
94                         break;
95                 }
96
97                 req_sects = min_t(sector_t, nr_sects, max_discard_sectors);
98
99                 /*
100                  * If splitting a request, and the next starting sector would be
101                  * misaligned, stop the discard at the previous aligned sector.
102                  */
103                 end_sect = sector + req_sects;
104                 tmp = end_sect;
105                 if (req_sects < nr_sects &&
106                     sector_div(tmp, granularity) != alignment) {
107                         end_sect = end_sect - alignment;
108                         sector_div(end_sect, granularity);
109                         end_sect = end_sect * granularity + alignment;
110                         req_sects = end_sect - sector;
111                 }
112
113                 bio->bi_sector = sector;
114                 bio->bi_end_io = bio_batch_end_io;
115                 bio->bi_bdev = bdev;
116                 bio->bi_private = &bb;
117
118                 bio->bi_size = req_sects << 9;
119                 nr_sects -= req_sects;
120                 sector = end_sect;
121
122                 atomic_inc(&bb.done);
123                 submit_bio(type, bio);
124         }
125         blk_finish_plug(&plug);
126
127         /* Wait for bios in-flight */
128         if (!atomic_dec_and_test(&bb.done))
129                 wait_for_completion_io(&wait);
130
131         if (!test_bit(BIO_UPTODATE, &bb.flags))
132                 ret = -EIO;
133
134         return ret;
135 }
136 EXPORT_SYMBOL(blkdev_issue_discard);
137
138 /**
139  * blkdev_issue_write_same - queue a write same operation
140  * @bdev:       target blockdev
141  * @sector:     start sector
142  * @nr_sects:   number of sectors to write
143  * @gfp_mask:   memory allocation flags (for bio_alloc)
144  * @page:       page containing data to write
145  *
146  * Description:
147  *    Issue a write same request for the sectors in question.
148  */
149 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
150                             sector_t nr_sects, gfp_t gfp_mask,
151                             struct page *page)
152 {
153         DECLARE_COMPLETION_ONSTACK(wait);
154         struct request_queue *q = bdev_get_queue(bdev);
155         unsigned int max_write_same_sectors;
156         struct bio_batch bb;
157         struct bio *bio;
158         int ret = 0;
159
160         if (!q)
161                 return -ENXIO;
162
163         max_write_same_sectors = q->limits.max_write_same_sectors;
164
165         if (max_write_same_sectors == 0)
166                 return -EOPNOTSUPP;
167
168         atomic_set(&bb.done, 1);
169         bb.flags = 1 << BIO_UPTODATE;
170         bb.wait = &wait;
171
172         while (nr_sects) {
173                 bio = bio_alloc(gfp_mask, 1);
174                 if (!bio) {
175                         ret = -ENOMEM;
176                         break;
177                 }
178
179                 bio->bi_sector = sector;
180                 bio->bi_end_io = bio_batch_end_io;
181                 bio->bi_bdev = bdev;
182                 bio->bi_private = &bb;
183                 bio->bi_vcnt = 1;
184                 bio->bi_io_vec->bv_page = page;
185                 bio->bi_io_vec->bv_offset = 0;
186                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
187
188                 if (nr_sects > max_write_same_sectors) {
189                         bio->bi_size = max_write_same_sectors << 9;
190                         nr_sects -= max_write_same_sectors;
191                         sector += max_write_same_sectors;
192                 } else {
193                         bio->bi_size = nr_sects << 9;
194                         nr_sects = 0;
195                 }
196
197                 atomic_inc(&bb.done);
198                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
199         }
200
201         /* Wait for bios in-flight */
202         if (!atomic_dec_and_test(&bb.done))
203                 wait_for_completion_io(&wait);
204
205         if (!test_bit(BIO_UPTODATE, &bb.flags))
206                 ret = -ENOTSUPP;
207
208         return ret;
209 }
210 EXPORT_SYMBOL(blkdev_issue_write_same);
211
212 /**
213  * blkdev_issue_zeroout - generate number of zero filed write bios
214  * @bdev:       blockdev to issue
215  * @sector:     start sector
216  * @nr_sects:   number of sectors to write
217  * @gfp_mask:   memory allocation flags (for bio_alloc)
218  *
219  * Description:
220  *  Generate and issue number of bios with zerofiled pages.
221  */
222
223 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
224                         sector_t nr_sects, gfp_t gfp_mask)
225 {
226         int ret;
227         struct bio *bio;
228         struct bio_batch bb;
229         unsigned int sz;
230         DECLARE_COMPLETION_ONSTACK(wait);
231
232         atomic_set(&bb.done, 1);
233         bb.flags = 1 << BIO_UPTODATE;
234         bb.wait = &wait;
235
236         ret = 0;
237         while (nr_sects != 0) {
238                 bio = bio_alloc(gfp_mask,
239                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
240                 if (!bio) {
241                         ret = -ENOMEM;
242                         break;
243                 }
244
245                 bio->bi_sector = sector;
246                 bio->bi_bdev   = bdev;
247                 bio->bi_end_io = bio_batch_end_io;
248                 bio->bi_private = &bb;
249
250                 while (nr_sects != 0) {
251                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
252                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
253                         nr_sects -= ret >> 9;
254                         sector += ret >> 9;
255                         if (ret < (sz << 9))
256                                 break;
257                 }
258                 ret = 0;
259                 atomic_inc(&bb.done);
260                 submit_bio(WRITE, bio);
261         }
262
263         /* Wait for bios in-flight */
264         if (!atomic_dec_and_test(&bb.done))
265                 wait_for_completion_io(&wait);
266
267         if (!test_bit(BIO_UPTODATE, &bb.flags))
268                 /* One of bios in the batch was completed with error.*/
269                 ret = -EIO;
270
271         return ret;
272 }
273
274 /**
275  * blkdev_issue_zeroout - zero-fill a block range
276  * @bdev:       blockdev to write
277  * @sector:     start sector
278  * @nr_sects:   number of sectors to write
279  * @gfp_mask:   memory allocation flags (for bio_alloc)
280  *
281  * Description:
282  *  Generate and issue number of bios with zerofiled pages.
283  */
284
285 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
286                          sector_t nr_sects, gfp_t gfp_mask)
287 {
288         if (bdev_write_same(bdev)) {
289                 unsigned char bdn[BDEVNAME_SIZE];
290
291                 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
292                                              ZERO_PAGE(0)))
293                         return 0;
294
295                 bdevname(bdev, bdn);
296                 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
297         }
298
299         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
300 }
301 EXPORT_SYMBOL(blkdev_issue_zeroout);