]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - block/blk-lib.c
md: raid1: improve write behind
[karo-tx-linux.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 static struct bio *next_bio(struct bio *bio, unsigned int nr_pages,
13                 gfp_t gfp)
14 {
15         struct bio *new = bio_alloc(gfp, nr_pages);
16
17         if (bio) {
18                 bio_chain(bio, new);
19                 submit_bio(bio);
20         }
21
22         return new;
23 }
24
25 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
26                 sector_t nr_sects, gfp_t gfp_mask, int flags,
27                 struct bio **biop)
28 {
29         struct request_queue *q = bdev_get_queue(bdev);
30         struct bio *bio = *biop;
31         unsigned int granularity;
32         unsigned int op;
33         int alignment;
34         sector_t bs_mask;
35
36         if (!q)
37                 return -ENXIO;
38
39         if (flags & BLKDEV_DISCARD_SECURE) {
40                 if (flags & BLKDEV_DISCARD_ZERO)
41                         return -EOPNOTSUPP;
42                 if (!blk_queue_secure_erase(q))
43                         return -EOPNOTSUPP;
44                 op = REQ_OP_SECURE_ERASE;
45         } else {
46                 if (!blk_queue_discard(q))
47                         return -EOPNOTSUPP;
48                 if ((flags & BLKDEV_DISCARD_ZERO) &&
49                     !q->limits.discard_zeroes_data)
50                         return -EOPNOTSUPP;
51                 op = REQ_OP_DISCARD;
52         }
53
54         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
55         if ((sector | nr_sects) & bs_mask)
56                 return -EINVAL;
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) % granularity;
61
62         while (nr_sects) {
63                 unsigned int req_sects;
64                 sector_t end_sect, tmp;
65
66                 /* Make sure bi_size doesn't overflow */
67                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
68
69                 /**
70                  * If splitting a request, and the next starting sector would be
71                  * misaligned, stop the discard at the previous aligned sector.
72                  */
73                 end_sect = sector + req_sects;
74                 tmp = end_sect;
75                 if (req_sects < nr_sects &&
76                     sector_div(tmp, granularity) != alignment) {
77                         end_sect = end_sect - alignment;
78                         sector_div(end_sect, granularity);
79                         end_sect = end_sect * granularity + alignment;
80                         req_sects = end_sect - sector;
81                 }
82
83                 bio = next_bio(bio, 0, gfp_mask);
84                 bio->bi_iter.bi_sector = sector;
85                 bio->bi_bdev = bdev;
86                 bio_set_op_attrs(bio, op, 0);
87
88                 bio->bi_iter.bi_size = req_sects << 9;
89                 nr_sects -= req_sects;
90                 sector = end_sect;
91
92                 /*
93                  * We can loop for a long time in here, if someone does
94                  * full device discards (like mkfs). Be nice and allow
95                  * us to schedule out to avoid softlocking if preempt
96                  * is disabled.
97                  */
98                 cond_resched();
99         }
100
101         *biop = bio;
102         return 0;
103 }
104 EXPORT_SYMBOL(__blkdev_issue_discard);
105
106 /**
107  * blkdev_issue_discard - queue a discard
108  * @bdev:       blockdev to issue discard for
109  * @sector:     start sector
110  * @nr_sects:   number of sectors to discard
111  * @gfp_mask:   memory allocation flags (for bio_alloc)
112  * @flags:      BLKDEV_IFL_* flags to control behaviour
113  *
114  * Description:
115  *    Issue a discard request for the sectors in question.
116  */
117 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
118                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
119 {
120         struct bio *bio = NULL;
121         struct blk_plug plug;
122         int ret;
123
124         blk_start_plug(&plug);
125         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
126                         &bio);
127         if (!ret && bio) {
128                 ret = submit_bio_wait(bio);
129                 if (ret == -EOPNOTSUPP && !(flags & BLKDEV_DISCARD_ZERO))
130                         ret = 0;
131                 bio_put(bio);
132         }
133         blk_finish_plug(&plug);
134
135         return ret;
136 }
137 EXPORT_SYMBOL(blkdev_issue_discard);
138
139 /**
140  * __blkdev_issue_write_same - generate number of bios with same page
141  * @bdev:       target blockdev
142  * @sector:     start sector
143  * @nr_sects:   number of sectors to write
144  * @gfp_mask:   memory allocation flags (for bio_alloc)
145  * @page:       page containing data to write
146  * @biop:       pointer to anchor bio
147  *
148  * Description:
149  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
150  */
151 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
152                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
153                 struct bio **biop)
154 {
155         struct request_queue *q = bdev_get_queue(bdev);
156         unsigned int max_write_same_sectors;
157         struct bio *bio = *biop;
158         sector_t bs_mask;
159
160         if (!q)
161                 return -ENXIO;
162
163         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
164         if ((sector | nr_sects) & bs_mask)
165                 return -EINVAL;
166
167         if (!bdev_write_same(bdev))
168                 return -EOPNOTSUPP;
169
170         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
171         max_write_same_sectors = UINT_MAX >> 9;
172
173         while (nr_sects) {
174                 bio = next_bio(bio, 1, gfp_mask);
175                 bio->bi_iter.bi_sector = sector;
176                 bio->bi_bdev = bdev;
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                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
182
183                 if (nr_sects > max_write_same_sectors) {
184                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
185                         nr_sects -= max_write_same_sectors;
186                         sector += max_write_same_sectors;
187                 } else {
188                         bio->bi_iter.bi_size = nr_sects << 9;
189                         nr_sects = 0;
190                 }
191                 cond_resched();
192         }
193
194         *biop = bio;
195         return 0;
196 }
197
198 /**
199  * blkdev_issue_write_same - queue a write same operation
200  * @bdev:       target blockdev
201  * @sector:     start sector
202  * @nr_sects:   number of sectors to write
203  * @gfp_mask:   memory allocation flags (for bio_alloc)
204  * @page:       page containing data
205  *
206  * Description:
207  *    Issue a write same request for the sectors in question.
208  */
209 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
210                                 sector_t nr_sects, gfp_t gfp_mask,
211                                 struct page *page)
212 {
213         struct bio *bio = NULL;
214         struct blk_plug plug;
215         int ret;
216
217         blk_start_plug(&plug);
218         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
219                         &bio);
220         if (ret == 0 && bio) {
221                 ret = submit_bio_wait(bio);
222                 bio_put(bio);
223         }
224         blk_finish_plug(&plug);
225         return ret;
226 }
227 EXPORT_SYMBOL(blkdev_issue_write_same);
228
229 /**
230  * __blkdev_issue_write_zeroes - generate number of bios with WRITE ZEROES
231  * @bdev:       blockdev to issue
232  * @sector:     start sector
233  * @nr_sects:   number of sectors to write
234  * @gfp_mask:   memory allocation flags (for bio_alloc)
235  * @biop:       pointer to anchor bio
236  *
237  * Description:
238  *  Generate and issue number of bios(REQ_OP_WRITE_ZEROES) with zerofiled pages.
239  */
240 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
241                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
242                 struct bio **biop)
243 {
244         struct bio *bio = *biop;
245         unsigned int max_write_zeroes_sectors;
246         struct request_queue *q = bdev_get_queue(bdev);
247
248         if (!q)
249                 return -ENXIO;
250
251         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
252         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
253
254         if (max_write_zeroes_sectors == 0)
255                 return -EOPNOTSUPP;
256
257         while (nr_sects) {
258                 bio = next_bio(bio, 0, gfp_mask);
259                 bio->bi_iter.bi_sector = sector;
260                 bio->bi_bdev = bdev;
261                 bio_set_op_attrs(bio, REQ_OP_WRITE_ZEROES, 0);
262
263                 if (nr_sects > max_write_zeroes_sectors) {
264                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
265                         nr_sects -= max_write_zeroes_sectors;
266                         sector += max_write_zeroes_sectors;
267                 } else {
268                         bio->bi_iter.bi_size = nr_sects << 9;
269                         nr_sects = 0;
270                 }
271                 cond_resched();
272         }
273
274         *biop = bio;
275         return 0;
276 }
277
278 /**
279  * __blkdev_issue_zeroout - generate number of zero filed write bios
280  * @bdev:       blockdev to issue
281  * @sector:     start sector
282  * @nr_sects:   number of sectors to write
283  * @gfp_mask:   memory allocation flags (for bio_alloc)
284  * @biop:       pointer to anchor bio
285  * @discard:    discard flag
286  *
287  * Description:
288  *  Generate and issue number of bios with zerofiled pages.
289  */
290 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
291                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
292                 bool discard)
293 {
294         int ret;
295         int bi_size = 0;
296         struct bio *bio = *biop;
297         unsigned int sz;
298         sector_t bs_mask;
299
300         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
301         if ((sector | nr_sects) & bs_mask)
302                 return -EINVAL;
303
304         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
305                         biop);
306         if (ret == 0 || (ret && ret != -EOPNOTSUPP))
307                 goto out;
308
309         ret = 0;
310         while (nr_sects != 0) {
311                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
312                                 gfp_mask);
313                 bio->bi_iter.bi_sector = sector;
314                 bio->bi_bdev   = bdev;
315                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
316
317                 while (nr_sects != 0) {
318                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
319                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
320                         nr_sects -= bi_size >> 9;
321                         sector += bi_size >> 9;
322                         if (bi_size < (sz << 9))
323                                 break;
324                 }
325                 cond_resched();
326         }
327
328         *biop = bio;
329 out:
330         return ret;
331 }
332 EXPORT_SYMBOL(__blkdev_issue_zeroout);
333
334 /**
335  * blkdev_issue_zeroout - zero-fill a block range
336  * @bdev:       blockdev to write
337  * @sector:     start sector
338  * @nr_sects:   number of sectors to write
339  * @gfp_mask:   memory allocation flags (for bio_alloc)
340  * @discard:    whether to discard the block range
341  *
342  * Description:
343  *  Zero-fill a block range.  If the discard flag is set and the block
344  *  device guarantees that subsequent READ operations to the block range
345  *  in question will return zeroes, the blocks will be discarded. Should
346  *  the discard request fail, if the discard flag is not set, or if
347  *  discard_zeroes_data is not supported, this function will resort to
348  *  zeroing the blocks manually, thus provisioning (allocating,
349  *  anchoring) them. If the block device supports WRITE ZEROES or WRITE SAME
350  *  command(s), blkdev_issue_zeroout() will use it to optimize the process of
351  *  clearing the block range. Otherwise the zeroing will be performed
352  *  using regular WRITE calls.
353  */
354 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
355                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
356 {
357         int ret;
358         struct bio *bio = NULL;
359         struct blk_plug plug;
360
361         if (discard) {
362                 if (!blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask,
363                                 BLKDEV_DISCARD_ZERO))
364                         return 0;
365         }
366
367         if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
368                         ZERO_PAGE(0)))
369                 return 0;
370
371         blk_start_plug(&plug);
372         ret = __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask,
373                         &bio, discard);
374         if (ret == 0 && bio) {
375                 ret = submit_bio_wait(bio);
376                 bio_put(bio);
377         }
378         blk_finish_plug(&plug);
379
380         return ret;
381 }
382 EXPORT_SYMBOL(blkdev_issue_zeroout);