Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to generic helpers functions
4 */
5#include <linux/kernel.h>
6#include <linux/module.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/scatterlist.h>
10
11#include "blk.h"
12
13static sector_t bio_discard_limit(struct block_device *bdev, sector_t sector)
14{
15 unsigned int discard_granularity = bdev_discard_granularity(bdev);
16 sector_t granularity_aligned_sector;
17
18 if (bdev_is_partition(bdev))
19 sector += bdev->bd_start_sect;
20
21 granularity_aligned_sector =
22 round_up(sector, discard_granularity >> SECTOR_SHIFT);
23
24 /*
25 * Make sure subsequent bios start aligned to the discard granularity if
26 * it needs to be split.
27 */
28 if (granularity_aligned_sector != sector)
29 return granularity_aligned_sector - sector;
30
31 /*
32 * Align the bio size to the discard granularity to make splitting the bio
33 * at discard granularity boundaries easier in the driver if needed.
34 */
35 return round_down(UINT_MAX, discard_granularity) >> SECTOR_SHIFT;
36}
37
38int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
39 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop)
40{
41 struct bio *bio = *biop;
42 sector_t bs_mask;
43
44 if (bdev_read_only(bdev))
45 return -EPERM;
46 if (!bdev_max_discard_sectors(bdev))
47 return -EOPNOTSUPP;
48
49 /* In case the discard granularity isn't set by buggy device driver */
50 if (WARN_ON_ONCE(!bdev_discard_granularity(bdev))) {
51 pr_err_ratelimited("%pg: Error: discard_granularity is 0.\n",
52 bdev);
53 return -EOPNOTSUPP;
54 }
55
56 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
57 if ((sector | nr_sects) & bs_mask)
58 return -EINVAL;
59
60 if (!nr_sects)
61 return -EINVAL;
62
63 while (nr_sects) {
64 sector_t req_sects =
65 min(nr_sects, bio_discard_limit(bdev, sector));
66
67 bio = blk_next_bio(bio, bdev, 0, REQ_OP_DISCARD, gfp_mask);
68 bio->bi_iter.bi_sector = sector;
69 bio->bi_iter.bi_size = req_sects << 9;
70 sector += req_sects;
71 nr_sects -= req_sects;
72
73 /*
74 * We can loop for a long time in here, if someone does
75 * full device discards (like mkfs). Be nice and allow
76 * us to schedule out to avoid softlocking if preempt
77 * is disabled.
78 */
79 cond_resched();
80 }
81
82 *biop = bio;
83 return 0;
84}
85EXPORT_SYMBOL(__blkdev_issue_discard);
86
87/**
88 * blkdev_issue_discard - queue a discard
89 * @bdev: blockdev to issue discard for
90 * @sector: start sector
91 * @nr_sects: number of sectors to discard
92 * @gfp_mask: memory allocation flags (for bio_alloc)
93 *
94 * Description:
95 * Issue a discard request for the sectors in question.
96 */
97int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
98 sector_t nr_sects, gfp_t gfp_mask)
99{
100 struct bio *bio = NULL;
101 struct blk_plug plug;
102 int ret;
103
104 blk_start_plug(&plug);
105 ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, &bio);
106 if (!ret && bio) {
107 ret = submit_bio_wait(bio);
108 if (ret == -EOPNOTSUPP)
109 ret = 0;
110 bio_put(bio);
111 }
112 blk_finish_plug(&plug);
113
114 return ret;
115}
116EXPORT_SYMBOL(blkdev_issue_discard);
117
118static int __blkdev_issue_write_zeroes(struct block_device *bdev,
119 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
120 struct bio **biop, unsigned flags)
121{
122 struct bio *bio = *biop;
123 unsigned int max_write_zeroes_sectors;
124
125 if (bdev_read_only(bdev))
126 return -EPERM;
127
128 /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
129 max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
130
131 if (max_write_zeroes_sectors == 0)
132 return -EOPNOTSUPP;
133
134 while (nr_sects) {
135 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
136 bio->bi_iter.bi_sector = sector;
137 if (flags & BLKDEV_ZERO_NOUNMAP)
138 bio->bi_opf |= REQ_NOUNMAP;
139
140 if (nr_sects > max_write_zeroes_sectors) {
141 bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
142 nr_sects -= max_write_zeroes_sectors;
143 sector += max_write_zeroes_sectors;
144 } else {
145 bio->bi_iter.bi_size = nr_sects << 9;
146 nr_sects = 0;
147 }
148 cond_resched();
149 }
150
151 *biop = bio;
152 return 0;
153}
154
155/*
156 * Convert a number of 512B sectors to a number of pages.
157 * The result is limited to a number of pages that can fit into a BIO.
158 * Also make sure that the result is always at least 1 (page) for the cases
159 * where nr_sects is lower than the number of sectors in a page.
160 */
161static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
162{
163 sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
164
165 return min(pages, (sector_t)BIO_MAX_VECS);
166}
167
168static int __blkdev_issue_zero_pages(struct block_device *bdev,
169 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
170 struct bio **biop)
171{
172 struct bio *bio = *biop;
173 int bi_size = 0;
174 unsigned int sz;
175
176 if (bdev_read_only(bdev))
177 return -EPERM;
178
179 while (nr_sects != 0) {
180 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
181 REQ_OP_WRITE, gfp_mask);
182 bio->bi_iter.bi_sector = sector;
183
184 while (nr_sects != 0) {
185 sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
186 bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
187 nr_sects -= bi_size >> 9;
188 sector += bi_size >> 9;
189 if (bi_size < sz)
190 break;
191 }
192 cond_resched();
193 }
194
195 *biop = bio;
196 return 0;
197}
198
199/**
200 * __blkdev_issue_zeroout - generate number of zero filed write bios
201 * @bdev: blockdev to issue
202 * @sector: start sector
203 * @nr_sects: number of sectors to write
204 * @gfp_mask: memory allocation flags (for bio_alloc)
205 * @biop: pointer to anchor bio
206 * @flags: controls detailed behavior
207 *
208 * Description:
209 * Zero-fill a block range, either using hardware offload or by explicitly
210 * writing zeroes to the device.
211 *
212 * If a device is using logical block provisioning, the underlying space will
213 * not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
214 *
215 * If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
216 * -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
217 */
218int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
219 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
220 unsigned flags)
221{
222 int ret;
223 sector_t bs_mask;
224
225 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
226 if ((sector | nr_sects) & bs_mask)
227 return -EINVAL;
228
229 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
230 biop, flags);
231 if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
232 return ret;
233
234 return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
235 biop);
236}
237EXPORT_SYMBOL(__blkdev_issue_zeroout);
238
239/**
240 * blkdev_issue_zeroout - zero-fill a block range
241 * @bdev: blockdev to write
242 * @sector: start sector
243 * @nr_sects: number of sectors to write
244 * @gfp_mask: memory allocation flags (for bio_alloc)
245 * @flags: controls detailed behavior
246 *
247 * Description:
248 * Zero-fill a block range, either using hardware offload or by explicitly
249 * writing zeroes to the device. See __blkdev_issue_zeroout() for the
250 * valid values for %flags.
251 */
252int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
253 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
254{
255 int ret = 0;
256 sector_t bs_mask;
257 struct bio *bio;
258 struct blk_plug plug;
259 bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
260
261 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
262 if ((sector | nr_sects) & bs_mask)
263 return -EINVAL;
264
265retry:
266 bio = NULL;
267 blk_start_plug(&plug);
268 if (try_write_zeroes) {
269 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
270 gfp_mask, &bio, flags);
271 } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
272 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
273 gfp_mask, &bio);
274 } else {
275 /* No zeroing offload support */
276 ret = -EOPNOTSUPP;
277 }
278 if (ret == 0 && bio) {
279 ret = submit_bio_wait(bio);
280 bio_put(bio);
281 }
282 blk_finish_plug(&plug);
283 if (ret && try_write_zeroes) {
284 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
285 try_write_zeroes = false;
286 goto retry;
287 }
288 if (!bdev_write_zeroes_sectors(bdev)) {
289 /*
290 * Zeroing offload support was indicated, but the
291 * device reported ILLEGAL REQUEST (for some devices
292 * there is no non-destructive way to verify whether
293 * WRITE ZEROES is actually supported).
294 */
295 ret = -EOPNOTSUPP;
296 }
297 }
298
299 return ret;
300}
301EXPORT_SYMBOL(blkdev_issue_zeroout);
302
303int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector,
304 sector_t nr_sects, gfp_t gfp)
305{
306 sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
307 unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev);
308 struct bio *bio = NULL;
309 struct blk_plug plug;
310 int ret = 0;
311
312 /* make sure that "len << SECTOR_SHIFT" doesn't overflow */
313 if (max_sectors > UINT_MAX >> SECTOR_SHIFT)
314 max_sectors = UINT_MAX >> SECTOR_SHIFT;
315 max_sectors &= ~bs_mask;
316
317 if (max_sectors == 0)
318 return -EOPNOTSUPP;
319 if ((sector | nr_sects) & bs_mask)
320 return -EINVAL;
321 if (bdev_read_only(bdev))
322 return -EPERM;
323
324 blk_start_plug(&plug);
325 for (;;) {
326 unsigned int len = min_t(sector_t, nr_sects, max_sectors);
327
328 bio = blk_next_bio(bio, bdev, 0, REQ_OP_SECURE_ERASE, gfp);
329 bio->bi_iter.bi_sector = sector;
330 bio->bi_iter.bi_size = len << SECTOR_SHIFT;
331
332 sector += len;
333 nr_sects -= len;
334 if (!nr_sects) {
335 ret = submit_bio_wait(bio);
336 bio_put(bio);
337 break;
338 }
339 cond_resched();
340 }
341 blk_finish_plug(&plug);
342
343 return ret;
344}
345EXPORT_SYMBOL(blkdev_issue_secure_erase);
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
12struct bio_batch {
13 atomic_t done;
14 int error;
15 struct completion *wait;
16};
17
18static void bio_batch_end_io(struct bio *bio)
19{
20 struct bio_batch *bb = bio->bi_private;
21
22 if (bio->bi_error && bio->bi_error != -EOPNOTSUPP)
23 bb->error = bio->bi_error;
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 */
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);
45 int type = REQ_WRITE | REQ_DISCARD;
46 unsigned int granularity;
47 int 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) % granularity;
62
63 if (flags & BLKDEV_DISCARD_SECURE) {
64 if (!blk_queue_secdiscard(q))
65 return -EOPNOTSUPP;
66 type |= REQ_SECURE;
67 }
68
69 atomic_set(&bb.done, 1);
70 bb.error = 0;
71 bb.wait = &wait;
72
73 blk_start_plug(&plug);
74 while (nr_sects) {
75 unsigned int req_sects;
76 sector_t end_sect, tmp;
77
78 bio = bio_alloc(gfp_mask, 1);
79 if (!bio) {
80 ret = -ENOMEM;
81 break;
82 }
83
84 /* Make sure bi_size doesn't overflow */
85 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
86
87 /*
88 * If splitting a request, and the next starting sector would be
89 * misaligned, stop the discard at the previous aligned sector.
90 */
91 end_sect = sector + req_sects;
92 tmp = end_sect;
93 if (req_sects < nr_sects &&
94 sector_div(tmp, granularity) != alignment) {
95 end_sect = end_sect - alignment;
96 sector_div(end_sect, granularity);
97 end_sect = end_sect * granularity + alignment;
98 req_sects = end_sect - sector;
99 }
100
101 bio->bi_iter.bi_sector = sector;
102 bio->bi_end_io = bio_batch_end_io;
103 bio->bi_bdev = bdev;
104 bio->bi_private = &bb;
105
106 bio->bi_iter.bi_size = req_sects << 9;
107 nr_sects -= req_sects;
108 sector = end_sect;
109
110 atomic_inc(&bb.done);
111 submit_bio(type, bio);
112
113 /*
114 * We can loop for a long time in here, if someone does
115 * full device discards (like mkfs). Be nice and allow
116 * us to schedule out to avoid softlocking if preempt
117 * is disabled.
118 */
119 cond_resched();
120 }
121 blk_finish_plug(&plug);
122
123 /* Wait for bios in-flight */
124 if (!atomic_dec_and_test(&bb.done))
125 wait_for_completion_io(&wait);
126
127 if (bb.error)
128 return bb.error;
129 return ret;
130}
131EXPORT_SYMBOL(blkdev_issue_discard);
132
133/**
134 * blkdev_issue_write_same - queue a write same operation
135 * @bdev: target blockdev
136 * @sector: start sector
137 * @nr_sects: number of sectors to write
138 * @gfp_mask: memory allocation flags (for bio_alloc)
139 * @page: page containing data to write
140 *
141 * Description:
142 * Issue a write same request for the sectors in question.
143 */
144int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
145 sector_t nr_sects, gfp_t gfp_mask,
146 struct page *page)
147{
148 DECLARE_COMPLETION_ONSTACK(wait);
149 struct request_queue *q = bdev_get_queue(bdev);
150 unsigned int max_write_same_sectors;
151 struct bio_batch bb;
152 struct bio *bio;
153 int ret = 0;
154
155 if (!q)
156 return -ENXIO;
157
158 /* Ensure that max_write_same_sectors doesn't overflow bi_size */
159 max_write_same_sectors = UINT_MAX >> 9;
160
161 atomic_set(&bb.done, 1);
162 bb.error = 0;
163 bb.wait = &wait;
164
165 while (nr_sects) {
166 bio = bio_alloc(gfp_mask, 1);
167 if (!bio) {
168 ret = -ENOMEM;
169 break;
170 }
171
172 bio->bi_iter.bi_sector = sector;
173 bio->bi_end_io = bio_batch_end_io;
174 bio->bi_bdev = bdev;
175 bio->bi_private = &bb;
176 bio->bi_vcnt = 1;
177 bio->bi_io_vec->bv_page = page;
178 bio->bi_io_vec->bv_offset = 0;
179 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
180
181 if (nr_sects > max_write_same_sectors) {
182 bio->bi_iter.bi_size = max_write_same_sectors << 9;
183 nr_sects -= max_write_same_sectors;
184 sector += max_write_same_sectors;
185 } else {
186 bio->bi_iter.bi_size = nr_sects << 9;
187 nr_sects = 0;
188 }
189
190 atomic_inc(&bb.done);
191 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
192 }
193
194 /* Wait for bios in-flight */
195 if (!atomic_dec_and_test(&bb.done))
196 wait_for_completion_io(&wait);
197
198 if (bb.error)
199 return bb.error;
200 return ret;
201}
202EXPORT_SYMBOL(blkdev_issue_write_same);
203
204/**
205 * blkdev_issue_zeroout - generate number of zero filed write bios
206 * @bdev: blockdev to issue
207 * @sector: start sector
208 * @nr_sects: number of sectors to write
209 * @gfp_mask: memory allocation flags (for bio_alloc)
210 *
211 * Description:
212 * Generate and issue number of bios with zerofiled pages.
213 */
214
215static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
216 sector_t nr_sects, gfp_t gfp_mask)
217{
218 int ret;
219 struct bio *bio;
220 struct bio_batch bb;
221 unsigned int sz;
222 DECLARE_COMPLETION_ONSTACK(wait);
223
224 atomic_set(&bb.done, 1);
225 bb.error = 0;
226 bb.wait = &wait;
227
228 ret = 0;
229 while (nr_sects != 0) {
230 bio = bio_alloc(gfp_mask,
231 min(nr_sects, (sector_t)BIO_MAX_PAGES));
232 if (!bio) {
233 ret = -ENOMEM;
234 break;
235 }
236
237 bio->bi_iter.bi_sector = sector;
238 bio->bi_bdev = bdev;
239 bio->bi_end_io = bio_batch_end_io;
240 bio->bi_private = &bb;
241
242 while (nr_sects != 0) {
243 sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
244 ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
245 nr_sects -= ret >> 9;
246 sector += ret >> 9;
247 if (ret < (sz << 9))
248 break;
249 }
250 ret = 0;
251 atomic_inc(&bb.done);
252 submit_bio(WRITE, bio);
253 }
254
255 /* Wait for bios in-flight */
256 if (!atomic_dec_and_test(&bb.done))
257 wait_for_completion_io(&wait);
258
259 if (bb.error)
260 return bb.error;
261 return ret;
262}
263
264/**
265 * blkdev_issue_zeroout - zero-fill a block range
266 * @bdev: blockdev to write
267 * @sector: start sector
268 * @nr_sects: number of sectors to write
269 * @gfp_mask: memory allocation flags (for bio_alloc)
270 * @discard: whether to discard the block range
271 *
272 * Description:
273 * Zero-fill a block range. If the discard flag is set and the block
274 * device guarantees that subsequent READ operations to the block range
275 * in question will return zeroes, the blocks will be discarded. Should
276 * the discard request fail, if the discard flag is not set, or if
277 * discard_zeroes_data is not supported, this function will resort to
278 * zeroing the blocks manually, thus provisioning (allocating,
279 * anchoring) them. If the block device supports the WRITE SAME command
280 * blkdev_issue_zeroout() will use it to optimize the process of
281 * clearing the block range. Otherwise the zeroing will be performed
282 * using regular WRITE calls.
283 */
284
285int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
286 sector_t nr_sects, gfp_t gfp_mask, bool discard)
287{
288 struct request_queue *q = bdev_get_queue(bdev);
289
290 if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
291 blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
292 return 0;
293
294 if (bdev_write_same(bdev) &&
295 blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
296 ZERO_PAGE(0)) == 0)
297 return 0;
298
299 return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
300}
301EXPORT_SYMBOL(blkdev_issue_zeroout);