Loading...
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 unsigned long flags;
15 struct completion *wait;
16};
17
18static 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 */
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 max_discard_sectors, 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 /*
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 -= 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 blk_start_plug(&plug);
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_iter.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_iter.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 * We can loop for a long time in here, if someone does
125 * full device discards (like mkfs). Be nice and allow
126 * us to schedule out to avoid softlocking if preempt
127 * is disabled.
128 */
129 cond_resched();
130 }
131 blk_finish_plug(&plug);
132
133 /* Wait for bios in-flight */
134 if (!atomic_dec_and_test(&bb.done))
135 wait_for_completion_io(&wait);
136
137 if (!test_bit(BIO_UPTODATE, &bb.flags))
138 ret = -EIO;
139
140 return ret;
141}
142EXPORT_SYMBOL(blkdev_issue_discard);
143
144/**
145 * blkdev_issue_write_same - queue a write same operation
146 * @bdev: target blockdev
147 * @sector: start sector
148 * @nr_sects: number of sectors to write
149 * @gfp_mask: memory allocation flags (for bio_alloc)
150 * @page: page containing data to write
151 *
152 * Description:
153 * Issue a write same request for the sectors in question.
154 */
155int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
156 sector_t nr_sects, gfp_t gfp_mask,
157 struct page *page)
158{
159 DECLARE_COMPLETION_ONSTACK(wait);
160 struct request_queue *q = bdev_get_queue(bdev);
161 unsigned int max_write_same_sectors;
162 struct bio_batch bb;
163 struct bio *bio;
164 int ret = 0;
165
166 if (!q)
167 return -ENXIO;
168
169 max_write_same_sectors = q->limits.max_write_same_sectors;
170
171 if (max_write_same_sectors == 0)
172 return -EOPNOTSUPP;
173
174 atomic_set(&bb.done, 1);
175 bb.flags = 1 << BIO_UPTODATE;
176 bb.wait = &wait;
177
178 while (nr_sects) {
179 bio = bio_alloc(gfp_mask, 1);
180 if (!bio) {
181 ret = -ENOMEM;
182 break;
183 }
184
185 bio->bi_iter.bi_sector = sector;
186 bio->bi_end_io = bio_batch_end_io;
187 bio->bi_bdev = bdev;
188 bio->bi_private = &bb;
189 bio->bi_vcnt = 1;
190 bio->bi_io_vec->bv_page = page;
191 bio->bi_io_vec->bv_offset = 0;
192 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
193
194 if (nr_sects > max_write_same_sectors) {
195 bio->bi_iter.bi_size = max_write_same_sectors << 9;
196 nr_sects -= max_write_same_sectors;
197 sector += max_write_same_sectors;
198 } else {
199 bio->bi_iter.bi_size = nr_sects << 9;
200 nr_sects = 0;
201 }
202
203 atomic_inc(&bb.done);
204 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
205 }
206
207 /* Wait for bios in-flight */
208 if (!atomic_dec_and_test(&bb.done))
209 wait_for_completion_io(&wait);
210
211 if (!test_bit(BIO_UPTODATE, &bb.flags))
212 ret = -ENOTSUPP;
213
214 return ret;
215}
216EXPORT_SYMBOL(blkdev_issue_write_same);
217
218/**
219 * blkdev_issue_zeroout - generate number of zero filed write bios
220 * @bdev: blockdev to issue
221 * @sector: start sector
222 * @nr_sects: number of sectors to write
223 * @gfp_mask: memory allocation flags (for bio_alloc)
224 *
225 * Description:
226 * Generate and issue number of bios with zerofiled pages.
227 */
228
229int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
230 sector_t nr_sects, gfp_t gfp_mask)
231{
232 int ret;
233 struct bio *bio;
234 struct bio_batch bb;
235 unsigned int sz;
236 DECLARE_COMPLETION_ONSTACK(wait);
237
238 atomic_set(&bb.done, 1);
239 bb.flags = 1 << BIO_UPTODATE;
240 bb.wait = &wait;
241
242 ret = 0;
243 while (nr_sects != 0) {
244 bio = bio_alloc(gfp_mask,
245 min(nr_sects, (sector_t)BIO_MAX_PAGES));
246 if (!bio) {
247 ret = -ENOMEM;
248 break;
249 }
250
251 bio->bi_iter.bi_sector = sector;
252 bio->bi_bdev = bdev;
253 bio->bi_end_io = bio_batch_end_io;
254 bio->bi_private = &bb;
255
256 while (nr_sects != 0) {
257 sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
258 ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
259 nr_sects -= ret >> 9;
260 sector += ret >> 9;
261 if (ret < (sz << 9))
262 break;
263 }
264 ret = 0;
265 atomic_inc(&bb.done);
266 submit_bio(WRITE, bio);
267 }
268
269 /* Wait for bios in-flight */
270 if (!atomic_dec_and_test(&bb.done))
271 wait_for_completion_io(&wait);
272
273 if (!test_bit(BIO_UPTODATE, &bb.flags))
274 /* One of bios in the batch was completed with error.*/
275 ret = -EIO;
276
277 return ret;
278}
279
280/**
281 * blkdev_issue_zeroout - zero-fill a block range
282 * @bdev: blockdev to write
283 * @sector: start sector
284 * @nr_sects: number of sectors to write
285 * @gfp_mask: memory allocation flags (for bio_alloc)
286 *
287 * Description:
288 * Generate and issue number of bios with zerofiled pages.
289 */
290
291int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
292 sector_t nr_sects, gfp_t gfp_mask)
293{
294 if (bdev_write_same(bdev)) {
295 unsigned char bdn[BDEVNAME_SIZE];
296
297 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
298 ZERO_PAGE(0)))
299 return 0;
300
301 bdevname(bdev, bdn);
302 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
303 }
304
305 return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
306}
307EXPORT_SYMBOL(blkdev_issue_zeroout);
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_sectors;
124
125 if (bdev_read_only(bdev))
126 return -EPERM;
127
128 /* Ensure that max_sectors doesn't overflow bi_size */
129 max_sectors = bdev_write_zeroes_sectors(bdev);
130
131 if (max_sectors == 0)
132 return -EOPNOTSUPP;
133
134 while (nr_sects) {
135 unsigned int len = min_t(sector_t, nr_sects, max_sectors);
136
137 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
138 bio->bi_iter.bi_sector = sector;
139 if (flags & BLKDEV_ZERO_NOUNMAP)
140 bio->bi_opf |= REQ_NOUNMAP;
141
142 bio->bi_iter.bi_size = len << SECTOR_SHIFT;
143 nr_sects -= len;
144 sector += len;
145 cond_resched();
146 }
147
148 *biop = bio;
149 return 0;
150}
151
152/*
153 * Convert a number of 512B sectors to a number of pages.
154 * The result is limited to a number of pages that can fit into a BIO.
155 * Also make sure that the result is always at least 1 (page) for the cases
156 * where nr_sects is lower than the number of sectors in a page.
157 */
158static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
159{
160 sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
161
162 return min(pages, (sector_t)BIO_MAX_VECS);
163}
164
165static int __blkdev_issue_zero_pages(struct block_device *bdev,
166 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
167 struct bio **biop)
168{
169 struct bio *bio = *biop;
170 int bi_size = 0;
171 unsigned int sz;
172
173 if (bdev_read_only(bdev))
174 return -EPERM;
175
176 while (nr_sects != 0) {
177 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
178 REQ_OP_WRITE, gfp_mask);
179 bio->bi_iter.bi_sector = sector;
180
181 while (nr_sects != 0) {
182 sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
183 bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
184 nr_sects -= bi_size >> 9;
185 sector += bi_size >> 9;
186 if (bi_size < sz)
187 break;
188 }
189 cond_resched();
190 }
191
192 *biop = bio;
193 return 0;
194}
195
196/**
197 * __blkdev_issue_zeroout - generate number of zero filed write bios
198 * @bdev: blockdev to issue
199 * @sector: start sector
200 * @nr_sects: number of sectors to write
201 * @gfp_mask: memory allocation flags (for bio_alloc)
202 * @biop: pointer to anchor bio
203 * @flags: controls detailed behavior
204 *
205 * Description:
206 * Zero-fill a block range, either using hardware offload or by explicitly
207 * writing zeroes to the device.
208 *
209 * If a device is using logical block provisioning, the underlying space will
210 * not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
211 *
212 * If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
213 * -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
214 */
215int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
216 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
217 unsigned flags)
218{
219 int ret;
220 sector_t bs_mask;
221
222 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
223 if ((sector | nr_sects) & bs_mask)
224 return -EINVAL;
225
226 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
227 biop, flags);
228 if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
229 return ret;
230
231 return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
232 biop);
233}
234EXPORT_SYMBOL(__blkdev_issue_zeroout);
235
236/**
237 * blkdev_issue_zeroout - zero-fill a block range
238 * @bdev: blockdev to write
239 * @sector: start sector
240 * @nr_sects: number of sectors to write
241 * @gfp_mask: memory allocation flags (for bio_alloc)
242 * @flags: controls detailed behavior
243 *
244 * Description:
245 * Zero-fill a block range, either using hardware offload or by explicitly
246 * writing zeroes to the device. See __blkdev_issue_zeroout() for the
247 * valid values for %flags.
248 */
249int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
250 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
251{
252 int ret = 0;
253 sector_t bs_mask;
254 struct bio *bio;
255 struct blk_plug plug;
256 bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
257
258 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
259 if ((sector | nr_sects) & bs_mask)
260 return -EINVAL;
261
262retry:
263 bio = NULL;
264 blk_start_plug(&plug);
265 if (try_write_zeroes) {
266 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
267 gfp_mask, &bio, flags);
268 } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
269 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
270 gfp_mask, &bio);
271 } else {
272 /* No zeroing offload support */
273 ret = -EOPNOTSUPP;
274 }
275 if (ret == 0 && bio) {
276 ret = submit_bio_wait(bio);
277 bio_put(bio);
278 }
279 blk_finish_plug(&plug);
280 if (ret && try_write_zeroes) {
281 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
282 try_write_zeroes = false;
283 goto retry;
284 }
285 if (!bdev_write_zeroes_sectors(bdev)) {
286 /*
287 * Zeroing offload support was indicated, but the
288 * device reported ILLEGAL REQUEST (for some devices
289 * there is no non-destructive way to verify whether
290 * WRITE ZEROES is actually supported).
291 */
292 ret = -EOPNOTSUPP;
293 }
294 }
295
296 return ret;
297}
298EXPORT_SYMBOL(blkdev_issue_zeroout);
299
300int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector,
301 sector_t nr_sects, gfp_t gfp)
302{
303 sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
304 unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev);
305 struct bio *bio = NULL;
306 struct blk_plug plug;
307 int ret = 0;
308
309 /* make sure that "len << SECTOR_SHIFT" doesn't overflow */
310 if (max_sectors > UINT_MAX >> SECTOR_SHIFT)
311 max_sectors = UINT_MAX >> SECTOR_SHIFT;
312 max_sectors &= ~bs_mask;
313
314 if (max_sectors == 0)
315 return -EOPNOTSUPP;
316 if ((sector | nr_sects) & bs_mask)
317 return -EINVAL;
318 if (bdev_read_only(bdev))
319 return -EPERM;
320
321 blk_start_plug(&plug);
322 while (nr_sects) {
323 unsigned int len = min_t(sector_t, nr_sects, max_sectors);
324
325 bio = blk_next_bio(bio, bdev, 0, REQ_OP_SECURE_ERASE, gfp);
326 bio->bi_iter.bi_sector = sector;
327 bio->bi_iter.bi_size = len << SECTOR_SHIFT;
328
329 sector += len;
330 nr_sects -= len;
331 cond_resched();
332 }
333 if (bio) {
334 ret = submit_bio_wait(bio);
335 bio_put(bio);
336 }
337 blk_finish_plug(&plug);
338
339 return ret;
340}
341EXPORT_SYMBOL(blkdev_issue_secure_erase);