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
38struct bio *blk_alloc_discard_bio(struct block_device *bdev,
39		sector_t *sector, sector_t *nr_sects, gfp_t gfp_mask)
40{
41	sector_t bio_sects = min(*nr_sects, bio_discard_limit(bdev, *sector));
42	struct bio *bio;
43
44	if (!bio_sects)
45		return NULL;
46
47	bio = bio_alloc(bdev, 0, REQ_OP_DISCARD, gfp_mask);
48	if (!bio)
49		return NULL;
50	bio->bi_iter.bi_sector = *sector;
51	bio->bi_iter.bi_size = bio_sects << SECTOR_SHIFT;
52	*sector += bio_sects;
53	*nr_sects -= bio_sects;
54	/*
55	 * We can loop for a long time in here if someone does full device
56	 * discards (like mkfs).  Be nice and allow us to schedule out to avoid
57	 * softlocking if preempt is disabled.
58	 */
59	cond_resched();
60	return bio;
61}
62
63int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
64		sector_t nr_sects, gfp_t gfp_mask, struct bio **biop)
65{
66	struct bio *bio;
67
68	while ((bio = blk_alloc_discard_bio(bdev, &sector, &nr_sects,
69			gfp_mask)))
70		*biop = bio_chain_and_submit(*biop, bio);
71	return 0;
72}
73EXPORT_SYMBOL(__blkdev_issue_discard);
74
75/**
76 * blkdev_issue_discard - queue a discard
77 * @bdev:	blockdev to issue discard for
78 * @sector:	start sector
79 * @nr_sects:	number of sectors to discard
80 * @gfp_mask:	memory allocation flags (for bio_alloc)
81 *
82 * Description:
83 *    Issue a discard request for the sectors in question.
84 */
85int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
86		sector_t nr_sects, gfp_t gfp_mask)
87{
88	struct bio *bio = NULL;
89	struct blk_plug plug;
90	int ret;
91
92	blk_start_plug(&plug);
93	ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, &bio);
94	if (!ret && bio) {
95		ret = submit_bio_wait(bio);
96		if (ret == -EOPNOTSUPP)
97			ret = 0;
98		bio_put(bio);
99	}
100	blk_finish_plug(&plug);
101
102	return ret;
103}
104EXPORT_SYMBOL(blkdev_issue_discard);
105
106static int __blkdev_issue_write_zeroes(struct block_device *bdev,
107		sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
108		struct bio **biop, unsigned flags)
109{
110	struct bio *bio = *biop;
111	unsigned int max_sectors;
112
113	if (bdev_read_only(bdev))
114		return -EPERM;
115
116	/* Ensure that max_sectors doesn't overflow bi_size */
117	max_sectors = bdev_write_zeroes_sectors(bdev);
118
119	if (max_sectors == 0)
120		return -EOPNOTSUPP;
121
122	while (nr_sects) {
123		unsigned int len = min_t(sector_t, nr_sects, max_sectors);
124
125		bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
126		bio->bi_iter.bi_sector = sector;
127		if (flags & BLKDEV_ZERO_NOUNMAP)
128			bio->bi_opf |= REQ_NOUNMAP;
129
130		bio->bi_iter.bi_size = len << SECTOR_SHIFT;
131		nr_sects -= len;
132		sector += len;
133		cond_resched();
134	}
135
136	*biop = bio;
137	return 0;
138}
139
140/*
141 * Convert a number of 512B sectors to a number of pages.
142 * The result is limited to a number of pages that can fit into a BIO.
143 * Also make sure that the result is always at least 1 (page) for the cases
144 * where nr_sects is lower than the number of sectors in a page.
145 */
146static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
147{
148	sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
149
150	return min(pages, (sector_t)BIO_MAX_VECS);
151}
152
153static int __blkdev_issue_zero_pages(struct block_device *bdev,
154		sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
155		struct bio **biop)
156{
157	struct bio *bio = *biop;
158	int bi_size = 0;
159	unsigned int sz;
160
161	if (bdev_read_only(bdev))
162		return -EPERM;
163
164	while (nr_sects != 0) {
165		bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
166				   REQ_OP_WRITE, gfp_mask);
167		bio->bi_iter.bi_sector = sector;
168
169		while (nr_sects != 0) {
170			sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
171			bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
172			nr_sects -= bi_size >> 9;
173			sector += bi_size >> 9;
174			if (bi_size < sz)
175				break;
176		}
177		cond_resched();
178	}
179
180	*biop = bio;
181	return 0;
182}
183
184/**
185 * __blkdev_issue_zeroout - generate number of zero filed write bios
186 * @bdev:	blockdev to issue
187 * @sector:	start sector
188 * @nr_sects:	number of sectors to write
189 * @gfp_mask:	memory allocation flags (for bio_alloc)
190 * @biop:	pointer to anchor bio
191 * @flags:	controls detailed behavior
192 *
193 * Description:
194 *  Zero-fill a block range, either using hardware offload or by explicitly
195 *  writing zeroes to the device.
196 *
197 *  If a device is using logical block provisioning, the underlying space will
198 *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
199 *
200 *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
201 *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
202 */
203int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
204		sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
205		unsigned flags)
206{
207	int ret;
208	sector_t bs_mask;
209
210	bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
211	if ((sector | nr_sects) & bs_mask)
212		return -EINVAL;
213
214	ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
215			biop, flags);
216	if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
217		return ret;
218
219	return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
220					 biop);
221}
222EXPORT_SYMBOL(__blkdev_issue_zeroout);
223
224/**
225 * blkdev_issue_zeroout - zero-fill a block range
226 * @bdev:	blockdev to write
227 * @sector:	start sector
228 * @nr_sects:	number of sectors to write
229 * @gfp_mask:	memory allocation flags (for bio_alloc)
230 * @flags:	controls detailed behavior
231 *
232 * Description:
233 *  Zero-fill a block range, either using hardware offload or by explicitly
234 *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
235 *  valid values for %flags.
236 */
237int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
238		sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
239{
240	int ret = 0;
241	sector_t bs_mask;
242	struct bio *bio;
243	struct blk_plug plug;
244	bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
245
246	bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
247	if ((sector | nr_sects) & bs_mask)
248		return -EINVAL;
249
250retry:
251	bio = NULL;
252	blk_start_plug(&plug);
253	if (try_write_zeroes) {
254		ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
255						  gfp_mask, &bio, flags);
256	} else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
257		ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
258						gfp_mask, &bio);
259	} else {
260		/* No zeroing offload support */
261		ret = -EOPNOTSUPP;
262	}
263	if (ret == 0 && bio) {
264		ret = submit_bio_wait(bio);
265		bio_put(bio);
266	}
267	blk_finish_plug(&plug);
268	if (ret && try_write_zeroes) {
269		if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
270			try_write_zeroes = false;
271			goto retry;
272		}
273		if (!bdev_write_zeroes_sectors(bdev)) {
274			/*
275			 * Zeroing offload support was indicated, but the
276			 * device reported ILLEGAL REQUEST (for some devices
277			 * there is no non-destructive way to verify whether
278			 * WRITE ZEROES is actually supported).
279			 */
280			ret = -EOPNOTSUPP;
281		}
282	}
283
284	return ret;
285}
286EXPORT_SYMBOL(blkdev_issue_zeroout);
287
288int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector,
289		sector_t nr_sects, gfp_t gfp)
290{
291	sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
292	unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev);
293	struct bio *bio = NULL;
294	struct blk_plug plug;
295	int ret = 0;
296
297	/* make sure that "len << SECTOR_SHIFT" doesn't overflow */
298	if (max_sectors > UINT_MAX >> SECTOR_SHIFT)
299		max_sectors = UINT_MAX >> SECTOR_SHIFT;
300	max_sectors &= ~bs_mask;
301
302	if (max_sectors == 0)
303		return -EOPNOTSUPP;
304	if ((sector | nr_sects) & bs_mask)
305		return -EINVAL;
306	if (bdev_read_only(bdev))
307		return -EPERM;
308
309	blk_start_plug(&plug);
310	while (nr_sects) {
311		unsigned int len = min_t(sector_t, nr_sects, max_sectors);
312
313		bio = blk_next_bio(bio, bdev, 0, REQ_OP_SECURE_ERASE, gfp);
314		bio->bi_iter.bi_sector = sector;
315		bio->bi_iter.bi_size = len << SECTOR_SHIFT;
316
317		sector += len;
318		nr_sects -= len;
319		cond_resched();
320	}
321	if (bio) {
322		ret = submit_bio_wait(bio);
323		bio_put(bio);
324	}
325	blk_finish_plug(&plug);
326
327	return ret;
328}
329EXPORT_SYMBOL(blkdev_issue_secure_erase);
330