xref: /openbmc/linux/block/blk-lib.c (revision 288dab8a35a0bde426a09870943c8d3ee3a50dab)
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 	enum req_op op;
33 	int alignment;
34 
35 	if (!q)
36 		return -ENXIO;
37 
38 	if (flags & BLKDEV_DISCARD_SECURE) {
39 		if (!blk_queue_secure_erase(q))
40 			return -EOPNOTSUPP;
41 		op = REQ_OP_SECURE_ERASE;
42 	} else {
43 		if (!blk_queue_discard(q))
44 			return -EOPNOTSUPP;
45 		op = REQ_OP_DISCARD;
46 	}
47 
48 	/* Zero-sector (unknown) and one-sector granularities are the same.  */
49 	granularity = max(q->limits.discard_granularity >> 9, 1U);
50 	alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
51 
52 	while (nr_sects) {
53 		unsigned int req_sects;
54 		sector_t end_sect, tmp;
55 
56 		/* Make sure bi_size doesn't overflow */
57 		req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
58 
59 		/**
60 		 * If splitting a request, and the next starting sector would be
61 		 * misaligned, stop the discard at the previous aligned sector.
62 		 */
63 		end_sect = sector + req_sects;
64 		tmp = end_sect;
65 		if (req_sects < nr_sects &&
66 		    sector_div(tmp, granularity) != alignment) {
67 			end_sect = end_sect - alignment;
68 			sector_div(end_sect, granularity);
69 			end_sect = end_sect * granularity + alignment;
70 			req_sects = end_sect - sector;
71 		}
72 
73 		bio = next_bio(bio, 1, gfp_mask);
74 		bio->bi_iter.bi_sector = sector;
75 		bio->bi_bdev = bdev;
76 		bio_set_op_attrs(bio, op, 0);
77 
78 		bio->bi_iter.bi_size = req_sects << 9;
79 		nr_sects -= req_sects;
80 		sector = end_sect;
81 
82 		/*
83 		 * We can loop for a long time in here, if someone does
84 		 * full device discards (like mkfs). Be nice and allow
85 		 * us to schedule out to avoid softlocking if preempt
86 		 * is disabled.
87 		 */
88 		cond_resched();
89 	}
90 
91 	*biop = bio;
92 	return 0;
93 }
94 EXPORT_SYMBOL(__blkdev_issue_discard);
95 
96 /**
97  * blkdev_issue_discard - queue a discard
98  * @bdev:	blockdev to issue discard for
99  * @sector:	start sector
100  * @nr_sects:	number of sectors to discard
101  * @gfp_mask:	memory allocation flags (for bio_alloc)
102  * @flags:	BLKDEV_IFL_* flags to control behaviour
103  *
104  * Description:
105  *    Issue a discard request for the sectors in question.
106  */
107 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
108 		sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
109 {
110 	struct bio *bio = NULL;
111 	struct blk_plug plug;
112 	int ret;
113 
114 	blk_start_plug(&plug);
115 	ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
116 			&bio);
117 	if (!ret && bio) {
118 		ret = submit_bio_wait(bio);
119 		if (ret == -EOPNOTSUPP)
120 			ret = 0;
121 	}
122 	blk_finish_plug(&plug);
123 
124 	return ret;
125 }
126 EXPORT_SYMBOL(blkdev_issue_discard);
127 
128 /**
129  * blkdev_issue_write_same - queue a write same operation
130  * @bdev:	target blockdev
131  * @sector:	start sector
132  * @nr_sects:	number of sectors to write
133  * @gfp_mask:	memory allocation flags (for bio_alloc)
134  * @page:	page containing data to write
135  *
136  * Description:
137  *    Issue a write same request for the sectors in question.
138  */
139 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
140 			    sector_t nr_sects, gfp_t gfp_mask,
141 			    struct page *page)
142 {
143 	struct request_queue *q = bdev_get_queue(bdev);
144 	unsigned int max_write_same_sectors;
145 	struct bio *bio = NULL;
146 	int ret = 0;
147 
148 	if (!q)
149 		return -ENXIO;
150 
151 	/* Ensure that max_write_same_sectors doesn't overflow bi_size */
152 	max_write_same_sectors = UINT_MAX >> 9;
153 
154 	while (nr_sects) {
155 		bio = next_bio(bio, 1, gfp_mask);
156 		bio->bi_iter.bi_sector = sector;
157 		bio->bi_bdev = bdev;
158 		bio->bi_vcnt = 1;
159 		bio->bi_io_vec->bv_page = page;
160 		bio->bi_io_vec->bv_offset = 0;
161 		bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
162 		bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
163 
164 		if (nr_sects > max_write_same_sectors) {
165 			bio->bi_iter.bi_size = max_write_same_sectors << 9;
166 			nr_sects -= max_write_same_sectors;
167 			sector += max_write_same_sectors;
168 		} else {
169 			bio->bi_iter.bi_size = nr_sects << 9;
170 			nr_sects = 0;
171 		}
172 	}
173 
174 	if (bio)
175 		ret = submit_bio_wait(bio);
176 	return ret != -EOPNOTSUPP ? ret : 0;
177 }
178 EXPORT_SYMBOL(blkdev_issue_write_same);
179 
180 /**
181  * blkdev_issue_zeroout - generate number of zero filed write bios
182  * @bdev:	blockdev to issue
183  * @sector:	start sector
184  * @nr_sects:	number of sectors to write
185  * @gfp_mask:	memory allocation flags (for bio_alloc)
186  *
187  * Description:
188  *  Generate and issue number of bios with zerofiled pages.
189  */
190 
191 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
192 				  sector_t nr_sects, gfp_t gfp_mask)
193 {
194 	int ret;
195 	struct bio *bio = NULL;
196 	unsigned int sz;
197 
198 	while (nr_sects != 0) {
199 		bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
200 				gfp_mask);
201 		bio->bi_iter.bi_sector = sector;
202 		bio->bi_bdev   = bdev;
203 		bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
204 
205 		while (nr_sects != 0) {
206 			sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
207 			ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
208 			nr_sects -= ret >> 9;
209 			sector += ret >> 9;
210 			if (ret < (sz << 9))
211 				break;
212 		}
213 	}
214 
215 	if (bio)
216 		return submit_bio_wait(bio);
217 	return 0;
218 }
219 
220 /**
221  * blkdev_issue_zeroout - zero-fill a block range
222  * @bdev:	blockdev to write
223  * @sector:	start sector
224  * @nr_sects:	number of sectors to write
225  * @gfp_mask:	memory allocation flags (for bio_alloc)
226  * @discard:	whether to discard the block range
227  *
228  * Description:
229  *  Zero-fill a block range.  If the discard flag is set and the block
230  *  device guarantees that subsequent READ operations to the block range
231  *  in question will return zeroes, the blocks will be discarded. Should
232  *  the discard request fail, if the discard flag is not set, or if
233  *  discard_zeroes_data is not supported, this function will resort to
234  *  zeroing the blocks manually, thus provisioning (allocating,
235  *  anchoring) them. If the block device supports the WRITE SAME command
236  *  blkdev_issue_zeroout() will use it to optimize the process of
237  *  clearing the block range. Otherwise the zeroing will be performed
238  *  using regular WRITE calls.
239  */
240 
241 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
242 			 sector_t nr_sects, gfp_t gfp_mask, bool discard)
243 {
244 	struct request_queue *q = bdev_get_queue(bdev);
245 
246 	if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
247 	    blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
248 		return 0;
249 
250 	if (bdev_write_same(bdev) &&
251 	    blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
252 				    ZERO_PAGE(0)) == 0)
253 		return 0;
254 
255 	return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
256 }
257 EXPORT_SYMBOL(blkdev_issue_zeroout);
258