block/blk-lib.c

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