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, 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                 char dev_name[BDEVNAME_SIZE];
  52
  53                 bdevname(bdev, dev_name);
  54                 pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  55                 return -EOPNOTSUPP;
  56         }
  57
  58         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  59         if ((sector | nr_sects) & bs_mask)
  60                 return -EINVAL;
  61
  62         if (!nr_sects)
  63                 return -EINVAL;
  64
  65         while (nr_sects) {
  66                 sector_t req_sects =
  67                         min(nr_sects, bio_discard_limit(bdev, sector));
  68
  69                 bio = blk_next_bio(bio, bdev, 0, REQ_OP_DISCARD, gfp_mask);
  70                 bio->bi_iter.bi_sector = sector;
  71                 bio->bi_iter.bi_size = req_sects << 9;
  72                 sector += req_sects;
  73                 nr_sects -= req_sects;
  74
  75                 /*
  76                  * We can loop for a long time in here, if someone does
  77                  * full device discards (like mkfs). Be nice and allow
  78                  * us to schedule out to avoid softlocking if preempt
  79                  * is disabled.
  80                  */
  81                 cond_resched();
  82         }
  83
  84         *biop = bio;
  85         return 0;
  86 }
  87 EXPORT_SYMBOL(__blkdev_issue_discard);
  88
  89 /**
  90  * blkdev_issue_discard - queue a discard
  91  * @bdev:       blockdev to issue discard for
  92  * @sector:     start sector
  93  * @nr_sects:   number of sectors to discard
  94  * @gfp_mask:   memory allocation flags (for bio_alloc)
  95  *
  96  * Description:
  97  *    Issue a discard request for the sectors in question.
  98  */
  99 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 100                 sector_t nr_sects, gfp_t gfp_mask)
 101 {
 102         struct bio *bio = NULL;
 103         struct blk_plug plug;
 104         int ret;
 105
 106         blk_start_plug(&plug);
 107         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, &bio);
 108         if (!ret && bio) {
 109                 ret = submit_bio_wait(bio);
 110                 if (ret == -EOPNOTSUPP)
 111                         ret = 0;
 112                 bio_put(bio);
 113         }
 114         blk_finish_plug(&plug);
 115
 116         return ret;
 117 }
 118 EXPORT_SYMBOL(blkdev_issue_discard);
 119
 120 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 121                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 122                 struct bio **biop, unsigned flags)
 123 {
 124         struct bio *bio = *biop;
 125         unsigned int max_write_zeroes_sectors;
 126
 127         if (bdev_read_only(bdev))
 128                 return -EPERM;
 129
 130         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 131         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 132
 133         if (max_write_zeroes_sectors == 0)
 134                 return -EOPNOTSUPP;
 135
 136         while (nr_sects) {
 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                 if (nr_sects > max_write_zeroes_sectors) {
 143                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 144                         nr_sects -= max_write_zeroes_sectors;
 145                         sector += max_write_zeroes_sectors;
 146                 } else {
 147                         bio->bi_iter.bi_size = nr_sects << 9;
 148                         nr_sects = 0;
 149                 }
 150                 cond_resched();
 151         }
 152
 153         *biop = bio;
 154         return 0;
 155 }
 156
 157 /*
 158  * Convert a number of 512B sectors to a number of pages.
 159  * The result is limited to a number of pages that can fit into a BIO.
 160  * Also make sure that the result is always at least 1 (page) for the cases
 161  * where nr_sects is lower than the number of sectors in a page.
 162  */
 163 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 164 {
 165         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 166
 167         return min(pages, (sector_t)BIO_MAX_VECS);
 168 }
 169
 170 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 171                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 172                 struct bio **biop)
 173 {
 174         struct bio *bio = *biop;
 175         int bi_size = 0;
 176         unsigned int sz;
 177
 178         if (bdev_read_only(bdev))
 179                 return -EPERM;
 180
 181         while (nr_sects != 0) {
 182                 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
 183                                    REQ_OP_WRITE, gfp_mask);
 184                 bio->bi_iter.bi_sector = sector;
 185
 186                 while (nr_sects != 0) {
 187                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 188                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 189                         nr_sects -= bi_size >> 9;
 190                         sector += bi_size >> 9;
 191                         if (bi_size < sz)
 192                                 break;
 193                 }
 194                 cond_resched();
 195         }
 196
 197         *biop = bio;
 198         return 0;
 199 }
 200
 201 /**
 202  * __blkdev_issue_zeroout - generate number of zero filed write bios
 203  * @bdev:       blockdev to issue
 204  * @sector:     start sector
 205  * @nr_sects:   number of sectors to write
 206  * @gfp_mask:   memory allocation flags (for bio_alloc)
 207  * @biop:       pointer to anchor bio
 208  * @flags:      controls detailed behavior
 209  *
 210  * Description:
 211  *  Zero-fill a block range, either using hardware offload or by explicitly
 212  *  writing zeroes to the device.
 213  *
 214  *  If a device is using logical block provisioning, the underlying space will
 215  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 216  *
 217  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 218  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 219  */
 220 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 221                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 222                 unsigned flags)
 223 {
 224         int ret;
 225         sector_t bs_mask;
 226
 227         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 228         if ((sector | nr_sects) & bs_mask)
 229                 return -EINVAL;
 230
 231         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 232                         biop, flags);
 233         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 234                 return ret;
 235
 236         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 237                                          biop);
 238 }
 239 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 240
 241 /**
 242  * blkdev_issue_zeroout - zero-fill a block range
 243  * @bdev:       blockdev to write
 244  * @sector:     start sector
 245  * @nr_sects:   number of sectors to write
 246  * @gfp_mask:   memory allocation flags (for bio_alloc)
 247  * @flags:      controls detailed behavior
 248  *
 249  * Description:
 250  *  Zero-fill a block range, either using hardware offload or by explicitly
 251  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 252  *  valid values for %flags.
 253  */
 254 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 255                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 256 {
 257         int ret = 0;
 258         sector_t bs_mask;
 259         struct bio *bio;
 260         struct blk_plug plug;
 261         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 262
 263         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 264         if ((sector | nr_sects) & bs_mask)
 265                 return -EINVAL;
 266
 267 retry:
 268         bio = NULL;
 269         blk_start_plug(&plug);
 270         if (try_write_zeroes) {
 271                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 272                                                   gfp_mask, &bio, flags);
 273         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 274                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 275                                                 gfp_mask, &bio);
 276         } else {
 277                 /* No zeroing offload support */
 278                 ret = -EOPNOTSUPP;
 279         }
 280         if (ret == 0 && bio) {
 281                 ret = submit_bio_wait(bio);
 282                 bio_put(bio);
 283         }
 284         blk_finish_plug(&plug);
 285         if (ret && try_write_zeroes) {
 286                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 287                         try_write_zeroes = false;
 288                         goto retry;
 289                 }
 290                 if (!bdev_write_zeroes_sectors(bdev)) {
 291                         /*
 292                          * Zeroing offload support was indicated, but the
 293                          * device reported ILLEGAL REQUEST (for some devices
 294                          * there is no non-destructive way to verify whether
 295                          * WRITE ZEROES is actually supported).
 296                          */
 297                         ret = -EOPNOTSUPP;
 298                 }
 299         }
 300
 301         return ret;
 302 }
 303 EXPORT_SYMBOL(blkdev_issue_zeroout);
 304
 305 int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector,
 306                 sector_t nr_sects, gfp_t gfp)
 307 {
 308         sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 309         unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev);
 310         struct bio *bio = NULL;
 311         struct blk_plug plug;
 312         int ret = 0;
 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         for (;;) {
 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;
 328
 329                 sector += len << SECTOR_SHIFT;
 330                 nr_sects -= len << SECTOR_SHIFT;
 331                 if (!nr_sects) {
 332                         ret = submit_bio_wait(bio);
 333                         bio_put(bio);
 334                         break;
 335                 }
 336                 cond_resched();
 337         }
 338         blk_finish_plug(&plug);
 339
 340         return ret;
 341 }
 342 EXPORT_SYMBOL(blkdev_issue_secure_erase);