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 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  14                 sector_t nr_sects, gfp_t gfp_mask, int flags,
  15                 struct bio **biop)
  16 {
  17         struct request_queue *q = bdev_get_queue(bdev);
  18         struct bio *bio = *biop;
  19         unsigned int op;
  20         sector_t bs_mask, part_offset = 0;
  21
  22         if (bdev_read_only(bdev))
  23                 return -EPERM;
  24
  25         if (flags & BLKDEV_DISCARD_SECURE) {
  26                 if (!blk_queue_secure_erase(q))
  27                         return -EOPNOTSUPP;
  28                 op = REQ_OP_SECURE_ERASE;
  29         } else {
  30                 if (!blk_queue_discard(q))
  31                         return -EOPNOTSUPP;
  32                 op = REQ_OP_DISCARD;
  33         }
  34
  35         /* In case the discard granularity isn't set by buggy device driver */
  36         if (WARN_ON_ONCE(!q->limits.discard_granularity)) {
  37                 char dev_name[BDEVNAME_SIZE];
  38
  39                 bdevname(bdev, dev_name);
  40                 pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  41                 return -EOPNOTSUPP;
  42         }
  43
  44         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  45         if ((sector | nr_sects) & bs_mask)
  46                 return -EINVAL;
  47
  48         if (!nr_sects)
  49                 return -EINVAL;
  50
  51         /* In case the discard request is in a partition */
  52         if (bdev_is_partition(bdev))
  53                 part_offset = bdev->bd_start_sect;
  54
  55         while (nr_sects) {
  56                 sector_t granularity_aligned_lba, req_sects;
  57                 sector_t sector_mapped = sector + part_offset;
  58
  59                 granularity_aligned_lba = round_up(sector_mapped,
  60                                 q->limits.discard_granularity >> SECTOR_SHIFT);
  61
  62                 /*
  63                  * Check whether the discard bio starts at a discard_granularity
  64                  * aligned LBA,
  65                  * - If no: set (granularity_aligned_lba - sector_mapped) to
  66                  *   bi_size of the first split bio, then the second bio will
  67                  *   start at a discard_granularity aligned LBA on the device.
  68                  * - If yes: use bio_aligned_discard_max_sectors() as the max
  69                  *   possible bi_size of the first split bio. Then when this bio
  70                  *   is split in device drive, the split ones are very probably
  71                  *   to be aligned to discard_granularity of the device's queue.
  72                  */
  73                 if (granularity_aligned_lba == sector_mapped)
  74                         req_sects = min_t(sector_t, nr_sects,
  75                                           bio_aligned_discard_max_sectors(q));
  76                 else
  77                         req_sects = min_t(sector_t, nr_sects,
  78                                           granularity_aligned_lba - sector_mapped);
  79
  80                 WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
  81
  82                 bio = blk_next_bio(bio, bdev, 0, op, gfp_mask);
  83                 bio->bi_iter.bi_sector = sector;
  84                 bio->bi_iter.bi_size = req_sects << 9;
  85                 sector += req_sects;
  86                 nr_sects -= req_sects;
  87
  88                 /*
  89                  * We can loop for a long time in here, if someone does
  90                  * full device discards (like mkfs). Be nice and allow
  91                  * us to schedule out to avoid softlocking if preempt
  92                  * is disabled.
  93                  */
  94                 cond_resched();
  95         }
  96
  97         *biop = bio;
  98         return 0;
  99 }
 100 EXPORT_SYMBOL(__blkdev_issue_discard);
 101
 102 /**
 103  * blkdev_issue_discard - queue a discard
 104  * @bdev:       blockdev to issue discard for
 105  * @sector:     start sector
 106  * @nr_sects:   number of sectors to discard
 107  * @gfp_mask:   memory allocation flags (for bio_alloc)
 108  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 109  *
 110  * Description:
 111  *    Issue a discard request for the sectors in question.
 112  */
 113 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 114                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 115 {
 116         struct bio *bio = NULL;
 117         struct blk_plug plug;
 118         int ret;
 119
 120         blk_start_plug(&plug);
 121         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 122                         &bio);
 123         if (!ret && bio) {
 124                 ret = submit_bio_wait(bio);
 125                 if (ret == -EOPNOTSUPP)
 126                         ret = 0;
 127                 bio_put(bio);
 128         }
 129         blk_finish_plug(&plug);
 130
 131         return ret;
 132 }
 133 EXPORT_SYMBOL(blkdev_issue_discard);
 134
 135 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 136                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 137                 struct bio **biop, unsigned flags)
 138 {
 139         struct bio *bio = *biop;
 140         unsigned int max_write_zeroes_sectors;
 141
 142         if (bdev_read_only(bdev))
 143                 return -EPERM;
 144
 145         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 146         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 147
 148         if (max_write_zeroes_sectors == 0)
 149                 return -EOPNOTSUPP;
 150
 151         while (nr_sects) {
 152                 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
 153                 bio->bi_iter.bi_sector = sector;
 154                 if (flags & BLKDEV_ZERO_NOUNMAP)
 155                         bio->bi_opf |= REQ_NOUNMAP;
 156
 157                 if (nr_sects > max_write_zeroes_sectors) {
 158                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 159                         nr_sects -= max_write_zeroes_sectors;
 160                         sector += max_write_zeroes_sectors;
 161                 } else {
 162                         bio->bi_iter.bi_size = nr_sects << 9;
 163                         nr_sects = 0;
 164                 }
 165                 cond_resched();
 166         }
 167
 168         *biop = bio;
 169         return 0;
 170 }
 171
 172 /*
 173  * Convert a number of 512B sectors to a number of pages.
 174  * The result is limited to a number of pages that can fit into a BIO.
 175  * Also make sure that the result is always at least 1 (page) for the cases
 176  * where nr_sects is lower than the number of sectors in a page.
 177  */
 178 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 179 {
 180         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 181
 182         return min(pages, (sector_t)BIO_MAX_VECS);
 183 }
 184
 185 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 186                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 187                 struct bio **biop)
 188 {
 189         struct bio *bio = *biop;
 190         int bi_size = 0;
 191         unsigned int sz;
 192
 193         if (bdev_read_only(bdev))
 194                 return -EPERM;
 195
 196         while (nr_sects != 0) {
 197                 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
 198                                    REQ_OP_WRITE, gfp_mask);
 199                 bio->bi_iter.bi_sector = sector;
 200
 201                 while (nr_sects != 0) {
 202                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 203                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 204                         nr_sects -= bi_size >> 9;
 205                         sector += bi_size >> 9;
 206                         if (bi_size < sz)
 207                                 break;
 208                 }
 209                 cond_resched();
 210         }
 211
 212         *biop = bio;
 213         return 0;
 214 }
 215
 216 /**
 217  * __blkdev_issue_zeroout - generate number of zero filed write bios
 218  * @bdev:       blockdev to issue
 219  * @sector:     start sector
 220  * @nr_sects:   number of sectors to write
 221  * @gfp_mask:   memory allocation flags (for bio_alloc)
 222  * @biop:       pointer to anchor bio
 223  * @flags:      controls detailed behavior
 224  *
 225  * Description:
 226  *  Zero-fill a block range, either using hardware offload or by explicitly
 227  *  writing zeroes to the device.
 228  *
 229  *  If a device is using logical block provisioning, the underlying space will
 230  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 231  *
 232  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 233  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 234  */
 235 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 236                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 237                 unsigned flags)
 238 {
 239         int ret;
 240         sector_t bs_mask;
 241
 242         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 243         if ((sector | nr_sects) & bs_mask)
 244                 return -EINVAL;
 245
 246         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 247                         biop, flags);
 248         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 249                 return ret;
 250
 251         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 252                                          biop);
 253 }
 254 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 255
 256 /**
 257  * blkdev_issue_zeroout - zero-fill a block range
 258  * @bdev:       blockdev to write
 259  * @sector:     start sector
 260  * @nr_sects:   number of sectors to write
 261  * @gfp_mask:   memory allocation flags (for bio_alloc)
 262  * @flags:      controls detailed behavior
 263  *
 264  * Description:
 265  *  Zero-fill a block range, either using hardware offload or by explicitly
 266  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 267  *  valid values for %flags.
 268  */
 269 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 270                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 271 {
 272         int ret = 0;
 273         sector_t bs_mask;
 274         struct bio *bio;
 275         struct blk_plug plug;
 276         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 277
 278         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 279         if ((sector | nr_sects) & bs_mask)
 280                 return -EINVAL;
 281
 282 retry:
 283         bio = NULL;
 284         blk_start_plug(&plug);
 285         if (try_write_zeroes) {
 286                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 287                                                   gfp_mask, &bio, flags);
 288         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 289                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 290                                                 gfp_mask, &bio);
 291         } else {
 292                 /* No zeroing offload support */
 293                 ret = -EOPNOTSUPP;
 294         }
 295         if (ret == 0 && bio) {
 296                 ret = submit_bio_wait(bio);
 297                 bio_put(bio);
 298         }
 299         blk_finish_plug(&plug);
 300         if (ret && try_write_zeroes) {
 301                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 302                         try_write_zeroes = false;
 303                         goto retry;
 304                 }
 305                 if (!bdev_write_zeroes_sectors(bdev)) {
 306                         /*
 307                          * Zeroing offload support was indicated, but the
 308                          * device reported ILLEGAL REQUEST (for some devices
 309                          * there is no non-destructive way to verify whether
 310                          * WRITE ZEROES is actually supported).
 311                          */
 312                         ret = -EOPNOTSUPP;
 313                 }
 314         }
 315
 316         return ret;
 317 }
 318 EXPORT_SYMBOL(blkdev_issue_zeroout);