use-discard-if-possible-in-blkdev_issue_zeroout

   1 block: use discard if possible in blkdev_issue_zeroout()
   2
   3 If the block device supports discards and guarantees that subsequent
   4 reads will return zeros (sometimes known as DZAT, for Deterministic
   5 read Zeros After Trim), use this to implement blkdev_issue_zeroout()
   6
   7 Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
   8 ---
   9  block/blk-lib.c | 72 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++----------
  10  1 file changed, 62 insertions(+), 10 deletions(-)
  11
  12 diff --git a/block/blk-lib.c b/block/blk-lib.c
  13 index 2da76c9..62cbf28 100644
  14 --- a/block/blk-lib.c
  15 +++ b/block/blk-lib.c
  16 @@ -269,6 +269,32 @@ int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
  17         return ret;
  18  }
  19
  20 +static int issue_zeroout_or_write_same(struct block_device *bdev,
  21 +                                      sector_t sector,
  22 +                                      sector_t nr_sects, gfp_t gfp_mask)
  23 +{
  24 +       if (bdev_write_same(bdev)) {
  25 +               unsigned char bdn[BDEVNAME_SIZE];
  26 +
  27 +               if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
  28 +                                            ZERO_PAGE(0)))
  29 +                       return 0;
  30 +
  31 +               bdevname(bdev, bdn);
  32 +               pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
  33 +       }
  34 +
  35 +       return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
  36 +}
  37 +
  38 +/*
  39 + * Like sector_div except don't modify s.
  40 + */
  41 +static unsigned int sector_mod(sector_t s, unsigned int m)
  42 +{
  43 +       return sector_div(s, m);
  44 +}
  45 +
  46  /**
  47   * blkdev_issue_zeroout - zero-fill a block range
  48   * @bdev:      blockdev to write
  49 @@ -277,23 +303,49 @@ int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
  50   * @gfp_mask:  memory allocation flags (for bio_alloc)
  51   *
  52   * Description:
  53 - *  Generate and issue number of bios with zerofiled pages.
  54 + *  Issues bios which zeros the requested block range.
  55   */
  56 -
  57  int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
  58                          sector_t nr_sects, gfp_t gfp_mask)
  59  {
  60 -       if (bdev_write_same(bdev)) {
  61 -               unsigned char bdn[BDEVNAME_SIZE];
  62 +       struct request_queue *q = bdev_get_queue(bdev);
  63 +       unsigned int alignment, granularity;
  64 +       unsigned int c;
  65 +       int ret;
  66
  67 -               if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
  68 -                                            ZERO_PAGE(0)))
  69 -                       return 0;
  70 +       if (!q)
  71 +               return -ENXIO;
  72
  73 -               bdevname(bdev, bdn);
  74 -               pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
  75 +       if (!blk_queue_discard(q) || !queue_discard_zeroes_data(q) ||
  76 +           q->limits.discard_misaligned)
  77 +               return issue_zeroout_or_write_same(bdev, sector,
  78 +
  79 +                                                  nr_sects, gfp_mask);
  80 +
  81 +       alignment = q->limits.discard_alignment >> 9;
  82 +       granularity = q->limits.discard_granularity >> 9;
  83 +
  84 +       c = sector_mod(granularity + alignment - sector, granularity);
  85 +       if (c > nr_sects)
  86 +               c = nr_sects;
  87 +
  88 +       if (c) {
  89 +               int ret = issue_zeroout_or_write_same(bdev, sector,
  90 +                                                     c, gfp_mask);
  91 +               if (ret)
  92 +                       return ret;
  93 +               nr_sects -= c;
  94         }
  95 +       if (nr_sects == 0)
  96 +               return 0;
  97
  98 -       return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
  99 +       c = sector_mod(nr_sects, granularity);
 100 +
 101 +       ret = blkdev_issue_discard(bdev, sector, nr_sects - c, gfp_mask, 0);
 102 +       if (ret || c == 0)
 103 +               return ret;
 104 +
 105 +       return issue_zeroout_or_write_same(bdev, sector + nr_sects - c, c,
 106 +                                          gfp_mask);
 107  }
 108  EXPORT_SYMBOL(blkdev_issue_zeroout);