block/blk-lib.c

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