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 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         enum req_op op;
  33         int alignment;
  34
  35         if (!q)
  36                 return -ENXIO;
  37
  38         if (flags & BLKDEV_DISCARD_SECURE) {
  39                 if (flags & BLKDEV_DISCARD_ZERO)
  40                         return -EOPNOTSUPP;
  41                 if (!blk_queue_secure_erase(q))
  42                         return -EOPNOTSUPP;
  43                 op = REQ_OP_SECURE_ERASE;
  44         } else {
  45                 if (!blk_queue_discard(q))
  46                         return -EOPNOTSUPP;
  47                 if ((flags & BLKDEV_DISCARD_ZERO) &&
  48                     !q->limits.discard_zeroes_data)
  49                         return -EOPNOTSUPP;
  50                 op = REQ_OP_DISCARD;
  51         }
  52
  53         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  54         granularity = max(q->limits.discard_granularity >> 9, 1U);
  55         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  56
  57         while (nr_sects) {
  58                 unsigned int req_sects;
  59                 sector_t end_sect, tmp;
  60
  61                 /* Make sure bi_size doesn't overflow */
  62                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  63
  64                 /**
  65                  * If splitting a request, and the next starting sector would be
  66                  * misaligned, stop the discard at the previous aligned sector.
  67                  */
  68                 end_sect = sector + req_sects;
  69                 tmp = end_sect;
  70                 if (req_sects < nr_sects &&
  71                     sector_div(tmp, granularity) != alignment) {
  72                         end_sect = end_sect - alignment;
  73                         sector_div(end_sect, granularity);
  74                         end_sect = end_sect * granularity + alignment;
  75                         req_sects = end_sect - sector;
  76                 }
  77
  78                 bio = next_bio(bio, 1, gfp_mask);
  79                 bio->bi_iter.bi_sector = sector;
  80                 bio->bi_bdev = bdev;
  81                 bio_set_op_attrs(bio, op, 0);
  82
  83                 bio->bi_iter.bi_size = req_sects << 9;
  84                 nr_sects -= req_sects;
  85                 sector = end_sect;
  86
  87                 /*
  88                  * We can loop for a long time in here, if someone does
  89                  * full device discards (like mkfs). Be nice and allow
  90                  * us to schedule out to avoid softlocking if preempt
  91                  * is disabled.
  92                  */
  93                 cond_resched();
  94         }
  95
  96         *biop = bio;
  97         return 0;
  98 }
  99 EXPORT_SYMBOL(__blkdev_issue_discard);
 100
 101 /**
 102  * blkdev_issue_discard - queue a discard
 103  * @bdev:       blockdev to issue discard for
 104  * @sector:     start sector
 105  * @nr_sects:   number of sectors to discard
 106  * @gfp_mask:   memory allocation flags (for bio_alloc)
 107  * @flags:      BLKDEV_IFL_* flags to control behaviour
 108  *
 109  * Description:
 110  *    Issue a discard request for the sectors in question.
 111  */
 112 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 113                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 114 {
 115         struct bio *bio = NULL;
 116         struct blk_plug plug;
 117         int ret;
 118
 119         blk_start_plug(&plug);
 120         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 121                         &bio);
 122         if (!ret && bio) {
 123                 ret = submit_bio_wait(bio);
 124                 if (ret == -EOPNOTSUPP && !(flags & BLKDEV_DISCARD_ZERO))
 125                         ret = 0;
 126         }
 127         blk_finish_plug(&plug);
 128
 129         return ret;
 130 }
 131 EXPORT_SYMBOL(blkdev_issue_discard);
 132
 133 /**
 134  * blkdev_issue_write_same - queue a write same operation
 135  * @bdev:       target blockdev
 136  * @sector:     start sector
 137  * @nr_sects:   number of sectors to write
 138  * @gfp_mask:   memory allocation flags (for bio_alloc)
 139  * @page:       page containing data to write
 140  *
 141  * Description:
 142  *    Issue a write same request for the sectors in question.
 143  */
 144 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 145                             sector_t nr_sects, gfp_t gfp_mask,
 146                             struct page *page)
 147 {
 148         struct request_queue *q = bdev_get_queue(bdev);
 149         unsigned int max_write_same_sectors;
 150         struct bio *bio = NULL;
 151         int ret = 0;
 152
 153         if (!q)
 154                 return -ENXIO;
 155
 156         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 157         max_write_same_sectors = UINT_MAX >> 9;
 158
 159         while (nr_sects) {
 160                 bio = next_bio(bio, 1, gfp_mask);
 161                 bio->bi_iter.bi_sector = sector;
 162                 bio->bi_bdev = bdev;
 163                 bio->bi_vcnt = 1;
 164                 bio->bi_io_vec->bv_page = page;
 165                 bio->bi_io_vec->bv_offset = 0;
 166                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 167                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 168
 169                 if (nr_sects > max_write_same_sectors) {
 170                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 171                         nr_sects -= max_write_same_sectors;
 172                         sector += max_write_same_sectors;
 173                 } else {
 174                         bio->bi_iter.bi_size = nr_sects << 9;
 175                         nr_sects = 0;
 176                 }
 177         }
 178
 179         if (bio)
 180                 ret = submit_bio_wait(bio);
 181         return ret;
 182 }
 183 EXPORT_SYMBOL(blkdev_issue_write_same);
 184
 185 /**
 186  * blkdev_issue_zeroout - generate number of zero filed write bios
 187  * @bdev:       blockdev to issue
 188  * @sector:     start sector
 189  * @nr_sects:   number of sectors to write
 190  * @gfp_mask:   memory allocation flags (for bio_alloc)
 191  *
 192  * Description:
 193  *  Generate and issue number of bios with zerofiled pages.
 194  */
 195
 196 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 197                                   sector_t nr_sects, gfp_t gfp_mask)
 198 {
 199         int ret;
 200         struct bio *bio = NULL;
 201         unsigned int sz;
 202
 203         while (nr_sects != 0) {
 204                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
 205                                 gfp_mask);
 206                 bio->bi_iter.bi_sector = sector;
 207                 bio->bi_bdev   = bdev;
 208                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 209
 210                 while (nr_sects != 0) {
 211                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 212                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 213                         nr_sects -= ret >> 9;
 214                         sector += ret >> 9;
 215                         if (ret < (sz << 9))
 216                                 break;
 217                 }
 218         }
 219
 220         if (bio)
 221                 return submit_bio_wait(bio);
 222         return 0;
 223 }
 224
 225 /**
 226  * blkdev_issue_zeroout - zero-fill a block range
 227  * @bdev:       blockdev to write
 228  * @sector:     start sector
 229  * @nr_sects:   number of sectors to write
 230  * @gfp_mask:   memory allocation flags (for bio_alloc)
 231  * @discard:    whether to discard the block range
 232  *
 233  * Description:
 234  *  Zero-fill a block range.  If the discard flag is set and the block
 235  *  device guarantees that subsequent READ operations to the block range
 236  *  in question will return zeroes, the blocks will be discarded. Should
 237  *  the discard request fail, if the discard flag is not set, or if
 238  *  discard_zeroes_data is not supported, this function will resort to
 239  *  zeroing the blocks manually, thus provisioning (allocating,
 240  *  anchoring) them. If the block device supports the WRITE SAME command
 241  *  blkdev_issue_zeroout() will use it to optimize the process of
 242  *  clearing the block range. Otherwise the zeroing will be performed
 243  *  using regular WRITE calls.
 244  */
 245
 246 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 247                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 248 {
 249         if (discard) {
 250                 if (!blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask,
 251                                 BLKDEV_DISCARD_ZERO))
 252                         return 0;
 253         }
 254
 255         if (bdev_write_same(bdev) &&
 256             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 257                                     ZERO_PAGE(0)) == 0)
 258                 return 0;
 259
 260         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 261 }
 262 EXPORT_SYMBOL(blkdev_issue_zeroout);