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         sector_t bs_mask;
  35
  36         if (!q)
  37                 return -ENXIO;
  38
  39         if (flags & BLKDEV_DISCARD_SECURE) {
  40                 if (flags & BLKDEV_DISCARD_ZERO)
  41                         return -EOPNOTSUPP;
  42                 if (!blk_queue_secure_erase(q))
  43                         return -EOPNOTSUPP;
  44                 op = REQ_OP_SECURE_ERASE;
  45         } else {
  46                 if (!blk_queue_discard(q))
  47                         return -EOPNOTSUPP;
  48                 if ((flags & BLKDEV_DISCARD_ZERO) &&
  49                     !q->limits.discard_zeroes_data)
  50                         return -EOPNOTSUPP;
  51                 op = REQ_OP_DISCARD;
  52         }
  53
  54         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  55         if ((sector | nr_sects) & bs_mask)
  56                 return -EINVAL;
  57
  58         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  59         granularity = max(q->limits.discard_granularity >> 9, 1U);
  60         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  61
  62         while (nr_sects) {
  63                 unsigned int req_sects;
  64                 sector_t end_sect, tmp;
  65
  66                 /* Make sure bi_size doesn't overflow */
  67                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  68
  69                 /**
  70                  * If splitting a request, and the next starting sector would be
  71                  * misaligned, stop the discard at the previous aligned sector.
  72                  */
  73                 end_sect = sector + req_sects;
  74                 tmp = end_sect;
  75                 if (req_sects < nr_sects &&
  76                     sector_div(tmp, granularity) != alignment) {
  77                         end_sect = end_sect - alignment;
  78                         sector_div(end_sect, granularity);
  79                         end_sect = end_sect * granularity + alignment;
  80                         req_sects = end_sect - sector;
  81                 }
  82
  83                 bio = next_bio(bio, 1, gfp_mask);
  84                 bio->bi_iter.bi_sector = sector;
  85                 bio->bi_bdev = bdev;
  86                 bio_set_op_attrs(bio, op, 0);
  87
  88                 bio->bi_iter.bi_size = req_sects << 9;
  89                 nr_sects -= req_sects;
  90                 sector = end_sect;
  91
  92                 /*
  93                  * We can loop for a long time in here, if someone does
  94                  * full device discards (like mkfs). Be nice and allow
  95                  * us to schedule out to avoid softlocking if preempt
  96                  * is disabled.
  97                  */
  98                 cond_resched();
  99         }
 100
 101         *biop = bio;
 102         return 0;
 103 }
 104 EXPORT_SYMBOL(__blkdev_issue_discard);
 105
 106 /**
 107  * blkdev_issue_discard - queue a discard
 108  * @bdev:       blockdev to issue discard for
 109  * @sector:     start sector
 110  * @nr_sects:   number of sectors to discard
 111  * @gfp_mask:   memory allocation flags (for bio_alloc)
 112  * @flags:      BLKDEV_IFL_* flags to control behaviour
 113  *
 114  * Description:
 115  *    Issue a discard request for the sectors in question.
 116  */
 117 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 118                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 119 {
 120         struct bio *bio = NULL;
 121         struct blk_plug plug;
 122         int ret;
 123
 124         blk_start_plug(&plug);
 125         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 126                         &bio);
 127         if (!ret && bio) {
 128                 ret = submit_bio_wait(bio);
 129                 if (ret == -EOPNOTSUPP && !(flags & BLKDEV_DISCARD_ZERO))
 130                         ret = 0;
 131                 bio_put(bio);
 132         }
 133         blk_finish_plug(&plug);
 134
 135         return ret;
 136 }
 137 EXPORT_SYMBOL(blkdev_issue_discard);
 138
 139 /**
 140  * blkdev_issue_write_same - queue a write same operation
 141  * @bdev:       target blockdev
 142  * @sector:     start sector
 143  * @nr_sects:   number of sectors to write
 144  * @gfp_mask:   memory allocation flags (for bio_alloc)
 145  * @page:       page containing data to write
 146  *
 147  * Description:
 148  *    Issue a write same request for the sectors in question.
 149  */
 150 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 151                             sector_t nr_sects, gfp_t gfp_mask,
 152                             struct page *page)
 153 {
 154         struct request_queue *q = bdev_get_queue(bdev);
 155         unsigned int max_write_same_sectors;
 156         struct bio *bio = NULL;
 157         int ret = 0;
 158         sector_t bs_mask;
 159
 160         if (!q)
 161                 return -ENXIO;
 162
 163         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 164         if ((sector | nr_sects) & bs_mask)
 165                 return -EINVAL;
 166
 167         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 168         max_write_same_sectors = UINT_MAX >> 9;
 169
 170         while (nr_sects) {
 171                 bio = next_bio(bio, 1, gfp_mask);
 172                 bio->bi_iter.bi_sector = sector;
 173                 bio->bi_bdev = bdev;
 174                 bio->bi_vcnt = 1;
 175                 bio->bi_io_vec->bv_page = page;
 176                 bio->bi_io_vec->bv_offset = 0;
 177                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 178                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 179
 180                 if (nr_sects > max_write_same_sectors) {
 181                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 182                         nr_sects -= max_write_same_sectors;
 183                         sector += max_write_same_sectors;
 184                 } else {
 185                         bio->bi_iter.bi_size = nr_sects << 9;
 186                         nr_sects = 0;
 187                 }
 188         }
 189
 190         if (bio) {
 191                 ret = submit_bio_wait(bio);
 192                 bio_put(bio);
 193         }
 194         return ret;
 195 }
 196 EXPORT_SYMBOL(blkdev_issue_write_same);
 197
 198 /**
 199  * blkdev_issue_zeroout - generate number of zero filed write bios
 200  * @bdev:       blockdev to issue
 201  * @sector:     start sector
 202  * @nr_sects:   number of sectors to write
 203  * @gfp_mask:   memory allocation flags (for bio_alloc)
 204  *
 205  * Description:
 206  *  Generate and issue number of bios with zerofiled pages.
 207  */
 208
 209 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 210                                   sector_t nr_sects, gfp_t gfp_mask)
 211 {
 212         int ret;
 213         struct bio *bio = NULL;
 214         unsigned int sz;
 215         sector_t bs_mask;
 216
 217         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 218         if ((sector | nr_sects) & bs_mask)
 219                 return -EINVAL;
 220
 221         while (nr_sects != 0) {
 222                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
 223                                 gfp_mask);
 224                 bio->bi_iter.bi_sector = sector;
 225                 bio->bi_bdev   = bdev;
 226                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 227
 228                 while (nr_sects != 0) {
 229                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 230                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 231                         nr_sects -= ret >> 9;
 232                         sector += ret >> 9;
 233                         if (ret < (sz << 9))
 234                                 break;
 235                 }
 236         }
 237
 238         if (bio) {
 239                 ret = submit_bio_wait(bio);
 240                 bio_put(bio);
 241                 return ret;
 242         }
 243         return 0;
 244 }
 245
 246 /**
 247  * blkdev_issue_zeroout - zero-fill a block range
 248  * @bdev:       blockdev to write
 249  * @sector:     start sector
 250  * @nr_sects:   number of sectors to write
 251  * @gfp_mask:   memory allocation flags (for bio_alloc)
 252  * @discard:    whether to discard the block range
 253  *
 254  * Description:
 255  *  Zero-fill a block range.  If the discard flag is set and the block
 256  *  device guarantees that subsequent READ operations to the block range
 257  *  in question will return zeroes, the blocks will be discarded. Should
 258  *  the discard request fail, if the discard flag is not set, or if
 259  *  discard_zeroes_data is not supported, this function will resort to
 260  *  zeroing the blocks manually, thus provisioning (allocating,
 261  *  anchoring) them. If the block device supports the WRITE SAME command
 262  *  blkdev_issue_zeroout() will use it to optimize the process of
 263  *  clearing the block range. Otherwise the zeroing will be performed
 264  *  using regular WRITE calls.
 265  */
 266
 267 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 268                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 269 {
 270         if (discard) {
 271                 if (!blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask,
 272                                 BLKDEV_DISCARD_ZERO))
 273                         return 0;
 274         }
 275
 276         if (bdev_write_same(bdev) &&
 277             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 278                                     ZERO_PAGE(0)) == 0)
 279                 return 0;
 280
 281         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 282 }
 283 EXPORT_SYMBOL(blkdev_issue_zeroout);