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                 bio_put(bio);
 127         }
 128         blk_finish_plug(&plug);
 129
 130         return ret;
 131 }
 132 EXPORT_SYMBOL(blkdev_issue_discard);
 133
 134 /**
 135  * blkdev_issue_write_same - queue a write same operation
 136  * @bdev:       target blockdev
 137  * @sector:     start sector
 138  * @nr_sects:   number of sectors to write
 139  * @gfp_mask:   memory allocation flags (for bio_alloc)
 140  * @page:       page containing data to write
 141  *
 142  * Description:
 143  *    Issue a write same request for the sectors in question.
 144  */
 145 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 146                             sector_t nr_sects, gfp_t gfp_mask,
 147                             struct page *page)
 148 {
 149         struct request_queue *q = bdev_get_queue(bdev);
 150         unsigned int max_write_same_sectors;
 151         struct bio *bio = NULL;
 152         int ret = 0;
 153
 154         if (!q)
 155                 return -ENXIO;
 156
 157         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 158         max_write_same_sectors = UINT_MAX >> 9;
 159
 160         while (nr_sects) {
 161                 bio = next_bio(bio, 1, gfp_mask);
 162                 bio->bi_iter.bi_sector = sector;
 163                 bio->bi_bdev = bdev;
 164                 bio->bi_vcnt = 1;
 165                 bio->bi_io_vec->bv_page = page;
 166                 bio->bi_io_vec->bv_offset = 0;
 167                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 168                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 169
 170                 if (nr_sects > max_write_same_sectors) {
 171                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 172                         nr_sects -= max_write_same_sectors;
 173                         sector += max_write_same_sectors;
 174                 } else {
 175                         bio->bi_iter.bi_size = nr_sects << 9;
 176                         nr_sects = 0;
 177                 }
 178         }
 179
 180         if (bio) {
 181                 ret = submit_bio_wait(bio);
 182                 bio_put(bio);
 183         }
 184         return ret;
 185 }
 186 EXPORT_SYMBOL(blkdev_issue_write_same);
 187
 188 /**
 189  * blkdev_issue_zeroout - generate number of zero filed write bios
 190  * @bdev:       blockdev to issue
 191  * @sector:     start sector
 192  * @nr_sects:   number of sectors to write
 193  * @gfp_mask:   memory allocation flags (for bio_alloc)
 194  *
 195  * Description:
 196  *  Generate and issue number of bios with zerofiled pages.
 197  */
 198
 199 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 200                                   sector_t nr_sects, gfp_t gfp_mask)
 201 {
 202         int ret;
 203         struct bio *bio = NULL;
 204         unsigned int sz;
 205
 206         while (nr_sects != 0) {
 207                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
 208                                 gfp_mask);
 209                 bio->bi_iter.bi_sector = sector;
 210                 bio->bi_bdev   = bdev;
 211                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 212
 213                 while (nr_sects != 0) {
 214                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 215                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 216                         nr_sects -= ret >> 9;
 217                         sector += ret >> 9;
 218                         if (ret < (sz << 9))
 219                                 break;
 220                 }
 221         }
 222
 223         if (bio) {
 224                 ret = submit_bio_wait(bio);
 225                 bio_put(bio);
 226                 return ret;
 227         }
 228         return 0;
 229 }
 230
 231 /**
 232  * blkdev_issue_zeroout - zero-fill a block range
 233  * @bdev:       blockdev to write
 234  * @sector:     start sector
 235  * @nr_sects:   number of sectors to write
 236  * @gfp_mask:   memory allocation flags (for bio_alloc)
 237  * @discard:    whether to discard the block range
 238  *
 239  * Description:
 240  *  Zero-fill a block range.  If the discard flag is set and the block
 241  *  device guarantees that subsequent READ operations to the block range
 242  *  in question will return zeroes, the blocks will be discarded. Should
 243  *  the discard request fail, if the discard flag is not set, or if
 244  *  discard_zeroes_data is not supported, this function will resort to
 245  *  zeroing the blocks manually, thus provisioning (allocating,
 246  *  anchoring) them. If the block device supports the WRITE SAME command
 247  *  blkdev_issue_zeroout() will use it to optimize the process of
 248  *  clearing the block range. Otherwise the zeroing will be performed
 249  *  using regular WRITE calls.
 250  */
 251
 252 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 253                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 254 {
 255         if (discard) {
 256                 if (!blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask,
 257                                 BLKDEV_DISCARD_ZERO))
 258                         return 0;
 259         }
 260
 261         if (bdev_write_same(bdev) &&
 262             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 263                                     ZERO_PAGE(0)) == 0)
 264                 return 0;
 265
 266         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 267 }
 268 EXPORT_SYMBOL(blkdev_issue_zeroout);