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, int rw, 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(rw, 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 type, struct bio **biop)
  27 {
  28         struct request_queue *q = bdev_get_queue(bdev);
  29         struct bio *bio = *biop;
  30         unsigned int granularity;
  31         int alignment;
  32
  33         if (!q)
  34                 return -ENXIO;
  35         if (!blk_queue_discard(q))
  36                 return -EOPNOTSUPP;
  37         if ((type & REQ_SECURE) && !blk_queue_secdiscard(q))
  38                 return -EOPNOTSUPP;
  39
  40         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  41         granularity = max(q->limits.discard_granularity >> 9, 1U);
  42         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  43
  44         while (nr_sects) {
  45                 unsigned int req_sects;
  46                 sector_t end_sect, tmp;
  47
  48                 /* Make sure bi_size doesn't overflow */
  49                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  50
  51                 /**
  52                  * If splitting a request, and the next starting sector would be
  53                  * misaligned, stop the discard at the previous aligned sector.
  54                  */
  55                 end_sect = sector + req_sects;
  56                 tmp = end_sect;
  57                 if (req_sects < nr_sects &&
  58                     sector_div(tmp, granularity) != alignment) {
  59                         end_sect = end_sect - alignment;
  60                         sector_div(end_sect, granularity);
  61                         end_sect = end_sect * granularity + alignment;
  62                         req_sects = end_sect - sector;
  63                 }
  64
  65                 bio = next_bio(bio, type, 1, gfp_mask);
  66                 bio->bi_iter.bi_sector = sector;
  67                 bio->bi_bdev = bdev;
  68
  69                 bio->bi_iter.bi_size = req_sects << 9;
  70                 nr_sects -= req_sects;
  71                 sector = end_sect;
  72
  73                 /*
  74                  * We can loop for a long time in here, if someone does
  75                  * full device discards (like mkfs). Be nice and allow
  76                  * us to schedule out to avoid softlocking if preempt
  77                  * is disabled.
  78                  */
  79                 cond_resched();
  80         }
  81
  82         *biop = bio;
  83         return 0;
  84 }
  85 EXPORT_SYMBOL(__blkdev_issue_discard);
  86
  87 /**
  88  * blkdev_issue_discard - queue a discard
  89  * @bdev:       blockdev to issue discard for
  90  * @sector:     start sector
  91  * @nr_sects:   number of sectors to discard
  92  * @gfp_mask:   memory allocation flags (for bio_alloc)
  93  * @flags:      BLKDEV_IFL_* flags to control behaviour
  94  *
  95  * Description:
  96  *    Issue a discard request for the sectors in question.
  97  */
  98 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  99                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 100 {
 101         int type = REQ_WRITE | REQ_DISCARD;
 102         struct bio *bio = NULL;
 103         struct blk_plug plug;
 104         int ret;
 105
 106         if (flags & BLKDEV_DISCARD_SECURE)
 107                 type |= REQ_SECURE;
 108
 109         blk_start_plug(&plug);
 110         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, type,
 111                         &bio);
 112         if (!ret && bio) {
 113                 ret = submit_bio_wait(type, bio);
 114                 if (ret == -EOPNOTSUPP)
 115                         ret = 0;
 116                 bio_put(bio);
 117         }
 118         blk_finish_plug(&plug);
 119
 120         return ret;
 121 }
 122 EXPORT_SYMBOL(blkdev_issue_discard);
 123
 124 /**
 125  * blkdev_issue_write_same - queue a write same operation
 126  * @bdev:       target blockdev
 127  * @sector:     start sector
 128  * @nr_sects:   number of sectors to write
 129  * @gfp_mask:   memory allocation flags (for bio_alloc)
 130  * @page:       page containing data to write
 131  *
 132  * Description:
 133  *    Issue a write same request for the sectors in question.
 134  */
 135 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 136                             sector_t nr_sects, gfp_t gfp_mask,
 137                             struct page *page)
 138 {
 139         struct request_queue *q = bdev_get_queue(bdev);
 140         unsigned int max_write_same_sectors;
 141         struct bio *bio = NULL;
 142         int ret = 0;
 143
 144         if (!q)
 145                 return -ENXIO;
 146
 147         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 148         max_write_same_sectors = UINT_MAX >> 9;
 149
 150         while (nr_sects) {
 151                 bio = next_bio(bio, REQ_WRITE | REQ_WRITE_SAME, 1, gfp_mask);
 152                 bio->bi_iter.bi_sector = sector;
 153                 bio->bi_bdev = bdev;
 154                 bio->bi_vcnt = 1;
 155                 bio->bi_io_vec->bv_page = page;
 156                 bio->bi_io_vec->bv_offset = 0;
 157                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 158
 159                 if (nr_sects > max_write_same_sectors) {
 160                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 161                         nr_sects -= max_write_same_sectors;
 162                         sector += max_write_same_sectors;
 163                 } else {
 164                         bio->bi_iter.bi_size = nr_sects << 9;
 165                         nr_sects = 0;
 166                 }
 167         }
 168
 169         if (bio) {
 170                 ret = submit_bio_wait(REQ_WRITE | REQ_WRITE_SAME, bio);
 171                 bio_put(bio);
 172         }
 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, WRITE,
 197                                 min(nr_sects, (sector_t)BIO_MAX_PAGES),
 198                                 gfp_mask);
 199                 bio->bi_iter.bi_sector = sector;
 200                 bio->bi_bdev   = bdev;
 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                 ret = submit_bio_wait(WRITE, bio);
 214                 bio_put(bio);
 215                 return ret;
 216         }
 217         return 0;
 218 }
 219
 220 /**
 221  * blkdev_issue_zeroout - zero-fill a block range
 222  * @bdev:       blockdev to write
 223  * @sector:     start sector
 224  * @nr_sects:   number of sectors to write
 225  * @gfp_mask:   memory allocation flags (for bio_alloc)
 226  * @discard:    whether to discard the block range
 227  *
 228  * Description:
 229  *  Zero-fill a block range.  If the discard flag is set and the block
 230  *  device guarantees that subsequent READ operations to the block range
 231  *  in question will return zeroes, the blocks will be discarded. Should
 232  *  the discard request fail, if the discard flag is not set, or if
 233  *  discard_zeroes_data is not supported, this function will resort to
 234  *  zeroing the blocks manually, thus provisioning (allocating,
 235  *  anchoring) them. If the block device supports the WRITE SAME command
 236  *  blkdev_issue_zeroout() will use it to optimize the process of
 237  *  clearing the block range. Otherwise the zeroing will be performed
 238  *  using regular WRITE calls.
 239  */
 240
 241 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 242                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 243 {
 244         struct request_queue *q = bdev_get_queue(bdev);
 245
 246         if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
 247             blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
 248                 return 0;
 249
 250         if (bdev_write_same(bdev) &&
 251             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 252                                     ZERO_PAGE(0)) == 0)
 253                 return 0;
 254
 255         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 256 }
 257 EXPORT_SYMBOL(blkdev_issue_zeroout);