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                 /*
  67                  * Issue in chunks of the user defined max discard setting,
  68                  * ensuring that bi_size doesn't overflow
  69                  */
  70                 req_sects = min_t(sector_t, nr_sects,
  71                                         q->limits.max_discard_sectors);
  72                 if (!req_sects)
  73                         goto fail;
  74                 if (req_sects > UINT_MAX >> 9)
  75                         req_sects = UINT_MAX >> 9;
  76
  77                 /*
  78                  * If splitting a request, and the next starting sector would be
  79                  * misaligned, stop the discard at the previous aligned sector.
  80                  */
  81                 end_sect = sector + req_sects;
  82                 tmp = end_sect;
  83                 if (req_sects < nr_sects &&
  84                     sector_div(tmp, granularity) != alignment) {
  85                         end_sect = end_sect - alignment;
  86                         sector_div(end_sect, granularity);
  87                         end_sect = end_sect * granularity + alignment;
  88                         req_sects = end_sect - sector;
  89                 }
  90
  91                 bio = next_bio(bio, 1, gfp_mask);
  92                 bio->bi_iter.bi_sector = sector;
  93                 bio->bi_bdev = bdev;
  94                 bio_set_op_attrs(bio, op, 0);
  95
  96                 bio->bi_iter.bi_size = req_sects << 9;
  97                 nr_sects -= req_sects;
  98                 sector = end_sect;
  99
 100                 /*
 101                  * We can loop for a long time in here, if someone does
 102                  * full device discards (like mkfs). Be nice and allow
 103                  * us to schedule out to avoid softlocking if preempt
 104                  * is disabled.
 105                  */
 106                 cond_resched();
 107         }
 108
 109         *biop = bio;
 110         return 0;
 111
 112 fail:
 113         if (bio) {
 114                 submit_bio_wait(bio);
 115                 bio_put(bio);
 116         }
 117         *biop = NULL;
 118         return -EOPNOTSUPP;
 119 }
 120 EXPORT_SYMBOL(__blkdev_issue_discard);
 121
 122 /**
 123  * blkdev_issue_discard - queue a discard
 124  * @bdev:       blockdev to issue discard for
 125  * @sector:     start sector
 126  * @nr_sects:   number of sectors to discard
 127  * @gfp_mask:   memory allocation flags (for bio_alloc)
 128  * @flags:      BLKDEV_IFL_* flags to control behaviour
 129  *
 130  * Description:
 131  *    Issue a discard request for the sectors in question.
 132  */
 133 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 134                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 135 {
 136         struct bio *bio = NULL;
 137         struct blk_plug plug;
 138         int ret;
 139
 140         blk_start_plug(&plug);
 141         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 142                         &bio);
 143         if (!ret && bio) {
 144                 ret = submit_bio_wait(bio);
 145                 if (ret == -EOPNOTSUPP && !(flags & BLKDEV_DISCARD_ZERO))
 146                         ret = 0;
 147                 bio_put(bio);
 148         }
 149         blk_finish_plug(&plug);
 150
 151         return ret;
 152 }
 153 EXPORT_SYMBOL(blkdev_issue_discard);
 154
 155 /**
 156  * blkdev_issue_write_same - queue a write same operation
 157  * @bdev:       target blockdev
 158  * @sector:     start sector
 159  * @nr_sects:   number of sectors to write
 160  * @gfp_mask:   memory allocation flags (for bio_alloc)
 161  * @page:       page containing data to write
 162  *
 163  * Description:
 164  *    Issue a write same request for the sectors in question.
 165  */
 166 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 167                             sector_t nr_sects, gfp_t gfp_mask,
 168                             struct page *page)
 169 {
 170         struct request_queue *q = bdev_get_queue(bdev);
 171         unsigned int max_write_same_sectors;
 172         struct bio *bio = NULL;
 173         int ret = 0;
 174         sector_t bs_mask;
 175
 176         if (!q)
 177                 return -ENXIO;
 178
 179         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 180         if ((sector | nr_sects) & bs_mask)
 181                 return -EINVAL;
 182
 183         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 184         max_write_same_sectors = UINT_MAX >> 9;
 185
 186         while (nr_sects) {
 187                 bio = next_bio(bio, 1, gfp_mask);
 188                 bio->bi_iter.bi_sector = sector;
 189                 bio->bi_bdev = bdev;
 190                 bio->bi_vcnt = 1;
 191                 bio->bi_io_vec->bv_page = page;
 192                 bio->bi_io_vec->bv_offset = 0;
 193                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 194                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 195
 196                 if (nr_sects > max_write_same_sectors) {
 197                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 198                         nr_sects -= max_write_same_sectors;
 199                         sector += max_write_same_sectors;
 200                 } else {
 201                         bio->bi_iter.bi_size = nr_sects << 9;
 202                         nr_sects = 0;
 203                 }
 204         }
 205
 206         if (bio) {
 207                 ret = submit_bio_wait(bio);
 208                 bio_put(bio);
 209         }
 210         return ret;
 211 }
 212 EXPORT_SYMBOL(blkdev_issue_write_same);
 213
 214 /**
 215  * blkdev_issue_zeroout - generate number of zero filed write bios
 216  * @bdev:       blockdev to issue
 217  * @sector:     start sector
 218  * @nr_sects:   number of sectors to write
 219  * @gfp_mask:   memory allocation flags (for bio_alloc)
 220  *
 221  * Description:
 222  *  Generate and issue number of bios with zerofiled pages.
 223  */
 224
 225 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 226                                   sector_t nr_sects, gfp_t gfp_mask)
 227 {
 228         int ret;
 229         struct bio *bio = NULL;
 230         unsigned int sz;
 231         sector_t bs_mask;
 232
 233         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 234         if ((sector | nr_sects) & bs_mask)
 235                 return -EINVAL;
 236
 237         while (nr_sects != 0) {
 238                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
 239                                 gfp_mask);
 240                 bio->bi_iter.bi_sector = sector;
 241                 bio->bi_bdev   = bdev;
 242                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 243
 244                 while (nr_sects != 0) {
 245                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 246                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 247                         nr_sects -= ret >> 9;
 248                         sector += ret >> 9;
 249                         if (ret < (sz << 9))
 250                                 break;
 251                 }
 252         }
 253
 254         if (bio) {
 255                 ret = submit_bio_wait(bio);
 256                 bio_put(bio);
 257                 return ret;
 258         }
 259         return 0;
 260 }
 261
 262 /**
 263  * blkdev_issue_zeroout - zero-fill a block range
 264  * @bdev:       blockdev to write
 265  * @sector:     start sector
 266  * @nr_sects:   number of sectors to write
 267  * @gfp_mask:   memory allocation flags (for bio_alloc)
 268  * @discard:    whether to discard the block range
 269  *
 270  * Description:
 271  *  Zero-fill a block range.  If the discard flag is set and the block
 272  *  device guarantees that subsequent READ operations to the block range
 273  *  in question will return zeroes, the blocks will be discarded. Should
 274  *  the discard request fail, if the discard flag is not set, or if
 275  *  discard_zeroes_data is not supported, this function will resort to
 276  *  zeroing the blocks manually, thus provisioning (allocating,
 277  *  anchoring) them. If the block device supports the WRITE SAME command
 278  *  blkdev_issue_zeroout() will use it to optimize the process of
 279  *  clearing the block range. Otherwise the zeroing will be performed
 280  *  using regular WRITE calls.
 281  */
 282
 283 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 284                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 285 {
 286         if (discard) {
 287                 if (!blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask,
 288                                 BLKDEV_DISCARD_ZERO))
 289                         return 0;
 290         }
 291
 292         if (bdev_write_same(bdev) &&
 293             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 294                                     ZERO_PAGE(0)) == 0)
 295                 return 0;
 296
 297         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 298 }
 299 EXPORT_SYMBOL(blkdev_issue_zeroout);