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 struct bio_batch {
  13         atomic_t                done;
  14         int                     error;
  15         struct completion       *wait;
  16 };
  17
  18 static void bio_batch_end_io(struct bio *bio)
  19 {
  20         struct bio_batch *bb = bio->bi_private;
  21
  22         if (bio->bi_error && bio->bi_error != -EOPNOTSUPP)
  23                 bb->error = bio->bi_error;
  24         if (atomic_dec_and_test(&bb->done))
  25                 complete(bb->wait);
  26         bio_put(bio);
  27 }
  28
  29 /*
  30  * Ensure that max discard sectors doesn't overflow bi_size and hopefully
  31  * it is of the proper granularity as long as the granularity is a power
  32  * of two.
  33  */
  34 #define MAX_BIO_SECTORS ((1U << 31) >> 9)
  35
  36 /**
  37  * blkdev_issue_discard - queue a discard
  38  * @bdev:       blockdev to issue discard for
  39  * @sector:     start sector
  40  * @nr_sects:   number of sectors to discard
  41  * @gfp_mask:   memory allocation flags (for bio_alloc)
  42  * @flags:      BLKDEV_IFL_* flags to control behaviour
  43  *
  44  * Description:
  45  *    Issue a discard request for the sectors in question.
  46  */
  47 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  48                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
  49 {
  50         DECLARE_COMPLETION_ONSTACK(wait);
  51         struct request_queue *q = bdev_get_queue(bdev);
  52         int type = REQ_WRITE | REQ_DISCARD;
  53         struct bio_batch bb;
  54         struct bio *bio;
  55         int ret = 0;
  56         struct blk_plug plug;
  57
  58         if (!q)
  59                 return -ENXIO;
  60
  61         if (!blk_queue_discard(q))
  62                 return -EOPNOTSUPP;
  63
  64         if (flags & BLKDEV_DISCARD_SECURE) {
  65                 if (!blk_queue_secdiscard(q))
  66                         return -EOPNOTSUPP;
  67                 type |= REQ_SECURE;
  68         }
  69
  70         atomic_set(&bb.done, 1);
  71         bb.error = 0;
  72         bb.wait = &wait;
  73
  74         blk_start_plug(&plug);
  75         while (nr_sects) {
  76                 unsigned int req_sects;
  77                 sector_t end_sect;
  78
  79                 bio = bio_alloc(gfp_mask, 1);
  80                 if (!bio) {
  81                         ret = -ENOMEM;
  82                         break;
  83                 }
  84
  85                 req_sects = min_t(sector_t, nr_sects, MAX_BIO_SECTORS);
  86                 end_sect = sector + req_sects;
  87
  88                 bio->bi_iter.bi_sector = sector;
  89                 bio->bi_end_io = bio_batch_end_io;
  90                 bio->bi_bdev = bdev;
  91                 bio->bi_private = &bb;
  92
  93                 bio->bi_iter.bi_size = req_sects << 9;
  94                 nr_sects -= req_sects;
  95                 sector = end_sect;
  96
  97                 atomic_inc(&bb.done);
  98                 submit_bio(type, bio);
  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         blk_finish_plug(&plug);
 109
 110         /* Wait for bios in-flight */
 111         if (!atomic_dec_and_test(&bb.done))
 112                 wait_for_completion_io(&wait);
 113
 114         if (bb.error)
 115                 return bb.error;
 116         return ret;
 117 }
 118 EXPORT_SYMBOL(blkdev_issue_discard);
 119
 120 /**
 121  * blkdev_issue_write_same - queue a write same operation
 122  * @bdev:       target blockdev
 123  * @sector:     start sector
 124  * @nr_sects:   number of sectors to write
 125  * @gfp_mask:   memory allocation flags (for bio_alloc)
 126  * @page:       page containing data to write
 127  *
 128  * Description:
 129  *    Issue a write same request for the sectors in question.
 130  */
 131 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 132                             sector_t nr_sects, gfp_t gfp_mask,
 133                             struct page *page)
 134 {
 135         DECLARE_COMPLETION_ONSTACK(wait);
 136         struct request_queue *q = bdev_get_queue(bdev);
 137         unsigned int max_write_same_sectors;
 138         struct bio_batch bb;
 139         struct bio *bio;
 140         int ret = 0;
 141
 142         if (!q)
 143                 return -ENXIO;
 144
 145         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 146         max_write_same_sectors = UINT_MAX >> 9;
 147
 148         atomic_set(&bb.done, 1);
 149         bb.error = 0;
 150         bb.wait = &wait;
 151
 152         while (nr_sects) {
 153                 bio = bio_alloc(gfp_mask, 1);
 154                 if (!bio) {
 155                         ret = -ENOMEM;
 156                         break;
 157                 }
 158
 159                 bio->bi_iter.bi_sector = sector;
 160                 bio->bi_end_io = bio_batch_end_io;
 161                 bio->bi_bdev = bdev;
 162                 bio->bi_private = &bb;
 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
 168                 if (nr_sects > max_write_same_sectors) {
 169                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 170                         nr_sects -= max_write_same_sectors;
 171                         sector += max_write_same_sectors;
 172                 } else {
 173                         bio->bi_iter.bi_size = nr_sects << 9;
 174                         nr_sects = 0;
 175                 }
 176
 177                 atomic_inc(&bb.done);
 178                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
 179         }
 180
 181         /* Wait for bios in-flight */
 182         if (!atomic_dec_and_test(&bb.done))
 183                 wait_for_completion_io(&wait);
 184
 185         if (bb.error)
 186                 return bb.error;
 187         return ret;
 188 }
 189 EXPORT_SYMBOL(blkdev_issue_write_same);
 190
 191 /**
 192  * blkdev_issue_zeroout - generate number of zero filed write bios
 193  * @bdev:       blockdev to issue
 194  * @sector:     start sector
 195  * @nr_sects:   number of sectors to write
 196  * @gfp_mask:   memory allocation flags (for bio_alloc)
 197  *
 198  * Description:
 199  *  Generate and issue number of bios with zerofiled pages.
 200  */
 201
 202 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 203                                   sector_t nr_sects, gfp_t gfp_mask)
 204 {
 205         int ret;
 206         struct bio *bio;
 207         struct bio_batch bb;
 208         unsigned int sz;
 209         DECLARE_COMPLETION_ONSTACK(wait);
 210
 211         atomic_set(&bb.done, 1);
 212         bb.error = 0;
 213         bb.wait = &wait;
 214
 215         ret = 0;
 216         while (nr_sects != 0) {
 217                 bio = bio_alloc(gfp_mask,
 218                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
 219                 if (!bio) {
 220                         ret = -ENOMEM;
 221                         break;
 222                 }
 223
 224                 bio->bi_iter.bi_sector = sector;
 225                 bio->bi_bdev   = bdev;
 226                 bio->bi_end_io = bio_batch_end_io;
 227                 bio->bi_private = &bb;
 228
 229                 while (nr_sects != 0) {
 230                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 231                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 232                         nr_sects -= ret >> 9;
 233                         sector += ret >> 9;
 234                         if (ret < (sz << 9))
 235                                 break;
 236                 }
 237                 ret = 0;
 238                 atomic_inc(&bb.done);
 239                 submit_bio(WRITE, bio);
 240         }
 241
 242         /* Wait for bios in-flight */
 243         if (!atomic_dec_and_test(&bb.done))
 244                 wait_for_completion_io(&wait);
 245
 246         if (bb.error)
 247                 return bb.error;
 248         return ret;
 249 }
 250
 251 /**
 252  * blkdev_issue_zeroout - zero-fill a block range
 253  * @bdev:       blockdev to write
 254  * @sector:     start sector
 255  * @nr_sects:   number of sectors to write
 256  * @gfp_mask:   memory allocation flags (for bio_alloc)
 257  * @discard:    whether to discard the block range
 258  *
 259  * Description:
 260  *  Zero-fill a block range.  If the discard flag is set and the block
 261  *  device guarantees that subsequent READ operations to the block range
 262  *  in question will return zeroes, the blocks will be discarded. Should
 263  *  the discard request fail, if the discard flag is not set, or if
 264  *  discard_zeroes_data is not supported, this function will resort to
 265  *  zeroing the blocks manually, thus provisioning (allocating,
 266  *  anchoring) them. If the block device supports the WRITE SAME command
 267  *  blkdev_issue_zeroout() will use it to optimize the process of
 268  *  clearing the block range. Otherwise the zeroing will be performed
 269  *  using regular WRITE calls.
 270  */
 271
 272 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 273                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 274 {
 275         struct request_queue *q = bdev_get_queue(bdev);
 276
 277         if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
 278             blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
 279                 return 0;
 280
 281         if (bdev_write_same(bdev) &&
 282             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 283                                     ZERO_PAGE(0)) == 0)
 284                 return 0;
 285
 286         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 287 }
 288 EXPORT_SYMBOL(blkdev_issue_zeroout);