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  * blkdev_issue_discard - queue a discard
  31  * @bdev:       blockdev to issue discard for
  32  * @sector:     start sector
  33  * @nr_sects:   number of sectors to discard
  34  * @gfp_mask:   memory allocation flags (for bio_alloc)
  35  * @flags:      BLKDEV_IFL_* flags to control behaviour
  36  *
  37  * Description:
  38  *    Issue a discard request for the sectors in question.
  39  */
  40 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  41                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
  42 {
  43         DECLARE_COMPLETION_ONSTACK(wait);
  44         struct request_queue *q = bdev_get_queue(bdev);
  45         int type = REQ_WRITE | REQ_DISCARD;
  46         unsigned int granularity;
  47         int alignment;
  48         struct bio_batch bb;
  49         struct bio *bio;
  50         int ret = 0;
  51         struct blk_plug plug;
  52
  53         if (!q)
  54                 return -ENXIO;
  55
  56         if (!blk_queue_discard(q))
  57                 return -EOPNOTSUPP;
  58
  59         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  60         granularity = max(q->limits.discard_granularity >> 9, 1U);
  61         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  62
  63         if (flags & BLKDEV_DISCARD_SECURE) {
  64                 if (!blk_queue_secdiscard(q))
  65                         return -EOPNOTSUPP;
  66                 type |= REQ_SECURE;
  67         }
  68
  69         atomic_set(&bb.done, 1);
  70         bb.error = 0;
  71         bb.wait = &wait;
  72
  73         blk_start_plug(&plug);
  74         while (nr_sects) {
  75                 unsigned int req_sects;
  76                 sector_t end_sect, tmp;
  77
  78                 bio = bio_alloc(gfp_mask, 1);
  79                 if (!bio) {
  80                         ret = -ENOMEM;
  81                         break;
  82                 }
  83
  84                 /* Make sure bi_size doesn't overflow */
  85                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  86
  87                 /*
  88                  * If splitting a request, and the next starting sector would be
  89                  * misaligned, stop the discard at the previous aligned sector.
  90                  */
  91                 end_sect = sector + req_sects;
  92                 tmp = end_sect;
  93                 if (req_sects < nr_sects &&
  94                     sector_div(tmp, granularity) != alignment) {
  95                         end_sect = end_sect - alignment;
  96                         sector_div(end_sect, granularity);
  97                         end_sect = end_sect * granularity + alignment;
  98                         req_sects = end_sect - sector;
  99                 }
 100
 101                 bio->bi_iter.bi_sector = sector;
 102                 bio->bi_end_io = bio_batch_end_io;
 103                 bio->bi_bdev = bdev;
 104                 bio->bi_private = &bb;
 105
 106                 bio->bi_iter.bi_size = req_sects << 9;
 107                 nr_sects -= req_sects;
 108                 sector = end_sect;
 109
 110                 atomic_inc(&bb.done);
 111                 submit_bio(type, bio);
 112
 113                 /*
 114                  * We can loop for a long time in here, if someone does
 115                  * full device discards (like mkfs). Be nice and allow
 116                  * us to schedule out to avoid softlocking if preempt
 117                  * is disabled.
 118                  */
 119                 cond_resched();
 120         }
 121         blk_finish_plug(&plug);
 122
 123         /* Wait for bios in-flight */
 124         if (!atomic_dec_and_test(&bb.done))
 125                 wait_for_completion_io(&wait);
 126
 127         if (bb.error)
 128                 return bb.error;
 129         return ret;
 130 }
 131 EXPORT_SYMBOL(blkdev_issue_discard);
 132
 133 /**
 134  * blkdev_issue_write_same - queue a write same operation
 135  * @bdev:       target blockdev
 136  * @sector:     start sector
 137  * @nr_sects:   number of sectors to write
 138  * @gfp_mask:   memory allocation flags (for bio_alloc)
 139  * @page:       page containing data to write
 140  *
 141  * Description:
 142  *    Issue a write same request for the sectors in question.
 143  */
 144 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 145                             sector_t nr_sects, gfp_t gfp_mask,
 146                             struct page *page)
 147 {
 148         DECLARE_COMPLETION_ONSTACK(wait);
 149         struct request_queue *q = bdev_get_queue(bdev);
 150         unsigned int max_write_same_sectors;
 151         struct bio_batch bb;
 152         struct bio *bio;
 153         int ret = 0;
 154
 155         if (!q)
 156                 return -ENXIO;
 157
 158         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 159         max_write_same_sectors = UINT_MAX >> 9;
 160
 161         atomic_set(&bb.done, 1);
 162         bb.error = 0;
 163         bb.wait = &wait;
 164
 165         while (nr_sects) {
 166                 bio = bio_alloc(gfp_mask, 1);
 167                 if (!bio) {
 168                         ret = -ENOMEM;
 169                         break;
 170                 }
 171
 172                 bio->bi_iter.bi_sector = sector;
 173                 bio->bi_end_io = bio_batch_end_io;
 174                 bio->bi_bdev = bdev;
 175                 bio->bi_private = &bb;
 176                 bio->bi_vcnt = 1;
 177                 bio->bi_io_vec->bv_page = page;
 178                 bio->bi_io_vec->bv_offset = 0;
 179                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 180
 181                 if (nr_sects > max_write_same_sectors) {
 182                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 183                         nr_sects -= max_write_same_sectors;
 184                         sector += max_write_same_sectors;
 185                 } else {
 186                         bio->bi_iter.bi_size = nr_sects << 9;
 187                         nr_sects = 0;
 188                 }
 189
 190                 atomic_inc(&bb.done);
 191                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
 192         }
 193
 194         /* Wait for bios in-flight */
 195         if (!atomic_dec_and_test(&bb.done))
 196                 wait_for_completion_io(&wait);
 197
 198         if (bb.error)
 199                 return bb.error;
 200         return ret;
 201 }
 202 EXPORT_SYMBOL(blkdev_issue_write_same);
 203
 204 /**
 205  * blkdev_issue_zeroout - generate number of zero filed write bios
 206  * @bdev:       blockdev to issue
 207  * @sector:     start sector
 208  * @nr_sects:   number of sectors to write
 209  * @gfp_mask:   memory allocation flags (for bio_alloc)
 210  *
 211  * Description:
 212  *  Generate and issue number of bios with zerofiled pages.
 213  */
 214
 215 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 216                                   sector_t nr_sects, gfp_t gfp_mask)
 217 {
 218         int ret;
 219         struct bio *bio;
 220         struct bio_batch bb;
 221         unsigned int sz;
 222         DECLARE_COMPLETION_ONSTACK(wait);
 223
 224         atomic_set(&bb.done, 1);
 225         bb.error = 0;
 226         bb.wait = &wait;
 227
 228         ret = 0;
 229         while (nr_sects != 0) {
 230                 bio = bio_alloc(gfp_mask,
 231                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
 232                 if (!bio) {
 233                         ret = -ENOMEM;
 234                         break;
 235                 }
 236
 237                 bio->bi_iter.bi_sector = sector;
 238                 bio->bi_bdev   = bdev;
 239                 bio->bi_end_io = bio_batch_end_io;
 240                 bio->bi_private = &bb;
 241
 242                 while (nr_sects != 0) {
 243                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 244                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 245                         nr_sects -= ret >> 9;
 246                         sector += ret >> 9;
 247                         if (ret < (sz << 9))
 248                                 break;
 249                 }
 250                 ret = 0;
 251                 atomic_inc(&bb.done);
 252                 submit_bio(WRITE, bio);
 253         }
 254
 255         /* Wait for bios in-flight */
 256         if (!atomic_dec_and_test(&bb.done))
 257                 wait_for_completion_io(&wait);
 258
 259         if (bb.error)
 260                 return bb.error;
 261         return ret;
 262 }
 263
 264 /**
 265  * blkdev_issue_zeroout - zero-fill a block range
 266  * @bdev:       blockdev to write
 267  * @sector:     start sector
 268  * @nr_sects:   number of sectors to write
 269  * @gfp_mask:   memory allocation flags (for bio_alloc)
 270  * @discard:    whether to discard the block range
 271  *
 272  * Description:
 273  *  Zero-fill a block range.  If the discard flag is set and the block
 274  *  device guarantees that subsequent READ operations to the block range
 275  *  in question will return zeroes, the blocks will be discarded. Should
 276  *  the discard request fail, if the discard flag is not set, or if
 277  *  discard_zeroes_data is not supported, this function will resort to
 278  *  zeroing the blocks manually, thus provisioning (allocating,
 279  *  anchoring) them. If the block device supports the WRITE SAME command
 280  *  blkdev_issue_zeroout() will use it to optimize the process of
 281  *  clearing the block range. Otherwise the zeroing will be performed
 282  *  using regular WRITE calls.
 283  */
 284
 285 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 286                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 287 {
 288         struct request_queue *q = bdev_get_queue(bdev);
 289
 290         if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
 291             blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
 292                 return 0;
 293
 294         if (bdev_write_same(bdev) &&
 295             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 296                                     ZERO_PAGE(0)) == 0)
 297                 return 0;
 298
 299         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 300 }
 301 EXPORT_SYMBOL(blkdev_issue_zeroout);