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 max_discard_sectors, 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         /*
  64          * Ensure that max_discard_sectors is of the proper
  65          * granularity, so that requests stay aligned after a split.
  66          */
  67         max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
  68         max_discard_sectors -= max_discard_sectors % granularity;
  69         if (unlikely(!max_discard_sectors)) {
  70                 /* Avoid infinite loop below. Being cautious never hurts. */
  71                 return -EOPNOTSUPP;
  72         }
  73
  74         if (flags & BLKDEV_DISCARD_SECURE) {
  75                 if (!blk_queue_secdiscard(q))
  76                         return -EOPNOTSUPP;
  77                 type |= REQ_SECURE;
  78         }
  79
  80         atomic_set(&bb.done, 1);
  81         bb.error = 0;
  82         bb.wait = &wait;
  83
  84         blk_start_plug(&plug);
  85         while (nr_sects) {
  86                 unsigned int req_sects;
  87                 sector_t end_sect, tmp;
  88
  89                 bio = bio_alloc(gfp_mask, 1);
  90                 if (!bio) {
  91                         ret = -ENOMEM;
  92                         break;
  93                 }
  94
  95                 req_sects = min_t(sector_t, nr_sects, max_discard_sectors);
  96
  97                 /*
  98                  * If splitting a request, and the next starting sector would be
  99                  * misaligned, stop the discard at the previous aligned sector.
 100                  */
 101                 end_sect = sector + req_sects;
 102                 tmp = end_sect;
 103                 if (req_sects < nr_sects &&
 104                     sector_div(tmp, granularity) != alignment) {
 105                         end_sect = end_sect - alignment;
 106                         sector_div(end_sect, granularity);
 107                         end_sect = end_sect * granularity + alignment;
 108                         req_sects = end_sect - sector;
 109                 }
 110
 111                 bio->bi_iter.bi_sector = sector;
 112                 bio->bi_end_io = bio_batch_end_io;
 113                 bio->bi_bdev = bdev;
 114                 bio->bi_private = &bb;
 115
 116                 bio->bi_iter.bi_size = req_sects << 9;
 117                 nr_sects -= req_sects;
 118                 sector = end_sect;
 119
 120                 atomic_inc(&bb.done);
 121                 submit_bio(type, bio);
 122
 123                 /*
 124                  * We can loop for a long time in here, if someone does
 125                  * full device discards (like mkfs). Be nice and allow
 126                  * us to schedule out to avoid softlocking if preempt
 127                  * is disabled.
 128                  */
 129                 cond_resched();
 130         }
 131         blk_finish_plug(&plug);
 132
 133         /* Wait for bios in-flight */
 134         if (!atomic_dec_and_test(&bb.done))
 135                 wait_for_completion_io(&wait);
 136
 137         if (bb.error)
 138                 return bb.error;
 139         return ret;
 140 }
 141 EXPORT_SYMBOL(blkdev_issue_discard);
 142
 143 /**
 144  * blkdev_issue_write_same - queue a write same operation
 145  * @bdev:       target blockdev
 146  * @sector:     start sector
 147  * @nr_sects:   number of sectors to write
 148  * @gfp_mask:   memory allocation flags (for bio_alloc)
 149  * @page:       page containing data to write
 150  *
 151  * Description:
 152  *    Issue a write same request for the sectors in question.
 153  */
 154 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 155                             sector_t nr_sects, gfp_t gfp_mask,
 156                             struct page *page)
 157 {
 158         DECLARE_COMPLETION_ONSTACK(wait);
 159         struct request_queue *q = bdev_get_queue(bdev);
 160         unsigned int max_write_same_sectors;
 161         struct bio_batch bb;
 162         struct bio *bio;
 163         int ret = 0;
 164
 165         if (!q)
 166                 return -ENXIO;
 167
 168         max_write_same_sectors = q->limits.max_write_same_sectors;
 169
 170         if (max_write_same_sectors == 0)
 171                 return -EOPNOTSUPP;
 172
 173         atomic_set(&bb.done, 1);
 174         bb.error = 0;
 175         bb.wait = &wait;
 176
 177         while (nr_sects) {
 178                 bio = bio_alloc(gfp_mask, 1);
 179                 if (!bio) {
 180                         ret = -ENOMEM;
 181                         break;
 182                 }
 183
 184                 bio->bi_iter.bi_sector = sector;
 185                 bio->bi_end_io = bio_batch_end_io;
 186                 bio->bi_bdev = bdev;
 187                 bio->bi_private = &bb;
 188                 bio->bi_vcnt = 1;
 189                 bio->bi_io_vec->bv_page = page;
 190                 bio->bi_io_vec->bv_offset = 0;
 191                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 192
 193                 if (nr_sects > max_write_same_sectors) {
 194                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 195                         nr_sects -= max_write_same_sectors;
 196                         sector += max_write_same_sectors;
 197                 } else {
 198                         bio->bi_iter.bi_size = nr_sects << 9;
 199                         nr_sects = 0;
 200                 }
 201
 202                 atomic_inc(&bb.done);
 203                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
 204         }
 205
 206         /* Wait for bios in-flight */
 207         if (!atomic_dec_and_test(&bb.done))
 208                 wait_for_completion_io(&wait);
 209
 210         if (bb.error)
 211                 return bb.error;
 212         return ret;
 213 }
 214 EXPORT_SYMBOL(blkdev_issue_write_same);
 215
 216 /**
 217  * blkdev_issue_zeroout - generate number of zero filed write bios
 218  * @bdev:       blockdev to issue
 219  * @sector:     start sector
 220  * @nr_sects:   number of sectors to write
 221  * @gfp_mask:   memory allocation flags (for bio_alloc)
 222  *
 223  * Description:
 224  *  Generate and issue number of bios with zerofiled pages.
 225  */
 226
 227 static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 228                                   sector_t nr_sects, gfp_t gfp_mask)
 229 {
 230         int ret;
 231         struct bio *bio;
 232         struct bio_batch bb;
 233         unsigned int sz;
 234         DECLARE_COMPLETION_ONSTACK(wait);
 235
 236         atomic_set(&bb.done, 1);
 237         bb.error = 0;
 238         bb.wait = &wait;
 239
 240         ret = 0;
 241         while (nr_sects != 0) {
 242                 bio = bio_alloc(gfp_mask,
 243                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
 244                 if (!bio) {
 245                         ret = -ENOMEM;
 246                         break;
 247                 }
 248
 249                 bio->bi_iter.bi_sector = sector;
 250                 bio->bi_bdev   = bdev;
 251                 bio->bi_end_io = bio_batch_end_io;
 252                 bio->bi_private = &bb;
 253
 254                 while (nr_sects != 0) {
 255                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 256                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 257                         nr_sects -= ret >> 9;
 258                         sector += ret >> 9;
 259                         if (ret < (sz << 9))
 260                                 break;
 261                 }
 262                 ret = 0;
 263                 atomic_inc(&bb.done);
 264                 submit_bio(WRITE, bio);
 265         }
 266
 267         /* Wait for bios in-flight */
 268         if (!atomic_dec_and_test(&bb.done))
 269                 wait_for_completion_io(&wait);
 270
 271         if (bb.error)
 272                 return bb.error;
 273         return ret;
 274 }
 275
 276 /**
 277  * blkdev_issue_zeroout - zero-fill a block range
 278  * @bdev:       blockdev to write
 279  * @sector:     start sector
 280  * @nr_sects:   number of sectors to write
 281  * @gfp_mask:   memory allocation flags (for bio_alloc)
 282  * @discard:    whether to discard the block range
 283  *
 284  * Description:
 285  *  Zero-fill a block range.  If the discard flag is set and the block
 286  *  device guarantees that subsequent READ operations to the block range
 287  *  in question will return zeroes, the blocks will be discarded. Should
 288  *  the discard request fail, if the discard flag is not set, or if
 289  *  discard_zeroes_data is not supported, this function will resort to
 290  *  zeroing the blocks manually, thus provisioning (allocating,
 291  *  anchoring) them. If the block device supports the WRITE SAME command
 292  *  blkdev_issue_zeroout() will use it to optimize the process of
 293  *  clearing the block range. Otherwise the zeroing will be performed
 294  *  using regular WRITE calls.
 295  */
 296
 297 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 298                          sector_t nr_sects, gfp_t gfp_mask, bool discard)
 299 {
 300         struct request_queue *q = bdev_get_queue(bdev);
 301
 302         if (discard && blk_queue_discard(q) && q->limits.discard_zeroes_data &&
 303             blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, 0) == 0)
 304                 return 0;
 305
 306         if (bdev_write_same(bdev) &&
 307             blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 308                                     ZERO_PAGE(0)) == 0)
 309                 return 0;
 310
 311         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 312 }
 313 EXPORT_SYMBOL(blkdev_issue_zeroout);