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         unsigned long           flags;
  15         struct completion       *wait;
  16 };
  17
  18 static void bio_batch_end_io(struct bio *bio, int err)
  19 {
  20         struct bio_batch *bb = bio->bi_private;
  21
  22         if (err && (err != -EOPNOTSUPP))
  23                 clear_bit(BIO_UPTODATE, &bb->flags);
  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.flags = 1 << BIO_UPTODATE;
  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_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_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         blk_finish_plug(&plug);
 124
 125         /* Wait for bios in-flight */
 126         if (!atomic_dec_and_test(&bb.done))
 127                 wait_for_completion_io(&wait);
 128
 129         if (!test_bit(BIO_UPTODATE, &bb.flags))
 130                 ret = -EIO;
 131
 132         return ret;
 133 }
 134 EXPORT_SYMBOL(blkdev_issue_discard);
 135
 136 /**
 137  * blkdev_issue_write_same - queue a write same operation
 138  * @bdev:       target blockdev
 139  * @sector:     start sector
 140  * @nr_sects:   number of sectors to write
 141  * @gfp_mask:   memory allocation flags (for bio_alloc)
 142  * @page:       page containing data to write
 143  *
 144  * Description:
 145  *    Issue a write same request for the sectors in question.
 146  */
 147 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 148                             sector_t nr_sects, gfp_t gfp_mask,
 149                             struct page *page)
 150 {
 151         DECLARE_COMPLETION_ONSTACK(wait);
 152         struct request_queue *q = bdev_get_queue(bdev);
 153         unsigned int max_write_same_sectors;
 154         struct bio_batch bb;
 155         struct bio *bio;
 156         int ret = 0;
 157
 158         if (!q)
 159                 return -ENXIO;
 160
 161         max_write_same_sectors = q->limits.max_write_same_sectors;
 162
 163         if (max_write_same_sectors == 0)
 164                 return -EOPNOTSUPP;
 165
 166         atomic_set(&bb.done, 1);
 167         bb.flags = 1 << BIO_UPTODATE;
 168         bb.wait = &wait;
 169
 170         while (nr_sects) {
 171                 bio = bio_alloc(gfp_mask, 1);
 172                 if (!bio) {
 173                         ret = -ENOMEM;
 174                         break;
 175                 }
 176
 177                 bio->bi_sector = sector;
 178                 bio->bi_end_io = bio_batch_end_io;
 179                 bio->bi_bdev = bdev;
 180                 bio->bi_private = &bb;
 181                 bio->bi_vcnt = 1;
 182                 bio->bi_io_vec->bv_page = page;
 183                 bio->bi_io_vec->bv_offset = 0;
 184                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 185
 186                 if (nr_sects > max_write_same_sectors) {
 187                         bio->bi_size = max_write_same_sectors << 9;
 188                         nr_sects -= max_write_same_sectors;
 189                         sector += max_write_same_sectors;
 190                 } else {
 191                         bio->bi_size = nr_sects << 9;
 192                         nr_sects = 0;
 193                 }
 194
 195                 atomic_inc(&bb.done);
 196                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
 197         }
 198
 199         /* Wait for bios in-flight */
 200         if (!atomic_dec_and_test(&bb.done))
 201                 wait_for_completion_io(&wait);
 202
 203         if (!test_bit(BIO_UPTODATE, &bb.flags))
 204                 ret = -ENOTSUPP;
 205
 206         return ret;
 207 }
 208 EXPORT_SYMBOL(blkdev_issue_write_same);
 209
 210 /**
 211  * blkdev_issue_zeroout - generate number of zero filed write bios
 212  * @bdev:       blockdev to issue
 213  * @sector:     start sector
 214  * @nr_sects:   number of sectors to write
 215  * @gfp_mask:   memory allocation flags (for bio_alloc)
 216  *
 217  * Description:
 218  *  Generate and issue number of bios with zerofiled pages.
 219  */
 220
 221 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 222                         sector_t nr_sects, gfp_t gfp_mask)
 223 {
 224         int ret;
 225         struct bio *bio;
 226         struct bio_batch bb;
 227         unsigned int sz;
 228         DECLARE_COMPLETION_ONSTACK(wait);
 229
 230         atomic_set(&bb.done, 1);
 231         bb.flags = 1 << BIO_UPTODATE;
 232         bb.wait = &wait;
 233
 234         ret = 0;
 235         while (nr_sects != 0) {
 236                 bio = bio_alloc(gfp_mask,
 237                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
 238                 if (!bio) {
 239                         ret = -ENOMEM;
 240                         break;
 241                 }
 242
 243                 bio->bi_sector = sector;
 244                 bio->bi_bdev   = bdev;
 245                 bio->bi_end_io = bio_batch_end_io;
 246                 bio->bi_private = &bb;
 247
 248                 while (nr_sects != 0) {
 249                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 250                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 251                         nr_sects -= ret >> 9;
 252                         sector += ret >> 9;
 253                         if (ret < (sz << 9))
 254                                 break;
 255                 }
 256                 ret = 0;
 257                 atomic_inc(&bb.done);
 258                 submit_bio(WRITE, bio);
 259         }
 260
 261         /* Wait for bios in-flight */
 262         if (!atomic_dec_and_test(&bb.done))
 263                 wait_for_completion_io(&wait);
 264
 265         if (!test_bit(BIO_UPTODATE, &bb.flags))
 266                 /* One of bios in the batch was completed with error.*/
 267                 ret = -EIO;
 268
 269         return ret;
 270 }
 271
 272 /**
 273  * blkdev_issue_zeroout - zero-fill a block range
 274  * @bdev:       blockdev to write
 275  * @sector:     start sector
 276  * @nr_sects:   number of sectors to write
 277  * @gfp_mask:   memory allocation flags (for bio_alloc)
 278  *
 279  * Description:
 280  *  Generate and issue number of bios with zerofiled pages.
 281  */
 282
 283 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 284                          sector_t nr_sects, gfp_t gfp_mask)
 285 {
 286         if (bdev_write_same(bdev)) {
 287                 unsigned char bdn[BDEVNAME_SIZE];
 288
 289                 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 290                                              ZERO_PAGE(0)))
 291                         return 0;
 292
 293                 bdevname(bdev, bdn);
 294                 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
 295         }
 296
 297         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 298 }
 299 EXPORT_SYMBOL(blkdev_issue_zeroout);