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         sector_t max_discard_sectors;
  47         sector_t granularity, 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;
  62         alignment = sector_div(alignment, granularity);
  63
  64         /*
  65          * Ensure that max_discard_sectors is of the proper
  66          * granularity, so that requests stay aligned after a split.
  67          */
  68         max_discard_sectors = min(q->limits.max_discard_sectors, UINT_MAX >> 9);
  69         sector_div(max_discard_sectors, granularity);
  70         max_discard_sectors *= granularity;
  71         if (unlikely(!max_discard_sectors)) {
  72                 /* Avoid infinite loop below. Being cautious never hurts. */
  73                 return -EOPNOTSUPP;
  74         }
  75
  76         if (flags & BLKDEV_DISCARD_SECURE) {
  77                 if (!blk_queue_secdiscard(q))
  78                         return -EOPNOTSUPP;
  79                 type |= REQ_SECURE;
  80         }
  81
  82         atomic_set(&bb.done, 1);
  83         bb.flags = 1 << BIO_UPTODATE;
  84         bb.wait = &wait;
  85
  86         blk_start_plug(&plug);
  87         while (nr_sects) {
  88                 unsigned int req_sects;
  89                 sector_t end_sect, tmp;
  90
  91                 bio = bio_alloc(gfp_mask, 1);
  92                 if (!bio) {
  93                         ret = -ENOMEM;
  94                         break;
  95                 }
  96
  97                 req_sects = min_t(sector_t, nr_sects, max_discard_sectors);
  98
  99                 /*
 100                  * If splitting a request, and the next starting sector would be
 101                  * misaligned, stop the discard at the previous aligned sector.
 102                  */
 103                 end_sect = sector + req_sects;
 104                 tmp = end_sect;
 105                 if (req_sects < nr_sects &&
 106                     sector_div(tmp, granularity) != alignment) {
 107                         end_sect = end_sect - alignment;
 108                         sector_div(end_sect, granularity);
 109                         end_sect = end_sect * granularity + alignment;
 110                         req_sects = end_sect - sector;
 111                 }
 112
 113                 bio->bi_sector = sector;
 114                 bio->bi_end_io = bio_batch_end_io;
 115                 bio->bi_bdev = bdev;
 116                 bio->bi_private = &bb;
 117
 118                 bio->bi_size = req_sects << 9;
 119                 nr_sects -= req_sects;
 120                 sector = end_sect;
 121
 122                 atomic_inc(&bb.done);
 123                 submit_bio(type, bio);
 124
 125                 /*
 126                  * We can loop for a long time in here, if someone does
 127                  * full device discards (like mkfs). Be nice and allow
 128                  * us to schedule out to avoid softlocking if preempt
 129                  * is disabled.
 130                  */
 131                 cond_resched();
 132         }
 133         blk_finish_plug(&plug);
 134
 135         /* Wait for bios in-flight */
 136         if (!atomic_dec_and_test(&bb.done))
 137                 wait_for_completion_io(&wait);
 138
 139         if (!test_bit(BIO_UPTODATE, &bb.flags))
 140                 ret = -EIO;
 141
 142         return ret;
 143 }
 144 EXPORT_SYMBOL(blkdev_issue_discard);
 145
 146 /**
 147  * blkdev_issue_write_same - queue a write same operation
 148  * @bdev:       target blockdev
 149  * @sector:     start sector
 150  * @nr_sects:   number of sectors to write
 151  * @gfp_mask:   memory allocation flags (for bio_alloc)
 152  * @page:       page containing data to write
 153  *
 154  * Description:
 155  *    Issue a write same request for the sectors in question.
 156  */
 157 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 158                             sector_t nr_sects, gfp_t gfp_mask,
 159                             struct page *page)
 160 {
 161         DECLARE_COMPLETION_ONSTACK(wait);
 162         struct request_queue *q = bdev_get_queue(bdev);
 163         unsigned int max_write_same_sectors;
 164         struct bio_batch bb;
 165         struct bio *bio;
 166         int ret = 0;
 167
 168         if (!q)
 169                 return -ENXIO;
 170
 171         max_write_same_sectors = q->limits.max_write_same_sectors;
 172
 173         if (max_write_same_sectors == 0)
 174                 return -EOPNOTSUPP;
 175
 176         atomic_set(&bb.done, 1);
 177         bb.flags = 1 << BIO_UPTODATE;
 178         bb.wait = &wait;
 179
 180         while (nr_sects) {
 181                 bio = bio_alloc(gfp_mask, 1);
 182                 if (!bio) {
 183                         ret = -ENOMEM;
 184                         break;
 185                 }
 186
 187                 bio->bi_sector = sector;
 188                 bio->bi_end_io = bio_batch_end_io;
 189                 bio->bi_bdev = bdev;
 190                 bio->bi_private = &bb;
 191                 bio->bi_vcnt = 1;
 192                 bio->bi_io_vec->bv_page = page;
 193                 bio->bi_io_vec->bv_offset = 0;
 194                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 195
 196                 if (nr_sects > max_write_same_sectors) {
 197                         bio->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_size = nr_sects << 9;
 202                         nr_sects = 0;
 203                 }
 204
 205                 atomic_inc(&bb.done);
 206                 submit_bio(REQ_WRITE | REQ_WRITE_SAME, bio);
 207         }
 208
 209         /* Wait for bios in-flight */
 210         if (!atomic_dec_and_test(&bb.done))
 211                 wait_for_completion_io(&wait);
 212
 213         if (!test_bit(BIO_UPTODATE, &bb.flags))
 214                 ret = -ENOTSUPP;
 215
 216         return ret;
 217 }
 218 EXPORT_SYMBOL(blkdev_issue_write_same);
 219
 220 /**
 221  * blkdev_issue_zeroout - generate number of zero filed write bios
 222  * @bdev:       blockdev to issue
 223  * @sector:     start sector
 224  * @nr_sects:   number of sectors to write
 225  * @gfp_mask:   memory allocation flags (for bio_alloc)
 226  *
 227  * Description:
 228  *  Generate and issue number of bios with zerofiled pages.
 229  */
 230
 231 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 232                         sector_t nr_sects, gfp_t gfp_mask)
 233 {
 234         int ret;
 235         struct bio *bio;
 236         struct bio_batch bb;
 237         unsigned int sz;
 238         DECLARE_COMPLETION_ONSTACK(wait);
 239
 240         atomic_set(&bb.done, 1);
 241         bb.flags = 1 << BIO_UPTODATE;
 242         bb.wait = &wait;
 243
 244         ret = 0;
 245         while (nr_sects != 0) {
 246                 bio = bio_alloc(gfp_mask,
 247                                 min(nr_sects, (sector_t)BIO_MAX_PAGES));
 248                 if (!bio) {
 249                         ret = -ENOMEM;
 250                         break;
 251                 }
 252
 253                 bio->bi_sector = sector;
 254                 bio->bi_bdev   = bdev;
 255                 bio->bi_end_io = bio_batch_end_io;
 256                 bio->bi_private = &bb;
 257
 258                 while (nr_sects != 0) {
 259                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 260                         ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 261                         nr_sects -= ret >> 9;
 262                         sector += ret >> 9;
 263                         if (ret < (sz << 9))
 264                                 break;
 265                 }
 266                 ret = 0;
 267                 atomic_inc(&bb.done);
 268                 submit_bio(WRITE, bio);
 269         }
 270
 271         /* Wait for bios in-flight */
 272         if (!atomic_dec_and_test(&bb.done))
 273                 wait_for_completion_io(&wait);
 274
 275         if (!test_bit(BIO_UPTODATE, &bb.flags))
 276                 /* One of bios in the batch was completed with error.*/
 277                 ret = -EIO;
 278
 279         return ret;
 280 }
 281
 282 /**
 283  * blkdev_issue_zeroout - zero-fill a block range
 284  * @bdev:       blockdev to write
 285  * @sector:     start sector
 286  * @nr_sects:   number of sectors to write
 287  * @gfp_mask:   memory allocation flags (for bio_alloc)
 288  *
 289  * Description:
 290  *  Generate and issue number of bios with zerofiled pages.
 291  */
 292
 293 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 294                          sector_t nr_sects, gfp_t gfp_mask)
 295 {
 296         if (bdev_write_same(bdev)) {
 297                 unsigned char bdn[BDEVNAME_SIZE];
 298
 299                 if (!blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask,
 300                                              ZERO_PAGE(0)))
 301                         return 0;
 302
 303                 bdevname(bdev, bdn);
 304                 pr_err("%s: WRITE SAME failed. Manually zeroing.\n", bdn);
 305         }
 306
 307         return __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask);
 308 }
 309 EXPORT_SYMBOL(blkdev_issue_zeroout);