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 static struct bio *next_bio(struct bio *bio, unsigned int nr_pages,
  13                 gfp_t gfp)
  14 {
  15         struct bio *new = bio_alloc(gfp, nr_pages);
  16
  17         if (bio) {
  18                 bio_chain(bio, new);
  19                 submit_bio(bio);
  20         }
  21
  22         return new;
  23 }
  24
  25 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  26                 sector_t nr_sects, gfp_t gfp_mask, int flags,
  27                 struct bio **biop)
  28 {
  29         struct request_queue *q = bdev_get_queue(bdev);
  30         struct bio *bio = *biop;
  31         unsigned int granularity;
  32         unsigned int op;
  33         int alignment;
  34         sector_t bs_mask;
  35
  36         if (!q)
  37                 return -ENXIO;
  38
  39         if (flags & BLKDEV_DISCARD_SECURE) {
  40                 if (flags & BLKDEV_DISCARD_ZERO)
  41                         return -EOPNOTSUPP;
  42                 if (!blk_queue_secure_erase(q))
  43                         return -EOPNOTSUPP;
  44                 op = REQ_OP_SECURE_ERASE;
  45         } else {
  46                 if (!blk_queue_discard(q))
  47                         return -EOPNOTSUPP;
  48                 if ((flags & BLKDEV_DISCARD_ZERO) &&
  49                     !q->limits.discard_zeroes_data)
  50                         return -EOPNOTSUPP;
  51                 op = REQ_OP_DISCARD;
  52         }
  53
  54         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  55         if ((sector | nr_sects) & bs_mask)
  56                 return -EINVAL;
  57
  58         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  59         granularity = max(q->limits.discard_granularity >> 9, 1U);
  60         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  61
  62         while (nr_sects) {
  63                 unsigned int req_sects;
  64                 sector_t end_sect, tmp;
  65
  66                 /* Make sure bi_size doesn't overflow */
  67                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  68
  69                 /**
  70                  * If splitting a request, and the next starting sector would be
  71                  * misaligned, stop the discard at the previous aligned sector.
  72                  */
  73                 end_sect = sector + req_sects;
  74                 tmp = end_sect;
  75                 if (req_sects < nr_sects &&
  76                     sector_div(tmp, granularity) != alignment) {
  77                         end_sect = end_sect - alignment;
  78                         sector_div(end_sect, granularity);
  79                         end_sect = end_sect * granularity + alignment;
  80                         req_sects = end_sect - sector;
  81                 }
  82
  83                 bio = next_bio(bio, 0, gfp_mask);
  84                 bio->bi_iter.bi_sector = sector;
  85                 bio->bi_bdev = bdev;
  86                 bio_set_op_attrs(bio, op, 0);
  87
  88                 bio->bi_iter.bi_size = req_sects << 9;
  89                 nr_sects -= req_sects;
  90                 sector = end_sect;
  91
  92                 /*
  93                  * We can loop for a long time in here, if someone does
  94                  * full device discards (like mkfs). Be nice and allow
  95                  * us to schedule out to avoid softlocking if preempt
  96                  * is disabled.
  97                  */
  98                 cond_resched();
  99         }
 100
 101         *biop = bio;
 102         return 0;
 103 }
 104 EXPORT_SYMBOL(__blkdev_issue_discard);
 105
 106 /**
 107  * blkdev_issue_discard - queue a discard
 108  * @bdev:       blockdev to issue discard for
 109  * @sector:     start sector
 110  * @nr_sects:   number of sectors to discard
 111  * @gfp_mask:   memory allocation flags (for bio_alloc)
 112  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 113  *
 114  * Description:
 115  *    Issue a discard request for the sectors in question.
 116  */
 117 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 118                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 119 {
 120         struct bio *bio = NULL;
 121         struct blk_plug plug;
 122         int ret;
 123
 124         blk_start_plug(&plug);
 125         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 126                         &bio);
 127         if (!ret && bio) {
 128                 ret = submit_bio_wait(bio);
 129                 if (ret == -EOPNOTSUPP && !(flags & BLKDEV_DISCARD_ZERO))
 130                         ret = 0;
 131                 bio_put(bio);
 132         }
 133         blk_finish_plug(&plug);
 134
 135         return ret;
 136 }
 137 EXPORT_SYMBOL(blkdev_issue_discard);
 138
 139 /**
 140  * __blkdev_issue_write_same - generate number of bios with same page
 141  * @bdev:       target blockdev
 142  * @sector:     start sector
 143  * @nr_sects:   number of sectors to write
 144  * @gfp_mask:   memory allocation flags (for bio_alloc)
 145  * @page:       page containing data to write
 146  * @biop:       pointer to anchor bio
 147  *
 148  * Description:
 149  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 150  */
 151 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 152                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 153                 struct bio **biop)
 154 {
 155         struct request_queue *q = bdev_get_queue(bdev);
 156         unsigned int max_write_same_sectors;
 157         struct bio *bio = *biop;
 158         sector_t bs_mask;
 159
 160         if (!q)
 161                 return -ENXIO;
 162
 163         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 164         if ((sector | nr_sects) & bs_mask)
 165                 return -EINVAL;
 166
 167         if (!bdev_write_same(bdev))
 168                 return -EOPNOTSUPP;
 169
 170         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 171         max_write_same_sectors = UINT_MAX >> 9;
 172
 173         while (nr_sects) {
 174                 bio = next_bio(bio, 1, gfp_mask);
 175                 bio->bi_iter.bi_sector = sector;
 176                 bio->bi_bdev = bdev;
 177                 bio->bi_vcnt = 1;
 178                 bio->bi_io_vec->bv_page = page;
 179                 bio->bi_io_vec->bv_offset = 0;
 180                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 181                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 182
 183                 if (nr_sects > max_write_same_sectors) {
 184                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 185                         nr_sects -= max_write_same_sectors;
 186                         sector += max_write_same_sectors;
 187                 } else {
 188                         bio->bi_iter.bi_size = nr_sects << 9;
 189                         nr_sects = 0;
 190                 }
 191                 cond_resched();
 192         }
 193
 194         *biop = bio;
 195         return 0;
 196 }
 197
 198 /**
 199  * blkdev_issue_write_same - queue a write same operation
 200  * @bdev:       target blockdev
 201  * @sector:     start sector
 202  * @nr_sects:   number of sectors to write
 203  * @gfp_mask:   memory allocation flags (for bio_alloc)
 204  * @page:       page containing data
 205  *
 206  * Description:
 207  *    Issue a write same request for the sectors in question.
 208  */
 209 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 210                                 sector_t nr_sects, gfp_t gfp_mask,
 211                                 struct page *page)
 212 {
 213         struct bio *bio = NULL;
 214         struct blk_plug plug;
 215         int ret;
 216
 217         blk_start_plug(&plug);
 218         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 219                         &bio);
 220         if (ret == 0 && bio) {
 221                 ret = submit_bio_wait(bio);
 222                 bio_put(bio);
 223         }
 224         blk_finish_plug(&plug);
 225         return ret;
 226 }
 227 EXPORT_SYMBOL(blkdev_issue_write_same);
 228
 229 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 230                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 231                 struct bio **biop, unsigned flags)
 232 {
 233         struct bio *bio = *biop;
 234         unsigned int max_write_zeroes_sectors;
 235         struct request_queue *q = bdev_get_queue(bdev);
 236
 237         if (!q)
 238                 return -ENXIO;
 239
 240         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 241         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 242
 243         if (max_write_zeroes_sectors == 0)
 244                 return -EOPNOTSUPP;
 245
 246         while (nr_sects) {
 247                 bio = next_bio(bio, 0, gfp_mask);
 248                 bio->bi_iter.bi_sector = sector;
 249                 bio->bi_bdev = bdev;
 250                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 251                 if (flags & BLKDEV_ZERO_NOUNMAP)
 252                         bio->bi_opf |= REQ_NOUNMAP;
 253
 254                 if (nr_sects > max_write_zeroes_sectors) {
 255                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 256                         nr_sects -= max_write_zeroes_sectors;
 257                         sector += max_write_zeroes_sectors;
 258                 } else {
 259                         bio->bi_iter.bi_size = nr_sects << 9;
 260                         nr_sects = 0;
 261                 }
 262                 cond_resched();
 263         }
 264
 265         *biop = bio;
 266         return 0;
 267 }
 268
 269 /**
 270  * __blkdev_issue_zeroout - generate number of zero filed write bios
 271  * @bdev:       blockdev to issue
 272  * @sector:     start sector
 273  * @nr_sects:   number of sectors to write
 274  * @gfp_mask:   memory allocation flags (for bio_alloc)
 275  * @biop:       pointer to anchor bio
 276  * @flags:      controls detailed behavior
 277  *
 278  * Description:
 279  *  Zero-fill a block range, either using hardware offload or by explicitly
 280  *  writing zeroes to the device.
 281  *
 282  *  Note that this function may fail with -EOPNOTSUPP if the driver signals
 283  *  zeroing offload support, but the device fails to process the command (for
 284  *  some devices there is no non-destructive way to verify whether this
 285  *  operation is actually supported).  In this case the caller should call
 286  *  retry the call to blkdev_issue_zeroout() and the fallback path will be used.
 287  *
 288  *  If a device is using logical block provisioning, the underlying space will
 289  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 290  *
 291  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 292  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 293  */
 294 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 295                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 296                 unsigned flags)
 297 {
 298         int ret;
 299         int bi_size = 0;
 300         struct bio *bio = *biop;
 301         unsigned int sz;
 302         sector_t bs_mask;
 303
 304         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 305         if ((sector | nr_sects) & bs_mask)
 306                 return -EINVAL;
 307
 308         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 309                         biop, flags);
 310         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 311                 goto out;
 312
 313         ret = 0;
 314         while (nr_sects != 0) {
 315                 bio = next_bio(bio, min(nr_sects, (sector_t)BIO_MAX_PAGES),
 316                                 gfp_mask);
 317                 bio->bi_iter.bi_sector = sector;
 318                 bio->bi_bdev   = bdev;
 319                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 320
 321                 while (nr_sects != 0) {
 322                         sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
 323                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 324                         nr_sects -= bi_size >> 9;
 325                         sector += bi_size >> 9;
 326                         if (bi_size < (sz << 9))
 327                                 break;
 328                 }
 329                 cond_resched();
 330         }
 331
 332         *biop = bio;
 333 out:
 334         return ret;
 335 }
 336 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 337
 338 /**
 339  * blkdev_issue_zeroout - zero-fill a block range
 340  * @bdev:       blockdev to write
 341  * @sector:     start sector
 342  * @nr_sects:   number of sectors to write
 343  * @gfp_mask:   memory allocation flags (for bio_alloc)
 344  * @flags:      controls detailed behavior
 345  *
 346  * Description:
 347  *  Zero-fill a block range, either using hardware offload or by explicitly
 348  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 349  *  valid values for %flags.
 350  */
 351 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 352                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 353 {
 354         int ret;
 355         struct bio *bio = NULL;
 356         struct blk_plug plug;
 357
 358         blk_start_plug(&plug);
 359         ret = __blkdev_issue_zeroout(bdev, sector, nr_sects, gfp_mask,
 360                         &bio, flags);
 361         if (ret == 0 && bio) {
 362                 ret = submit_bio_wait(bio);
 363                 bio_put(bio);
 364         }
 365         blk_finish_plug(&plug);
 366
 367         return ret;
 368 }
 369 EXPORT_SYMBOL(blkdev_issue_zeroout);