block/blk-lib.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Functions related to generic helpers functions
   4  */
   5 #include <linux/kernel.h>
   6 #include <linux/module.h>
   7 #include <linux/bio.h>
   8 #include <linux/blkdev.h>
   9 #include <linux/scatterlist.h>
  10
  11 #include "blk.h"
  12
  13 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  14                 sector_t nr_sects, gfp_t gfp_mask, int flags,
  15                 struct bio **biop)
  16 {
  17         struct request_queue *q = bdev_get_queue(bdev);
  18         struct bio *bio = *biop;
  19         unsigned int op;
  20         sector_t bs_mask, part_offset = 0;
  21
  22         if (bdev_read_only(bdev))
  23                 return -EPERM;
  24
  25         if (flags & BLKDEV_DISCARD_SECURE) {
  26                 if (!blk_queue_secure_erase(q))
  27                         return -EOPNOTSUPP;
  28                 op = REQ_OP_SECURE_ERASE;
  29         } else {
  30                 if (!blk_queue_discard(q))
  31                         return -EOPNOTSUPP;
  32                 op = REQ_OP_DISCARD;
  33         }
  34
  35         /* In case the discard granularity isn't set by buggy device driver */
  36         if (WARN_ON_ONCE(!q->limits.discard_granularity)) {
  37                 char dev_name[BDEVNAME_SIZE];
  38
  39                 bdevname(bdev, dev_name);
  40                 pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  41                 return -EOPNOTSUPP;
  42         }
  43
  44         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  45         if ((sector | nr_sects) & bs_mask)
  46                 return -EINVAL;
  47
  48         if (!nr_sects)
  49                 return -EINVAL;
  50
  51         /* In case the discard request is in a partition */
  52         if (bdev_is_partition(bdev))
  53                 part_offset = bdev->bd_start_sect;
  54
  55         while (nr_sects) {
  56                 sector_t granularity_aligned_lba, req_sects;
  57                 sector_t sector_mapped = sector + part_offset;
  58
  59                 granularity_aligned_lba = round_up(sector_mapped,
  60                                 q->limits.discard_granularity >> SECTOR_SHIFT);
  61
  62                 /*
  63                  * Check whether the discard bio starts at a discard_granularity
  64                  * aligned LBA,
  65                  * - If no: set (granularity_aligned_lba - sector_mapped) to
  66                  *   bi_size of the first split bio, then the second bio will
  67                  *   start at a discard_granularity aligned LBA on the device.
  68                  * - If yes: use bio_aligned_discard_max_sectors() as the max
  69                  *   possible bi_size of the first split bio. Then when this bio
  70                  *   is split in device drive, the split ones are very probably
  71                  *   to be aligned to discard_granularity of the device's queue.
  72                  */
  73                 if (granularity_aligned_lba == sector_mapped)
  74                         req_sects = min_t(sector_t, nr_sects,
  75                                           bio_aligned_discard_max_sectors(q));
  76                 else
  77                         req_sects = min_t(sector_t, nr_sects,
  78                                           granularity_aligned_lba - sector_mapped);
  79
  80                 WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
  81
  82                 bio = blk_next_bio(bio, bdev, 0, op, gfp_mask);
  83                 bio->bi_iter.bi_sector = sector;
  84                 bio->bi_iter.bi_size = req_sects << 9;
  85                 sector += req_sects;
  86                 nr_sects -= req_sects;
  87
  88                 /*
  89                  * We can loop for a long time in here, if someone does
  90                  * full device discards (like mkfs). Be nice and allow
  91                  * us to schedule out to avoid softlocking if preempt
  92                  * is disabled.
  93                  */
  94                 cond_resched();
  95         }
  96
  97         *biop = bio;
  98         return 0;
  99 }
 100 EXPORT_SYMBOL(__blkdev_issue_discard);
 101
 102 /**
 103  * blkdev_issue_discard - queue a discard
 104  * @bdev:       blockdev to issue discard for
 105  * @sector:     start sector
 106  * @nr_sects:   number of sectors to discard
 107  * @gfp_mask:   memory allocation flags (for bio_alloc)
 108  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 109  *
 110  * Description:
 111  *    Issue a discard request for the sectors in question.
 112  */
 113 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 114                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 115 {
 116         struct bio *bio = NULL;
 117         struct blk_plug plug;
 118         int ret;
 119
 120         blk_start_plug(&plug);
 121         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 122                         &bio);
 123         if (!ret && bio) {
 124                 ret = submit_bio_wait(bio);
 125                 if (ret == -EOPNOTSUPP)
 126                         ret = 0;
 127                 bio_put(bio);
 128         }
 129         blk_finish_plug(&plug);
 130
 131         return ret;
 132 }
 133 EXPORT_SYMBOL(blkdev_issue_discard);
 134
 135 /**
 136  * __blkdev_issue_write_same - generate number of bios with same page
 137  * @bdev:       target blockdev
 138  * @sector:     start sector
 139  * @nr_sects:   number of sectors to write
 140  * @gfp_mask:   memory allocation flags (for bio_alloc)
 141  * @page:       page containing data to write
 142  * @biop:       pointer to anchor bio
 143  *
 144  * Description:
 145  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 146  */
 147 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 148                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 149                 struct bio **biop)
 150 {
 151         struct request_queue *q = bdev_get_queue(bdev);
 152         unsigned int max_write_same_sectors;
 153         struct bio *bio = *biop;
 154         sector_t bs_mask;
 155
 156         if (bdev_read_only(bdev))
 157                 return -EPERM;
 158
 159         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 160         if ((sector | nr_sects) & bs_mask)
 161                 return -EINVAL;
 162
 163         if (!bdev_write_same(bdev))
 164                 return -EOPNOTSUPP;
 165
 166         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 167         max_write_same_sectors = bio_allowed_max_sectors(q);
 168
 169         while (nr_sects) {
 170                 bio = blk_next_bio(bio, bdev, 1, REQ_OP_WRITE_SAME, gfp_mask);
 171                 bio->bi_iter.bi_sector = sector;
 172                 bio->bi_vcnt = 1;
 173                 bio->bi_io_vec->bv_page = page;
 174                 bio->bi_io_vec->bv_offset = 0;
 175                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 176
 177                 if (nr_sects > max_write_same_sectors) {
 178                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 179                         nr_sects -= max_write_same_sectors;
 180                         sector += max_write_same_sectors;
 181                 } else {
 182                         bio->bi_iter.bi_size = nr_sects << 9;
 183                         nr_sects = 0;
 184                 }
 185                 cond_resched();
 186         }
 187
 188         *biop = bio;
 189         return 0;
 190 }
 191
 192 /**
 193  * blkdev_issue_write_same - queue a write same operation
 194  * @bdev:       target blockdev
 195  * @sector:     start sector
 196  * @nr_sects:   number of sectors to write
 197  * @gfp_mask:   memory allocation flags (for bio_alloc)
 198  * @page:       page containing data
 199  *
 200  * Description:
 201  *    Issue a write same request for the sectors in question.
 202  */
 203 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 204                                 sector_t nr_sects, gfp_t gfp_mask,
 205                                 struct page *page)
 206 {
 207         struct bio *bio = NULL;
 208         struct blk_plug plug;
 209         int ret;
 210
 211         blk_start_plug(&plug);
 212         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 213                         &bio);
 214         if (ret == 0 && bio) {
 215                 ret = submit_bio_wait(bio);
 216                 bio_put(bio);
 217         }
 218         blk_finish_plug(&plug);
 219         return ret;
 220 }
 221 EXPORT_SYMBOL(blkdev_issue_write_same);
 222
 223 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 224                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 225                 struct bio **biop, unsigned flags)
 226 {
 227         struct bio *bio = *biop;
 228         unsigned int max_write_zeroes_sectors;
 229
 230         if (bdev_read_only(bdev))
 231                 return -EPERM;
 232
 233         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 234         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 235
 236         if (max_write_zeroes_sectors == 0)
 237                 return -EOPNOTSUPP;
 238
 239         while (nr_sects) {
 240                 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
 241                 bio->bi_iter.bi_sector = sector;
 242                 if (flags & BLKDEV_ZERO_NOUNMAP)
 243                         bio->bi_opf |= REQ_NOUNMAP;
 244
 245                 if (nr_sects > max_write_zeroes_sectors) {
 246                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 247                         nr_sects -= max_write_zeroes_sectors;
 248                         sector += max_write_zeroes_sectors;
 249                 } else {
 250                         bio->bi_iter.bi_size = nr_sects << 9;
 251                         nr_sects = 0;
 252                 }
 253                 cond_resched();
 254         }
 255
 256         *biop = bio;
 257         return 0;
 258 }
 259
 260 /*
 261  * Convert a number of 512B sectors to a number of pages.
 262  * The result is limited to a number of pages that can fit into a BIO.
 263  * Also make sure that the result is always at least 1 (page) for the cases
 264  * where nr_sects is lower than the number of sectors in a page.
 265  */
 266 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 267 {
 268         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 269
 270         return min(pages, (sector_t)BIO_MAX_VECS);
 271 }
 272
 273 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 274                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 275                 struct bio **biop)
 276 {
 277         struct bio *bio = *biop;
 278         int bi_size = 0;
 279         unsigned int sz;
 280
 281         if (bdev_read_only(bdev))
 282                 return -EPERM;
 283
 284         while (nr_sects != 0) {
 285                 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
 286                                    REQ_OP_WRITE, gfp_mask);
 287                 bio->bi_iter.bi_sector = sector;
 288
 289                 while (nr_sects != 0) {
 290                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 291                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 292                         nr_sects -= bi_size >> 9;
 293                         sector += bi_size >> 9;
 294                         if (bi_size < sz)
 295                                 break;
 296                 }
 297                 cond_resched();
 298         }
 299
 300         *biop = bio;
 301         return 0;
 302 }
 303
 304 /**
 305  * __blkdev_issue_zeroout - generate number of zero filed write bios
 306  * @bdev:       blockdev to issue
 307  * @sector:     start sector
 308  * @nr_sects:   number of sectors to write
 309  * @gfp_mask:   memory allocation flags (for bio_alloc)
 310  * @biop:       pointer to anchor bio
 311  * @flags:      controls detailed behavior
 312  *
 313  * Description:
 314  *  Zero-fill a block range, either using hardware offload or by explicitly
 315  *  writing zeroes to the device.
 316  *
 317  *  If a device is using logical block provisioning, the underlying space will
 318  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 319  *
 320  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 321  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 322  */
 323 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 324                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 325                 unsigned flags)
 326 {
 327         int ret;
 328         sector_t bs_mask;
 329
 330         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 331         if ((sector | nr_sects) & bs_mask)
 332                 return -EINVAL;
 333
 334         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 335                         biop, flags);
 336         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 337                 return ret;
 338
 339         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 340                                          biop);
 341 }
 342 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 343
 344 /**
 345  * blkdev_issue_zeroout - zero-fill a block range
 346  * @bdev:       blockdev to write
 347  * @sector:     start sector
 348  * @nr_sects:   number of sectors to write
 349  * @gfp_mask:   memory allocation flags (for bio_alloc)
 350  * @flags:      controls detailed behavior
 351  *
 352  * Description:
 353  *  Zero-fill a block range, either using hardware offload or by explicitly
 354  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 355  *  valid values for %flags.
 356  */
 357 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 358                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 359 {
 360         int ret = 0;
 361         sector_t bs_mask;
 362         struct bio *bio;
 363         struct blk_plug plug;
 364         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 365
 366         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 367         if ((sector | nr_sects) & bs_mask)
 368                 return -EINVAL;
 369
 370 retry:
 371         bio = NULL;
 372         blk_start_plug(&plug);
 373         if (try_write_zeroes) {
 374                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 375                                                   gfp_mask, &bio, flags);
 376         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 377                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 378                                                 gfp_mask, &bio);
 379         } else {
 380                 /* No zeroing offload support */
 381                 ret = -EOPNOTSUPP;
 382         }
 383         if (ret == 0 && bio) {
 384                 ret = submit_bio_wait(bio);
 385                 bio_put(bio);
 386         }
 387         blk_finish_plug(&plug);
 388         if (ret && try_write_zeroes) {
 389                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 390                         try_write_zeroes = false;
 391                         goto retry;
 392                 }
 393                 if (!bdev_write_zeroes_sectors(bdev)) {
 394                         /*
 395                          * Zeroing offload support was indicated, but the
 396                          * device reported ILLEGAL REQUEST (for some devices
 397                          * there is no non-destructive way to verify whether
 398                          * WRITE ZEROES is actually supported).
 399                          */
 400                         ret = -EOPNOTSUPP;
 401                 }
 402         }
 403
 404         return ret;
 405 }
 406 EXPORT_SYMBOL(blkdev_issue_zeroout);