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 (!q)
  23                 return -ENXIO;
  24
  25         if (bdev_read_only(bdev))
  26                 return -EPERM;
  27
  28         if (flags & BLKDEV_DISCARD_SECURE) {
  29                 if (!blk_queue_secure_erase(q))
  30                         return -EOPNOTSUPP;
  31                 op = REQ_OP_SECURE_ERASE;
  32         } else {
  33                 if (!blk_queue_discard(q))
  34                         return -EOPNOTSUPP;
  35                 op = REQ_OP_DISCARD;
  36         }
  37
  38         /* In case the discard granularity isn't set by buggy device driver */
  39         if (WARN_ON_ONCE(!q->limits.discard_granularity)) {
  40                 char dev_name[BDEVNAME_SIZE];
  41
  42                 bdevname(bdev, dev_name);
  43                 pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  44                 return -EOPNOTSUPP;
  45         }
  46
  47         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  48         if ((sector | nr_sects) & bs_mask)
  49                 return -EINVAL;
  50
  51         if (!nr_sects)
  52                 return -EINVAL;
  53
  54         /* In case the discard request is in a partition */
  55         if (bdev_is_partition(bdev))
  56                 part_offset = bdev->bd_start_sect;
  57
  58         while (nr_sects) {
  59                 sector_t granularity_aligned_lba, req_sects;
  60                 sector_t sector_mapped = sector + part_offset;
  61
  62                 granularity_aligned_lba = round_up(sector_mapped,
  63                                 q->limits.discard_granularity >> SECTOR_SHIFT);
  64
  65                 /*
  66                  * Check whether the discard bio starts at a discard_granularity
  67                  * aligned LBA,
  68                  * - If no: set (granularity_aligned_lba - sector_mapped) to
  69                  *   bi_size of the first split bio, then the second bio will
  70                  *   start at a discard_granularity aligned LBA on the device.
  71                  * - If yes: use bio_aligned_discard_max_sectors() as the max
  72                  *   possible bi_size of the first split bio. Then when this bio
  73                  *   is split in device drive, the split ones are very probably
  74                  *   to be aligned to discard_granularity of the device's queue.
  75                  */
  76                 if (granularity_aligned_lba == sector_mapped)
  77                         req_sects = min_t(sector_t, nr_sects,
  78                                           bio_aligned_discard_max_sectors(q));
  79                 else
  80                         req_sects = min_t(sector_t, nr_sects,
  81                                           granularity_aligned_lba - sector_mapped);
  82
  83                 WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
  84
  85                 bio = blk_next_bio(bio, bdev, 0, op, gfp_mask);
  86                 bio->bi_iter.bi_sector = sector;
  87                 bio->bi_iter.bi_size = req_sects << 9;
  88                 sector += req_sects;
  89                 nr_sects -= req_sects;
  90
  91                 /*
  92                  * We can loop for a long time in here, if someone does
  93                  * full device discards (like mkfs). Be nice and allow
  94                  * us to schedule out to avoid softlocking if preempt
  95                  * is disabled.
  96                  */
  97                 cond_resched();
  98         }
  99
 100         *biop = bio;
 101         return 0;
 102 }
 103 EXPORT_SYMBOL(__blkdev_issue_discard);
 104
 105 /**
 106  * blkdev_issue_discard - queue a discard
 107  * @bdev:       blockdev to issue discard for
 108  * @sector:     start sector
 109  * @nr_sects:   number of sectors to discard
 110  * @gfp_mask:   memory allocation flags (for bio_alloc)
 111  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 112  *
 113  * Description:
 114  *    Issue a discard request for the sectors in question.
 115  */
 116 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 117                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 118 {
 119         struct bio *bio = NULL;
 120         struct blk_plug plug;
 121         int ret;
 122
 123         blk_start_plug(&plug);
 124         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 125                         &bio);
 126         if (!ret && bio) {
 127                 ret = submit_bio_wait(bio);
 128                 if (ret == -EOPNOTSUPP)
 129                         ret = 0;
 130                 bio_put(bio);
 131         }
 132         blk_finish_plug(&plug);
 133
 134         return ret;
 135 }
 136 EXPORT_SYMBOL(blkdev_issue_discard);
 137
 138 /**
 139  * __blkdev_issue_write_same - generate number of bios with same page
 140  * @bdev:       target blockdev
 141  * @sector:     start sector
 142  * @nr_sects:   number of sectors to write
 143  * @gfp_mask:   memory allocation flags (for bio_alloc)
 144  * @page:       page containing data to write
 145  * @biop:       pointer to anchor bio
 146  *
 147  * Description:
 148  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 149  */
 150 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 151                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 152                 struct bio **biop)
 153 {
 154         struct request_queue *q = bdev_get_queue(bdev);
 155         unsigned int max_write_same_sectors;
 156         struct bio *bio = *biop;
 157         sector_t bs_mask;
 158
 159         if (!q)
 160                 return -ENXIO;
 161
 162         if (bdev_read_only(bdev))
 163                 return -EPERM;
 164
 165         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 166         if ((sector | nr_sects) & bs_mask)
 167                 return -EINVAL;
 168
 169         if (!bdev_write_same(bdev))
 170                 return -EOPNOTSUPP;
 171
 172         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 173         max_write_same_sectors = bio_allowed_max_sectors(q);
 174
 175         while (nr_sects) {
 176                 bio = blk_next_bio(bio, bdev, 1, REQ_OP_WRITE_SAME, gfp_mask);
 177                 bio->bi_iter.bi_sector = sector;
 178                 bio->bi_vcnt = 1;
 179                 bio->bi_io_vec->bv_page = page;
 180                 bio->bi_io_vec->bv_offset = 0;
 181                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 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         if (bdev_read_only(bdev))
 241                 return -EPERM;
 242
 243         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 244         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 245
 246         if (max_write_zeroes_sectors == 0)
 247                 return -EOPNOTSUPP;
 248
 249         while (nr_sects) {
 250                 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
 251                 bio->bi_iter.bi_sector = sector;
 252                 if (flags & BLKDEV_ZERO_NOUNMAP)
 253                         bio->bi_opf |= REQ_NOUNMAP;
 254
 255                 if (nr_sects > max_write_zeroes_sectors) {
 256                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 257                         nr_sects -= max_write_zeroes_sectors;
 258                         sector += max_write_zeroes_sectors;
 259                 } else {
 260                         bio->bi_iter.bi_size = nr_sects << 9;
 261                         nr_sects = 0;
 262                 }
 263                 cond_resched();
 264         }
 265
 266         *biop = bio;
 267         return 0;
 268 }
 269
 270 /*
 271  * Convert a number of 512B sectors to a number of pages.
 272  * The result is limited to a number of pages that can fit into a BIO.
 273  * Also make sure that the result is always at least 1 (page) for the cases
 274  * where nr_sects is lower than the number of sectors in a page.
 275  */
 276 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 277 {
 278         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 279
 280         return min(pages, (sector_t)BIO_MAX_VECS);
 281 }
 282
 283 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 284                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 285                 struct bio **biop)
 286 {
 287         struct request_queue *q = bdev_get_queue(bdev);
 288         struct bio *bio = *biop;
 289         int bi_size = 0;
 290         unsigned int sz;
 291
 292         if (!q)
 293                 return -ENXIO;
 294
 295         if (bdev_read_only(bdev))
 296                 return -EPERM;
 297
 298         while (nr_sects != 0) {
 299                 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects),
 300                                    REQ_OP_WRITE, gfp_mask);
 301                 bio->bi_iter.bi_sector = sector;
 302
 303                 while (nr_sects != 0) {
 304                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 305                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 306                         nr_sects -= bi_size >> 9;
 307                         sector += bi_size >> 9;
 308                         if (bi_size < sz)
 309                                 break;
 310                 }
 311                 cond_resched();
 312         }
 313
 314         *biop = bio;
 315         return 0;
 316 }
 317
 318 /**
 319  * __blkdev_issue_zeroout - generate number of zero filed write bios
 320  * @bdev:       blockdev to issue
 321  * @sector:     start sector
 322  * @nr_sects:   number of sectors to write
 323  * @gfp_mask:   memory allocation flags (for bio_alloc)
 324  * @biop:       pointer to anchor bio
 325  * @flags:      controls detailed behavior
 326  *
 327  * Description:
 328  *  Zero-fill a block range, either using hardware offload or by explicitly
 329  *  writing zeroes to the device.
 330  *
 331  *  If a device is using logical block provisioning, the underlying space will
 332  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 333  *
 334  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 335  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 336  */
 337 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 338                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 339                 unsigned flags)
 340 {
 341         int ret;
 342         sector_t bs_mask;
 343
 344         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 345         if ((sector | nr_sects) & bs_mask)
 346                 return -EINVAL;
 347
 348         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 349                         biop, flags);
 350         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 351                 return ret;
 352
 353         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 354                                          biop);
 355 }
 356 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 357
 358 /**
 359  * blkdev_issue_zeroout - zero-fill a block range
 360  * @bdev:       blockdev to write
 361  * @sector:     start sector
 362  * @nr_sects:   number of sectors to write
 363  * @gfp_mask:   memory allocation flags (for bio_alloc)
 364  * @flags:      controls detailed behavior
 365  *
 366  * Description:
 367  *  Zero-fill a block range, either using hardware offload or by explicitly
 368  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 369  *  valid values for %flags.
 370  */
 371 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 372                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 373 {
 374         int ret = 0;
 375         sector_t bs_mask;
 376         struct bio *bio;
 377         struct blk_plug plug;
 378         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 379
 380         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 381         if ((sector | nr_sects) & bs_mask)
 382                 return -EINVAL;
 383
 384 retry:
 385         bio = NULL;
 386         blk_start_plug(&plug);
 387         if (try_write_zeroes) {
 388                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 389                                                   gfp_mask, &bio, flags);
 390         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 391                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 392                                                 gfp_mask, &bio);
 393         } else {
 394                 /* No zeroing offload support */
 395                 ret = -EOPNOTSUPP;
 396         }
 397         if (ret == 0 && bio) {
 398                 ret = submit_bio_wait(bio);
 399                 bio_put(bio);
 400         }
 401         blk_finish_plug(&plug);
 402         if (ret && try_write_zeroes) {
 403                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 404                         try_write_zeroes = false;
 405                         goto retry;
 406                 }
 407                 if (!bdev_write_zeroes_sectors(bdev)) {
 408                         /*
 409                          * Zeroing offload support was indicated, but the
 410                          * device reported ILLEGAL REQUEST (for some devices
 411                          * there is no non-destructive way to verify whether
 412                          * WRITE ZEROES is actually supported).
 413                          */
 414                         ret = -EOPNOTSUPP;
 415                 }
 416         }
 417
 418         return ret;
 419 }
 420 EXPORT_SYMBOL(blkdev_issue_zeroout);