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 static sector_t bio_discard_limit(struct block_device *bdev, sector_t sector)
  14 {
  15         unsigned int discard_granularity = bdev_discard_granularity(bdev);
  16         sector_t granularity_aligned_sector;
  17
  18         if (bdev_is_partition(bdev))
  19                 sector += bdev->bd_start_sect;
  20
  21         granularity_aligned_sector =
  22                 round_up(sector, discard_granularity >> SECTOR_SHIFT);
  23
  24         /*
  25          * Make sure subsequent bios start aligned to the discard granularity if
  26          * it needs to be split.
  27          */
  28         if (granularity_aligned_sector != sector)
  29                 return granularity_aligned_sector - sector;
  30
  31         /*
  32          * Align the bio size to the discard granularity to make splitting the bio
  33          * at discard granularity boundaries easier in the driver if needed.
  34          */
  35         return round_down(UINT_MAX, discard_granularity) >> SECTOR_SHIFT;
  36 }
  37
  38 struct bio *blk_alloc_discard_bio(struct block_device *bdev,
  39                 sector_t *sector, sector_t *nr_sects, gfp_t gfp_mask)
  40 {
  41         sector_t bio_sects = min(*nr_sects, bio_discard_limit(bdev, *sector));
  42         struct bio *bio;
  43
  44         if (!bio_sects)
  45                 return NULL;
  46
  47         bio = bio_alloc(bdev, 0, REQ_OP_DISCARD, gfp_mask);
  48         if (!bio)
  49                 return NULL;
  50         bio->bi_iter.bi_sector = *sector;
  51         bio->bi_iter.bi_size = bio_sects << SECTOR_SHIFT;
  52         *sector += bio_sects;
  53         *nr_sects -= bio_sects;
  54         /*
  55          * We can loop for a long time in here if someone does full device
  56          * discards (like mkfs).  Be nice and allow us to schedule out to avoid
  57          * softlocking if preempt is disabled.
  58          */
  59         cond_resched();
  60         return bio;
  61 }
  62
  63 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  64                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop)
  65 {
  66         struct bio *bio;
  67
  68         while ((bio = blk_alloc_discard_bio(bdev, &sector, &nr_sects,
  69                         gfp_mask)))
  70                 *biop = bio_chain_and_submit(*biop, bio);
  71         return 0;
  72 }
  73 EXPORT_SYMBOL(__blkdev_issue_discard);
  74
  75 /**
  76  * blkdev_issue_discard - queue a discard
  77  * @bdev:       blockdev to issue discard for
  78  * @sector:     start sector
  79  * @nr_sects:   number of sectors to discard
  80  * @gfp_mask:   memory allocation flags (for bio_alloc)
  81  *
  82  * Description:
  83  *    Issue a discard request for the sectors in question.
  84  */
  85 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  86                 sector_t nr_sects, gfp_t gfp_mask)
  87 {
  88         struct bio *bio = NULL;
  89         struct blk_plug plug;
  90         int ret;
  91
  92         blk_start_plug(&plug);
  93         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, &bio);
  94         if (!ret && bio) {
  95                 ret = submit_bio_wait(bio);
  96                 if (ret == -EOPNOTSUPP)
  97                         ret = 0;
  98                 bio_put(bio);
  99         }
 100         blk_finish_plug(&plug);
 101
 102         return ret;
 103 }
 104 EXPORT_SYMBOL(blkdev_issue_discard);
 105
 106 static sector_t bio_write_zeroes_limit(struct block_device *bdev)
 107 {
 108         sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 109
 110         return min(bdev_write_zeroes_sectors(bdev),
 111                 (UINT_MAX >> SECTOR_SHIFT) & ~bs_mask);
 112 }
 113
 114 /*
 115  * There is no reliable way for the SCSI subsystem to determine whether a
 116  * device supports a WRITE SAME operation without actually performing a write
 117  * to media. As a result, write_zeroes is enabled by default and will be
 118  * disabled if a zeroing operation subsequently fails. This means that this
 119  * queue limit is likely to change at runtime.
 120  */
 121 static void __blkdev_issue_write_zeroes(struct block_device *bdev,
 122                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 123                 struct bio **biop, unsigned flags, sector_t limit)
 124 {
 125
 126         while (nr_sects) {
 127                 unsigned int len = min(nr_sects, limit);
 128                 struct bio *bio;
 129
 130                 if ((flags & BLKDEV_ZERO_KILLABLE) &&
 131                     fatal_signal_pending(current))
 132                         break;
 133
 134                 bio = bio_alloc(bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask);
 135                 bio->bi_iter.bi_sector = sector;
 136                 if (flags & BLKDEV_ZERO_NOUNMAP)
 137                         bio->bi_opf |= REQ_NOUNMAP;
 138
 139                 bio->bi_iter.bi_size = len << SECTOR_SHIFT;
 140                 *biop = bio_chain_and_submit(*biop, bio);
 141
 142                 nr_sects -= len;
 143                 sector += len;
 144                 cond_resched();
 145         }
 146 }
 147
 148 static int blkdev_issue_write_zeroes(struct block_device *bdev, sector_t sector,
 149                 sector_t nr_sects, gfp_t gfp, unsigned flags)
 150 {
 151         sector_t limit = bio_write_zeroes_limit(bdev);
 152         struct bio *bio = NULL;
 153         struct blk_plug plug;
 154         int ret = 0;
 155
 156         blk_start_plug(&plug);
 157         __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp, &bio,
 158                         flags, limit);
 159         if (bio) {
 160                 if ((flags & BLKDEV_ZERO_KILLABLE) &&
 161                     fatal_signal_pending(current)) {
 162                         bio_await_chain(bio);
 163                         blk_finish_plug(&plug);
 164                         return -EINTR;
 165                 }
 166                 ret = submit_bio_wait(bio);
 167                 bio_put(bio);
 168         }
 169         blk_finish_plug(&plug);
 170
 171         /*
 172          * For some devices there is no non-destructive way to verify whether
 173          * WRITE ZEROES is actually supported.  These will clear the capability
 174          * on an I/O error, in which case we'll turn any error into
 175          * "not supported" here.
 176          */
 177         if (ret && !bdev_write_zeroes_sectors(bdev))
 178                 return -EOPNOTSUPP;
 179         return ret;
 180 }
 181
 182 /*
 183  * Convert a number of 512B sectors to a number of pages.
 184  * The result is limited to a number of pages that can fit into a BIO.
 185  * Also make sure that the result is always at least 1 (page) for the cases
 186  * where nr_sects is lower than the number of sectors in a page.
 187  */
 188 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 189 {
 190         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 191
 192         return min(pages, (sector_t)BIO_MAX_VECS);
 193 }
 194
 195 static void __blkdev_issue_zero_pages(struct block_device *bdev,
 196                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 197                 struct bio **biop, unsigned int flags)
 198 {
 199         while (nr_sects) {
 200                 unsigned int nr_vecs = __blkdev_sectors_to_bio_pages(nr_sects);
 201                 struct bio *bio;
 202
 203                 bio = bio_alloc(bdev, nr_vecs, REQ_OP_WRITE, gfp_mask);
 204                 bio->bi_iter.bi_sector = sector;
 205
 206                 if ((flags & BLKDEV_ZERO_KILLABLE) &&
 207                     fatal_signal_pending(current))
 208                         break;
 209
 210                 do {
 211                         unsigned int len, added;
 212
 213                         len = min_t(sector_t,
 214                                 PAGE_SIZE, nr_sects << SECTOR_SHIFT);
 215                         added = bio_add_page(bio, ZERO_PAGE(0), len, 0);
 216                         if (added < len)
 217                                 break;
 218                         nr_sects -= added >> SECTOR_SHIFT;
 219                         sector += added >> SECTOR_SHIFT;
 220                 } while (nr_sects);
 221
 222                 *biop = bio_chain_and_submit(*biop, bio);
 223                 cond_resched();
 224         }
 225 }
 226
 227 static int blkdev_issue_zero_pages(struct block_device *bdev, sector_t sector,
 228                 sector_t nr_sects, gfp_t gfp, unsigned flags)
 229 {
 230         struct bio *bio = NULL;
 231         struct blk_plug plug;
 232         int ret = 0;
 233
 234         if (flags & BLKDEV_ZERO_NOFALLBACK)
 235                 return -EOPNOTSUPP;
 236
 237         blk_start_plug(&plug);
 238         __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp, &bio, flags);
 239         if (bio) {
 240                 if ((flags & BLKDEV_ZERO_KILLABLE) &&
 241                     fatal_signal_pending(current)) {
 242                         bio_await_chain(bio);
 243                         blk_finish_plug(&plug);
 244                         return -EINTR;
 245                 }
 246                 ret = submit_bio_wait(bio);
 247                 bio_put(bio);
 248         }
 249         blk_finish_plug(&plug);
 250
 251         return ret;
 252 }
 253
 254 /**
 255  * __blkdev_issue_zeroout - generate number of zero filed write bios
 256  * @bdev:       blockdev to issue
 257  * @sector:     start sector
 258  * @nr_sects:   number of sectors to write
 259  * @gfp_mask:   memory allocation flags (for bio_alloc)
 260  * @biop:       pointer to anchor bio
 261  * @flags:      controls detailed behavior
 262  *
 263  * Description:
 264  *  Zero-fill a block range, either using hardware offload or by explicitly
 265  *  writing zeroes to the device.
 266  *
 267  *  If a device is using logical block provisioning, the underlying space will
 268  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 269  *
 270  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 271  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 272  */
 273 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 274                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 275                 unsigned flags)
 276 {
 277         sector_t limit = bio_write_zeroes_limit(bdev);
 278
 279         if (bdev_read_only(bdev))
 280                 return -EPERM;
 281
 282         if (limit) {
 283                 __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 284                                 gfp_mask, biop, flags, limit);
 285         } else {
 286                 if (flags & BLKDEV_ZERO_NOFALLBACK)
 287                         return -EOPNOTSUPP;
 288                 __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 289                                 biop, flags);
 290         }
 291         return 0;
 292 }
 293 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 294
 295 /**
 296  * blkdev_issue_zeroout - zero-fill a block range
 297  * @bdev:       blockdev to write
 298  * @sector:     start sector
 299  * @nr_sects:   number of sectors to write
 300  * @gfp_mask:   memory allocation flags (for bio_alloc)
 301  * @flags:      controls detailed behavior
 302  *
 303  * Description:
 304  *  Zero-fill a block range, either using hardware offload or by explicitly
 305  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 306  *  valid values for %flags.
 307  */
 308 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 309                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 310 {
 311         int ret;
 312
 313         if ((sector | nr_sects) & ((bdev_logical_block_size(bdev) >> 9) - 1))
 314                 return -EINVAL;
 315         if (bdev_read_only(bdev))
 316                 return -EPERM;
 317
 318         if (bdev_write_zeroes_sectors(bdev)) {
 319                 ret = blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 320                                 gfp_mask, flags);
 321                 if (ret != -EOPNOTSUPP)
 322                         return ret;
 323         }
 324
 325         return blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask, flags);
 326 }
 327 EXPORT_SYMBOL(blkdev_issue_zeroout);
 328
 329 int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector,
 330                 sector_t nr_sects, gfp_t gfp)
 331 {
 332         sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 333         unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev);
 334         struct bio *bio = NULL;
 335         struct blk_plug plug;
 336         int ret = 0;
 337
 338         /* make sure that "len << SECTOR_SHIFT" doesn't overflow */
 339         if (max_sectors > UINT_MAX >> SECTOR_SHIFT)
 340                 max_sectors = UINT_MAX >> SECTOR_SHIFT;
 341         max_sectors &= ~bs_mask;
 342
 343         if (max_sectors == 0)
 344                 return -EOPNOTSUPP;
 345         if ((sector | nr_sects) & bs_mask)
 346                 return -EINVAL;
 347         if (bdev_read_only(bdev))
 348                 return -EPERM;
 349
 350         blk_start_plug(&plug);
 351         while (nr_sects) {
 352                 unsigned int len = min_t(sector_t, nr_sects, max_sectors);
 353
 354                 bio = blk_next_bio(bio, bdev, 0, REQ_OP_SECURE_ERASE, gfp);
 355                 bio->bi_iter.bi_sector = sector;
 356                 bio->bi_iter.bi_size = len << SECTOR_SHIFT;
 357
 358                 sector += len;
 359                 nr_sects -= len;
 360                 cond_resched();
 361         }
 362         if (bio) {
 363                 ret = submit_bio_wait(bio);
 364                 bio_put(bio);
 365         }
 366         blk_finish_plug(&plug);
 367
 368         return ret;
 369 }
 370 EXPORT_SYMBOL(blkdev_issue_secure_erase);