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 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, 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 op;
  32         sector_t bs_mask, part_offset = 0;
  33
  34         if (!q)
  35                 return -ENXIO;
  36
  37         if (bdev_read_only(bdev))
  38                 return -EPERM;
  39
  40         if (flags & BLKDEV_DISCARD_SECURE) {
  41                 if (!blk_queue_secure_erase(q))
  42                         return -EOPNOTSUPP;
  43                 op = REQ_OP_SECURE_ERASE;
  44         } else {
  45                 if (!blk_queue_discard(q))
  46                         return -EOPNOTSUPP;
  47                 op = REQ_OP_DISCARD;
  48         }
  49
  50         /* In case the discard granularity isn't set by buggy device driver */
  51         if (WARN_ON_ONCE(!q->limits.discard_granularity)) {
  52                 char dev_name[BDEVNAME_SIZE];
  53
  54                 bdevname(bdev, dev_name);
  55                 pr_err_ratelimited("%s: Error: discard_granularity is 0.\n", dev_name);
  56                 return -EOPNOTSUPP;
  57         }
  58
  59         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  60         if ((sector | nr_sects) & bs_mask)
  61                 return -EINVAL;
  62
  63         if (!nr_sects)
  64                 return -EINVAL;
  65
  66         /* In case the discard request is in a partition */
  67         if (bdev_is_partition(bdev))
  68                 part_offset = bdev->bd_start_sect;
  69
  70         while (nr_sects) {
  71                 sector_t granularity_aligned_lba, req_sects;
  72                 sector_t sector_mapped = sector + part_offset;
  73
  74                 granularity_aligned_lba = round_up(sector_mapped,
  75                                 q->limits.discard_granularity >> SECTOR_SHIFT);
  76
  77                 /*
  78                  * Check whether the discard bio starts at a discard_granularity
  79                  * aligned LBA,
  80                  * - If no: set (granularity_aligned_lba - sector_mapped) to
  81                  *   bi_size of the first split bio, then the second bio will
  82                  *   start at a discard_granularity aligned LBA on the device.
  83                  * - If yes: use bio_aligned_discard_max_sectors() as the max
  84                  *   possible bi_size of the first split bio. Then when this bio
  85                  *   is split in device drive, the split ones are very probably
  86                  *   to be aligned to discard_granularity of the device's queue.
  87                  */
  88                 if (granularity_aligned_lba == sector_mapped)
  89                         req_sects = min_t(sector_t, nr_sects,
  90                                           bio_aligned_discard_max_sectors(q));
  91                 else
  92                         req_sects = min_t(sector_t, nr_sects,
  93                                           granularity_aligned_lba - sector_mapped);
  94
  95                 WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
  96
  97                 bio = blk_next_bio(bio, 0, gfp_mask);
  98                 bio->bi_iter.bi_sector = sector;
  99                 bio_set_dev(bio, bdev);
 100                 bio_set_op_attrs(bio, op, 0);
 101
 102                 bio->bi_iter.bi_size = req_sects << 9;
 103                 sector += req_sects;
 104                 nr_sects -= req_sects;
 105
 106                 /*
 107                  * We can loop for a long time in here, if someone does
 108                  * full device discards (like mkfs). Be nice and allow
 109                  * us to schedule out to avoid softlocking if preempt
 110                  * is disabled.
 111                  */
 112                 cond_resched();
 113         }
 114
 115         *biop = bio;
 116         return 0;
 117 }
 118 EXPORT_SYMBOL(__blkdev_issue_discard);
 119
 120 /**
 121  * blkdev_issue_discard - queue a discard
 122  * @bdev:       blockdev to issue discard for
 123  * @sector:     start sector
 124  * @nr_sects:   number of sectors to discard
 125  * @gfp_mask:   memory allocation flags (for bio_alloc)
 126  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 127  *
 128  * Description:
 129  *    Issue a discard request for the sectors in question.
 130  */
 131 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 132                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 133 {
 134         struct bio *bio = NULL;
 135         struct blk_plug plug;
 136         int ret;
 137
 138         blk_start_plug(&plug);
 139         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 140                         &bio);
 141         if (!ret && bio) {
 142                 ret = submit_bio_wait(bio);
 143                 if (ret == -EOPNOTSUPP)
 144                         ret = 0;
 145                 bio_put(bio);
 146         }
 147         blk_finish_plug(&plug);
 148
 149         return ret;
 150 }
 151 EXPORT_SYMBOL(blkdev_issue_discard);
 152
 153 /**
 154  * __blkdev_issue_write_same - generate number of bios with same page
 155  * @bdev:       target blockdev
 156  * @sector:     start sector
 157  * @nr_sects:   number of sectors to write
 158  * @gfp_mask:   memory allocation flags (for bio_alloc)
 159  * @page:       page containing data to write
 160  * @biop:       pointer to anchor bio
 161  *
 162  * Description:
 163  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 164  */
 165 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 166                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 167                 struct bio **biop)
 168 {
 169         struct request_queue *q = bdev_get_queue(bdev);
 170         unsigned int max_write_same_sectors;
 171         struct bio *bio = *biop;
 172         sector_t bs_mask;
 173
 174         if (!q)
 175                 return -ENXIO;
 176
 177         if (bdev_read_only(bdev))
 178                 return -EPERM;
 179
 180         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 181         if ((sector | nr_sects) & bs_mask)
 182                 return -EINVAL;
 183
 184         if (!bdev_write_same(bdev))
 185                 return -EOPNOTSUPP;
 186
 187         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 188         max_write_same_sectors = bio_allowed_max_sectors(q);
 189
 190         while (nr_sects) {
 191                 bio = blk_next_bio(bio, 1, gfp_mask);
 192                 bio->bi_iter.bi_sector = sector;
 193                 bio_set_dev(bio, bdev);
 194                 bio->bi_vcnt = 1;
 195                 bio->bi_io_vec->bv_page = page;
 196                 bio->bi_io_vec->bv_offset = 0;
 197                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 198                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 199
 200                 if (nr_sects > max_write_same_sectors) {
 201                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 202                         nr_sects -= max_write_same_sectors;
 203                         sector += max_write_same_sectors;
 204                 } else {
 205                         bio->bi_iter.bi_size = nr_sects << 9;
 206                         nr_sects = 0;
 207                 }
 208                 cond_resched();
 209         }
 210
 211         *biop = bio;
 212         return 0;
 213 }
 214
 215 /**
 216  * blkdev_issue_write_same - queue a write same operation
 217  * @bdev:       target blockdev
 218  * @sector:     start sector
 219  * @nr_sects:   number of sectors to write
 220  * @gfp_mask:   memory allocation flags (for bio_alloc)
 221  * @page:       page containing data
 222  *
 223  * Description:
 224  *    Issue a write same request for the sectors in question.
 225  */
 226 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 227                                 sector_t nr_sects, gfp_t gfp_mask,
 228                                 struct page *page)
 229 {
 230         struct bio *bio = NULL;
 231         struct blk_plug plug;
 232         int ret;
 233
 234         blk_start_plug(&plug);
 235         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 236                         &bio);
 237         if (ret == 0 && bio) {
 238                 ret = submit_bio_wait(bio);
 239                 bio_put(bio);
 240         }
 241         blk_finish_plug(&plug);
 242         return ret;
 243 }
 244 EXPORT_SYMBOL(blkdev_issue_write_same);
 245
 246 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 247                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 248                 struct bio **biop, unsigned flags)
 249 {
 250         struct bio *bio = *biop;
 251         unsigned int max_write_zeroes_sectors;
 252         struct request_queue *q = bdev_get_queue(bdev);
 253
 254         if (!q)
 255                 return -ENXIO;
 256
 257         if (bdev_read_only(bdev))
 258                 return -EPERM;
 259
 260         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 261         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 262
 263         if (max_write_zeroes_sectors == 0)
 264                 return -EOPNOTSUPP;
 265
 266         while (nr_sects) {
 267                 bio = blk_next_bio(bio, 0, gfp_mask);
 268                 bio->bi_iter.bi_sector = sector;
 269                 bio_set_dev(bio, bdev);
 270                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 271                 if (flags & BLKDEV_ZERO_NOUNMAP)
 272                         bio->bi_opf |= REQ_NOUNMAP;
 273
 274                 if (nr_sects > max_write_zeroes_sectors) {
 275                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 276                         nr_sects -= max_write_zeroes_sectors;
 277                         sector += max_write_zeroes_sectors;
 278                 } else {
 279                         bio->bi_iter.bi_size = nr_sects << 9;
 280                         nr_sects = 0;
 281                 }
 282                 cond_resched();
 283         }
 284
 285         *biop = bio;
 286         return 0;
 287 }
 288
 289 /*
 290  * Convert a number of 512B sectors to a number of pages.
 291  * The result is limited to a number of pages that can fit into a BIO.
 292  * Also make sure that the result is always at least 1 (page) for the cases
 293  * where nr_sects is lower than the number of sectors in a page.
 294  */
 295 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 296 {
 297         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 298
 299         return min(pages, (sector_t)BIO_MAX_VECS);
 300 }
 301
 302 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 303                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 304                 struct bio **biop)
 305 {
 306         struct request_queue *q = bdev_get_queue(bdev);
 307         struct bio *bio = *biop;
 308         int bi_size = 0;
 309         unsigned int sz;
 310
 311         if (!q)
 312                 return -ENXIO;
 313
 314         if (bdev_read_only(bdev))
 315                 return -EPERM;
 316
 317         while (nr_sects != 0) {
 318                 bio = blk_next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
 319                                    gfp_mask);
 320                 bio->bi_iter.bi_sector = sector;
 321                 bio_set_dev(bio, bdev);
 322                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 323
 324                 while (nr_sects != 0) {
 325                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 326                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 327                         nr_sects -= bi_size >> 9;
 328                         sector += bi_size >> 9;
 329                         if (bi_size < sz)
 330                                 break;
 331                 }
 332                 cond_resched();
 333         }
 334
 335         *biop = bio;
 336         return 0;
 337 }
 338
 339 /**
 340  * __blkdev_issue_zeroout - generate number of zero filed write bios
 341  * @bdev:       blockdev to issue
 342  * @sector:     start sector
 343  * @nr_sects:   number of sectors to write
 344  * @gfp_mask:   memory allocation flags (for bio_alloc)
 345  * @biop:       pointer to anchor bio
 346  * @flags:      controls detailed behavior
 347  *
 348  * Description:
 349  *  Zero-fill a block range, either using hardware offload or by explicitly
 350  *  writing zeroes to the device.
 351  *
 352  *  If a device is using logical block provisioning, the underlying space will
 353  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 354  *
 355  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 356  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 357  */
 358 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 359                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 360                 unsigned flags)
 361 {
 362         int ret;
 363         sector_t bs_mask;
 364
 365         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 366         if ((sector | nr_sects) & bs_mask)
 367                 return -EINVAL;
 368
 369         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 370                         biop, flags);
 371         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 372                 return ret;
 373
 374         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 375                                          biop);
 376 }
 377 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 378
 379 /**
 380  * blkdev_issue_zeroout - zero-fill a block range
 381  * @bdev:       blockdev to write
 382  * @sector:     start sector
 383  * @nr_sects:   number of sectors to write
 384  * @gfp_mask:   memory allocation flags (for bio_alloc)
 385  * @flags:      controls detailed behavior
 386  *
 387  * Description:
 388  *  Zero-fill a block range, either using hardware offload or by explicitly
 389  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 390  *  valid values for %flags.
 391  */
 392 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 393                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 394 {
 395         int ret = 0;
 396         sector_t bs_mask;
 397         struct bio *bio;
 398         struct blk_plug plug;
 399         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 400
 401         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 402         if ((sector | nr_sects) & bs_mask)
 403                 return -EINVAL;
 404
 405 retry:
 406         bio = NULL;
 407         blk_start_plug(&plug);
 408         if (try_write_zeroes) {
 409                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 410                                                   gfp_mask, &bio, flags);
 411         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 412                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 413                                                 gfp_mask, &bio);
 414         } else {
 415                 /* No zeroing offload support */
 416                 ret = -EOPNOTSUPP;
 417         }
 418         if (ret == 0 && bio) {
 419                 ret = submit_bio_wait(bio);
 420                 bio_put(bio);
 421         }
 422         blk_finish_plug(&plug);
 423         if (ret && try_write_zeroes) {
 424                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 425                         try_write_zeroes = false;
 426                         goto retry;
 427                 }
 428                 if (!bdev_write_zeroes_sectors(bdev)) {
 429                         /*
 430                          * Zeroing offload support was indicated, but the
 431                          * device reported ILLEGAL REQUEST (for some devices
 432                          * there is no non-destructive way to verify whether
 433                          * WRITE ZEROES is actually supported).
 434                          */
 435                         ret = -EOPNOTSUPP;
 436                 }
 437         }
 438
 439         return ret;
 440 }
 441 EXPORT_SYMBOL(blkdev_issue_zeroout);