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 struct bio *next_bio(struct bio *bio, unsigned int nr_pages,
  14                 gfp_t gfp)
  15 {
  16         struct bio *new = bio_alloc(gfp, nr_pages);
  17
  18         if (bio) {
  19                 bio_chain(bio, new);
  20                 submit_bio(bio);
  21         }
  22
  23         return new;
  24 }
  25
  26 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  27                 sector_t nr_sects, gfp_t gfp_mask, int flags,
  28                 struct bio **biop)
  29 {
  30         struct request_queue *q = bdev_get_queue(bdev);
  31         struct bio *bio = *biop;
  32         unsigned int granularity;
  33         unsigned int op;
  34         int alignment;
  35         sector_t bs_mask;
  36
  37         if (!q)
  38                 return -ENXIO;
  39
  40         if (bdev_read_only(bdev))
  41                 return -EPERM;
  42
  43         if (flags & BLKDEV_DISCARD_SECURE) {
  44                 if (!blk_queue_secure_erase(q))
  45                         return -EOPNOTSUPP;
  46                 op = REQ_OP_SECURE_ERASE;
  47         } else {
  48                 if (!blk_queue_discard(q))
  49                         return -EOPNOTSUPP;
  50                 op = REQ_OP_DISCARD;
  51         }
  52
  53         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  54         if ((sector | nr_sects) & bs_mask)
  55                 return -EINVAL;
  56
  57         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  58         granularity = max(q->limits.discard_granularity >> 9, 1U);
  59         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  60
  61         while (nr_sects) {
  62                 unsigned int req_sects;
  63                 sector_t end_sect, tmp;
  64
  65                 /*
  66                  * Issue in chunks of the user defined max discard setting,
  67                  * ensuring that bi_size doesn't overflow
  68                  */
  69                 req_sects = min_t(sector_t, nr_sects,
  70                                         q->limits.max_discard_sectors);
  71                 if (req_sects > UINT_MAX >> 9)
  72                         req_sects = UINT_MAX >> 9;
  73
  74                 /*
  75                  * If splitting a request, and the next starting sector would be
  76                  * misaligned, stop the discard at the previous aligned sector.
  77                  */
  78                 end_sect = sector + req_sects;
  79                 tmp = end_sect;
  80                 if (req_sects < nr_sects &&
  81                     sector_div(tmp, granularity) != alignment) {
  82                         end_sect = end_sect - alignment;
  83                         sector_div(end_sect, granularity);
  84                         end_sect = end_sect * granularity + alignment;
  85                         req_sects = end_sect - sector;
  86                 }
  87
  88                 bio = next_bio(bio, 0, gfp_mask);
  89                 bio->bi_iter.bi_sector = sector;
  90                 bio_set_dev(bio, bdev);
  91                 bio_set_op_attrs(bio, op, 0);
  92
  93                 bio->bi_iter.bi_size = req_sects << 9;
  94                 nr_sects -= req_sects;
  95                 sector = end_sect;
  96
  97                 /*
  98                  * We can loop for a long time in here, if someone does
  99                  * full device discards (like mkfs). Be nice and allow
 100                  * us to schedule out to avoid softlocking if preempt
 101                  * is disabled.
 102                  */
 103                 cond_resched();
 104         }
 105
 106         *biop = bio;
 107         return 0;
 108 }
 109 EXPORT_SYMBOL(__blkdev_issue_discard);
 110
 111 /**
 112  * blkdev_issue_discard - queue a discard
 113  * @bdev:       blockdev to issue discard for
 114  * @sector:     start sector
 115  * @nr_sects:   number of sectors to discard
 116  * @gfp_mask:   memory allocation flags (for bio_alloc)
 117  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 118  *
 119  * Description:
 120  *    Issue a discard request for the sectors in question.
 121  */
 122 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 123                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 124 {
 125         struct bio *bio = NULL;
 126         struct blk_plug plug;
 127         int ret;
 128
 129         blk_start_plug(&plug);
 130         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 131                         &bio);
 132         if (!ret && bio) {
 133                 ret = submit_bio_wait(bio);
 134                 if (ret == -EOPNOTSUPP)
 135                         ret = 0;
 136                 bio_put(bio);
 137         }
 138         blk_finish_plug(&plug);
 139
 140         return ret;
 141 }
 142 EXPORT_SYMBOL(blkdev_issue_discard);
 143
 144 /**
 145  * __blkdev_issue_write_same - generate number of bios with same page
 146  * @bdev:       target blockdev
 147  * @sector:     start sector
 148  * @nr_sects:   number of sectors to write
 149  * @gfp_mask:   memory allocation flags (for bio_alloc)
 150  * @page:       page containing data to write
 151  * @biop:       pointer to anchor bio
 152  *
 153  * Description:
 154  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 155  */
 156 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 157                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 158                 struct bio **biop)
 159 {
 160         struct request_queue *q = bdev_get_queue(bdev);
 161         unsigned int max_write_same_sectors;
 162         struct bio *bio = *biop;
 163         sector_t bs_mask;
 164
 165         if (!q)
 166                 return -ENXIO;
 167
 168         if (bdev_read_only(bdev))
 169                 return -EPERM;
 170
 171         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 172         if ((sector | nr_sects) & bs_mask)
 173                 return -EINVAL;
 174
 175         if (!bdev_write_same(bdev))
 176                 return -EOPNOTSUPP;
 177
 178         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 179         max_write_same_sectors = UINT_MAX >> 9;
 180
 181         while (nr_sects) {
 182                 bio = next_bio(bio, 1, gfp_mask);
 183                 bio->bi_iter.bi_sector = sector;
 184                 bio_set_dev(bio, bdev);
 185                 bio->bi_vcnt = 1;
 186                 bio->bi_io_vec->bv_page = page;
 187                 bio->bi_io_vec->bv_offset = 0;
 188                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 189                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 190
 191                 if (nr_sects > max_write_same_sectors) {
 192                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 193                         nr_sects -= max_write_same_sectors;
 194                         sector += max_write_same_sectors;
 195                 } else {
 196                         bio->bi_iter.bi_size = nr_sects << 9;
 197                         nr_sects = 0;
 198                 }
 199                 cond_resched();
 200         }
 201
 202         *biop = bio;
 203         return 0;
 204 }
 205
 206 /**
 207  * blkdev_issue_write_same - queue a write same operation
 208  * @bdev:       target blockdev
 209  * @sector:     start sector
 210  * @nr_sects:   number of sectors to write
 211  * @gfp_mask:   memory allocation flags (for bio_alloc)
 212  * @page:       page containing data
 213  *
 214  * Description:
 215  *    Issue a write same request for the sectors in question.
 216  */
 217 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 218                                 sector_t nr_sects, gfp_t gfp_mask,
 219                                 struct page *page)
 220 {
 221         struct bio *bio = NULL;
 222         struct blk_plug plug;
 223         int ret;
 224
 225         blk_start_plug(&plug);
 226         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 227                         &bio);
 228         if (ret == 0 && bio) {
 229                 ret = submit_bio_wait(bio);
 230                 bio_put(bio);
 231         }
 232         blk_finish_plug(&plug);
 233         return ret;
 234 }
 235 EXPORT_SYMBOL(blkdev_issue_write_same);
 236
 237 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 238                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 239                 struct bio **biop, unsigned flags)
 240 {
 241         struct bio *bio = *biop;
 242         unsigned int max_write_zeroes_sectors;
 243         struct request_queue *q = bdev_get_queue(bdev);
 244
 245         if (!q)
 246                 return -ENXIO;
 247
 248         if (bdev_read_only(bdev))
 249                 return -EPERM;
 250
 251         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 252         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 253
 254         if (max_write_zeroes_sectors == 0)
 255                 return -EOPNOTSUPP;
 256
 257         while (nr_sects) {
 258                 bio = next_bio(bio, 0, gfp_mask);
 259                 bio->bi_iter.bi_sector = sector;
 260                 bio_set_dev(bio, bdev);
 261                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 262                 if (flags & BLKDEV_ZERO_NOUNMAP)
 263                         bio->bi_opf |= REQ_NOUNMAP;
 264
 265                 if (nr_sects > max_write_zeroes_sectors) {
 266                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 267                         nr_sects -= max_write_zeroes_sectors;
 268                         sector += max_write_zeroes_sectors;
 269                 } else {
 270                         bio->bi_iter.bi_size = nr_sects << 9;
 271                         nr_sects = 0;
 272                 }
 273                 cond_resched();
 274         }
 275
 276         *biop = bio;
 277         return 0;
 278 }
 279
 280 /*
 281  * Convert a number of 512B sectors to a number of pages.
 282  * The result is limited to a number of pages that can fit into a BIO.
 283  * Also make sure that the result is always at least 1 (page) for the cases
 284  * where nr_sects is lower than the number of sectors in a page.
 285  */
 286 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 287 {
 288         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 289
 290         return min(pages, (sector_t)BIO_MAX_PAGES);
 291 }
 292
 293 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 294                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 295                 struct bio **biop)
 296 {
 297         struct request_queue *q = bdev_get_queue(bdev);
 298         struct bio *bio = *biop;
 299         int bi_size = 0;
 300         unsigned int sz;
 301
 302         if (!q)
 303                 return -ENXIO;
 304
 305         if (bdev_read_only(bdev))
 306                 return -EPERM;
 307
 308         while (nr_sects != 0) {
 309                 bio = next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
 310                                gfp_mask);
 311                 bio->bi_iter.bi_sector = sector;
 312                 bio_set_dev(bio, bdev);
 313                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 314
 315                 while (nr_sects != 0) {
 316                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 317                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 318                         nr_sects -= bi_size >> 9;
 319                         sector += bi_size >> 9;
 320                         if (bi_size < sz)
 321                                 break;
 322                 }
 323                 cond_resched();
 324         }
 325
 326         *biop = bio;
 327         return 0;
 328 }
 329
 330 /**
 331  * __blkdev_issue_zeroout - generate number of zero filed write bios
 332  * @bdev:       blockdev to issue
 333  * @sector:     start sector
 334  * @nr_sects:   number of sectors to write
 335  * @gfp_mask:   memory allocation flags (for bio_alloc)
 336  * @biop:       pointer to anchor bio
 337  * @flags:      controls detailed behavior
 338  *
 339  * Description:
 340  *  Zero-fill a block range, either using hardware offload or by explicitly
 341  *  writing zeroes to the device.
 342  *
 343  *  If a device is using logical block provisioning, the underlying space will
 344  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 345  *
 346  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 347  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 348  */
 349 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 350                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 351                 unsigned flags)
 352 {
 353         int ret;
 354         sector_t bs_mask;
 355
 356         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 357         if ((sector | nr_sects) & bs_mask)
 358                 return -EINVAL;
 359
 360         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 361                         biop, flags);
 362         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 363                 return ret;
 364
 365         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 366                                          biop);
 367 }
 368 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 369
 370 /**
 371  * blkdev_issue_zeroout - zero-fill a block range
 372  * @bdev:       blockdev to write
 373  * @sector:     start sector
 374  * @nr_sects:   number of sectors to write
 375  * @gfp_mask:   memory allocation flags (for bio_alloc)
 376  * @flags:      controls detailed behavior
 377  *
 378  * Description:
 379  *  Zero-fill a block range, either using hardware offload or by explicitly
 380  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 381  *  valid values for %flags.
 382  */
 383 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 384                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 385 {
 386         int ret = 0;
 387         sector_t bs_mask;
 388         struct bio *bio;
 389         struct blk_plug plug;
 390         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 391
 392         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 393         if ((sector | nr_sects) & bs_mask)
 394                 return -EINVAL;
 395
 396 retry:
 397         bio = NULL;
 398         blk_start_plug(&plug);
 399         if (try_write_zeroes) {
 400                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 401                                                   gfp_mask, &bio, flags);
 402         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 403                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 404                                                 gfp_mask, &bio);
 405         } else {
 406                 /* No zeroing offload support */
 407                 ret = -EOPNOTSUPP;
 408         }
 409         if (ret == 0 && bio) {
 410                 ret = submit_bio_wait(bio);
 411                 bio_put(bio);
 412         }
 413         blk_finish_plug(&plug);
 414         if (ret && try_write_zeroes) {
 415                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 416                         try_write_zeroes = false;
 417                         goto retry;
 418                 }
 419                 if (!bdev_write_zeroes_sectors(bdev)) {
 420                         /*
 421                          * Zeroing offload support was indicated, but the
 422                          * device reported ILLEGAL REQUEST (for some devices
 423                          * there is no non-destructive way to verify whether
 424                          * WRITE ZEROES is actually supported).
 425                          */
 426                         ret = -EOPNOTSUPP;
 427                 }
 428         }
 429
 430         return ret;
 431 }
 432 EXPORT_SYMBOL(blkdev_issue_zeroout);