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                 /* Make sure bi_size doesn't overflow */
  66                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  67
  68                 /**
  69                  * If splitting a request, and the next starting sector would be
  70                  * misaligned, stop the discard at the previous aligned sector.
  71                  */
  72                 end_sect = sector + req_sects;
  73                 tmp = end_sect;
  74                 if (req_sects < nr_sects &&
  75                     sector_div(tmp, granularity) != alignment) {
  76                         end_sect = end_sect - alignment;
  77                         sector_div(end_sect, granularity);
  78                         end_sect = end_sect * granularity + alignment;
  79                         req_sects = end_sect - sector;
  80                 }
  81
  82                 bio = next_bio(bio, 0, gfp_mask);
  83                 bio->bi_iter.bi_sector = sector;
  84                 bio_set_dev(bio, bdev);
  85                 bio_set_op_attrs(bio, op, 0);
  86
  87                 bio->bi_iter.bi_size = req_sects << 9;
  88                 nr_sects -= req_sects;
  89                 sector = end_sect;
  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 = UINT_MAX >> 9;
 174
 175         while (nr_sects) {
 176                 bio = next_bio(bio, 1, gfp_mask);
 177                 bio->bi_iter.bi_sector = sector;
 178                 bio_set_dev(bio, bdev);
 179                 bio->bi_vcnt = 1;
 180                 bio->bi_io_vec->bv_page = page;
 181                 bio->bi_io_vec->bv_offset = 0;
 182                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 183                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 184
 185                 if (nr_sects > max_write_same_sectors) {
 186                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 187                         nr_sects -= max_write_same_sectors;
 188                         sector += max_write_same_sectors;
 189                 } else {
 190                         bio->bi_iter.bi_size = nr_sects << 9;
 191                         nr_sects = 0;
 192                 }
 193                 cond_resched();
 194         }
 195
 196         *biop = bio;
 197         return 0;
 198 }
 199
 200 /**
 201  * blkdev_issue_write_same - queue a write same operation
 202  * @bdev:       target blockdev
 203  * @sector:     start sector
 204  * @nr_sects:   number of sectors to write
 205  * @gfp_mask:   memory allocation flags (for bio_alloc)
 206  * @page:       page containing data
 207  *
 208  * Description:
 209  *    Issue a write same request for the sectors in question.
 210  */
 211 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 212                                 sector_t nr_sects, gfp_t gfp_mask,
 213                                 struct page *page)
 214 {
 215         struct bio *bio = NULL;
 216         struct blk_plug plug;
 217         int ret;
 218
 219         blk_start_plug(&plug);
 220         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 221                         &bio);
 222         if (ret == 0 && bio) {
 223                 ret = submit_bio_wait(bio);
 224                 bio_put(bio);
 225         }
 226         blk_finish_plug(&plug);
 227         return ret;
 228 }
 229 EXPORT_SYMBOL(blkdev_issue_write_same);
 230
 231 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 232                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 233                 struct bio **biop, unsigned flags)
 234 {
 235         struct bio *bio = *biop;
 236         unsigned int max_write_zeroes_sectors;
 237         struct request_queue *q = bdev_get_queue(bdev);
 238
 239         if (!q)
 240                 return -ENXIO;
 241
 242         if (bdev_read_only(bdev))
 243                 return -EPERM;
 244
 245         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 246         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 247
 248         if (max_write_zeroes_sectors == 0)
 249                 return -EOPNOTSUPP;
 250
 251         while (nr_sects) {
 252                 bio = next_bio(bio, 0, gfp_mask);
 253                 bio->bi_iter.bi_sector = sector;
 254                 bio_set_dev(bio, bdev);
 255                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 256                 if (flags & BLKDEV_ZERO_NOUNMAP)
 257                         bio->bi_opf |= REQ_NOUNMAP;
 258
 259                 if (nr_sects > max_write_zeroes_sectors) {
 260                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 261                         nr_sects -= max_write_zeroes_sectors;
 262                         sector += max_write_zeroes_sectors;
 263                 } else {
 264                         bio->bi_iter.bi_size = nr_sects << 9;
 265                         nr_sects = 0;
 266                 }
 267                 cond_resched();
 268         }
 269
 270         *biop = bio;
 271         return 0;
 272 }
 273
 274 /*
 275  * Convert a number of 512B sectors to a number of pages.
 276  * The result is limited to a number of pages that can fit into a BIO.
 277  * Also make sure that the result is always at least 1 (page) for the cases
 278  * where nr_sects is lower than the number of sectors in a page.
 279  */
 280 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 281 {
 282         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 283
 284         return min(pages, (sector_t)BIO_MAX_PAGES);
 285 }
 286
 287 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 288                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 289                 struct bio **biop)
 290 {
 291         struct request_queue *q = bdev_get_queue(bdev);
 292         struct bio *bio = *biop;
 293         int bi_size = 0;
 294         unsigned int sz;
 295
 296         if (!q)
 297                 return -ENXIO;
 298
 299         if (bdev_read_only(bdev))
 300                 return -EPERM;
 301
 302         while (nr_sects != 0) {
 303                 bio = next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
 304                                gfp_mask);
 305                 bio->bi_iter.bi_sector = sector;
 306                 bio_set_dev(bio, bdev);
 307                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 308
 309                 while (nr_sects != 0) {
 310                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 311                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 312                         nr_sects -= bi_size >> 9;
 313                         sector += bi_size >> 9;
 314                         if (bi_size < sz)
 315                                 break;
 316                 }
 317                 cond_resched();
 318         }
 319
 320         *biop = bio;
 321         return 0;
 322 }
 323
 324 /**
 325  * __blkdev_issue_zeroout - generate number of zero filed write bios
 326  * @bdev:       blockdev to issue
 327  * @sector:     start sector
 328  * @nr_sects:   number of sectors to write
 329  * @gfp_mask:   memory allocation flags (for bio_alloc)
 330  * @biop:       pointer to anchor bio
 331  * @flags:      controls detailed behavior
 332  *
 333  * Description:
 334  *  Zero-fill a block range, either using hardware offload or by explicitly
 335  *  writing zeroes to the device.
 336  *
 337  *  If a device is using logical block provisioning, the underlying space will
 338  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 339  *
 340  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 341  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 342  */
 343 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 344                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 345                 unsigned flags)
 346 {
 347         int ret;
 348         sector_t bs_mask;
 349
 350         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 351         if ((sector | nr_sects) & bs_mask)
 352                 return -EINVAL;
 353
 354         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 355                         biop, flags);
 356         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 357                 return ret;
 358
 359         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 360                                          biop);
 361 }
 362 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 363
 364 /**
 365  * blkdev_issue_zeroout - zero-fill a block range
 366  * @bdev:       blockdev to write
 367  * @sector:     start sector
 368  * @nr_sects:   number of sectors to write
 369  * @gfp_mask:   memory allocation flags (for bio_alloc)
 370  * @flags:      controls detailed behavior
 371  *
 372  * Description:
 373  *  Zero-fill a block range, either using hardware offload or by explicitly
 374  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 375  *  valid values for %flags.
 376  */
 377 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 378                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 379 {
 380         int ret = 0;
 381         sector_t bs_mask;
 382         struct bio *bio;
 383         struct blk_plug plug;
 384         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 385
 386         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 387         if ((sector | nr_sects) & bs_mask)
 388                 return -EINVAL;
 389
 390 retry:
 391         bio = NULL;
 392         blk_start_plug(&plug);
 393         if (try_write_zeroes) {
 394                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 395                                                   gfp_mask, &bio, flags);
 396         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 397                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 398                                                 gfp_mask, &bio);
 399         } else {
 400                 /* No zeroing offload support */
 401                 ret = -EOPNOTSUPP;
 402         }
 403         if (ret == 0 && bio) {
 404                 ret = submit_bio_wait(bio);
 405                 bio_put(bio);
 406         }
 407         blk_finish_plug(&plug);
 408         if (ret && try_write_zeroes) {
 409                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 410                         try_write_zeroes = false;
 411                         goto retry;
 412                 }
 413                 if (!bdev_write_zeroes_sectors(bdev)) {
 414                         /*
 415                          * Zeroing offload support was indicated, but the
 416                          * device reported ILLEGAL REQUEST (for some devices
 417                          * there is no non-destructive way to verify whether
 418                          * WRITE ZEROES is actually supported).
 419                          */
 420                         ret = -EOPNOTSUPP;
 421                 }
 422         }
 423
 424         return ret;
 425 }
 426 EXPORT_SYMBOL(blkdev_issue_zeroout);