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