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