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