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