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