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 granularity;
  33         unsigned int op;
  34         int alignment;
  35         sector_t bs_mask;
  36
  37         if (!q)
  38                 return -ENXIO;
  39
  40         if (bdev_read_only(bdev))
  41                 return -EPERM;
  42
  43         if (flags & BLKDEV_DISCARD_SECURE) {
  44                 if (!blk_queue_secure_erase(q))
  45                         return -EOPNOTSUPP;
  46                 op = REQ_OP_SECURE_ERASE;
  47         } else {
  48                 if (!blk_queue_discard(q))
  49                         return -EOPNOTSUPP;
  50                 op = REQ_OP_DISCARD;
  51         }
  52
  53         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  54         if ((sector | nr_sects) & bs_mask)
  55                 return -EINVAL;
  56
  57         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  58         granularity = max(q->limits.discard_granularity >> 9, 1U);
  59         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  60
  61         while (nr_sects) {
  62                 unsigned int req_sects;
  63                 sector_t end_sect, tmp;
  64
  65                 /*
  66                  * Issue in chunks of the user defined max discard setting,
  67                  * ensuring that bi_size doesn't overflow
  68                  */
  69                 req_sects = min_t(sector_t, nr_sects,
  70                                         q->limits.max_discard_sectors);
  71                 if (!req_sects)
  72                         goto fail;
  73                 if (req_sects > UINT_MAX >> 9)
  74                         req_sects = UINT_MAX >> 9;
  75
  76                 /*
  77                  * If splitting a request, and the next starting sector would be
  78                  * misaligned, stop the discard at the previous aligned sector.
  79                  */
  80                 end_sect = sector + req_sects;
  81                 tmp = end_sect;
  82                 if (req_sects < nr_sects &&
  83                     sector_div(tmp, granularity) != alignment) {
  84                         end_sect = end_sect - alignment;
  85                         sector_div(end_sect, granularity);
  86                         end_sect = end_sect * granularity + alignment;
  87                         req_sects = end_sect - sector;
  88                 }
  89
  90                 bio = next_bio(bio, 0, gfp_mask);
  91                 bio->bi_iter.bi_sector = sector;
  92                 bio_set_dev(bio, bdev);
  93                 bio_set_op_attrs(bio, op, 0);
  94
  95                 bio->bi_iter.bi_size = req_sects << 9;
  96                 nr_sects -= req_sects;
  97                 sector = end_sect;
  98
  99                 /*
 100                  * We can loop for a long time in here, if someone does
 101                  * full device discards (like mkfs). Be nice and allow
 102                  * us to schedule out to avoid softlocking if preempt
 103                  * is disabled.
 104                  */
 105                 cond_resched();
 106         }
 107
 108         *biop = bio;
 109         return 0;
 110
 111 fail:
 112         if (bio) {
 113                 submit_bio_wait(bio);
 114                 bio_put(bio);
 115         }
 116         *biop = NULL;
 117         return -EOPNOTSUPP;
 118 }
 119 EXPORT_SYMBOL(__blkdev_issue_discard);
 120
 121 /**
 122  * blkdev_issue_discard - queue a discard
 123  * @bdev:       blockdev to issue discard for
 124  * @sector:     start sector
 125  * @nr_sects:   number of sectors to discard
 126  * @gfp_mask:   memory allocation flags (for bio_alloc)
 127  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 128  *
 129  * Description:
 130  *    Issue a discard request for the sectors in question.
 131  */
 132 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 133                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 134 {
 135         struct bio *bio = NULL;
 136         struct blk_plug plug;
 137         int ret;
 138
 139         blk_start_plug(&plug);
 140         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 141                         &bio);
 142         if (!ret && bio) {
 143                 ret = submit_bio_wait(bio);
 144                 if (ret == -EOPNOTSUPP)
 145                         ret = 0;
 146                 bio_put(bio);
 147         }
 148         blk_finish_plug(&plug);
 149
 150         return ret;
 151 }
 152 EXPORT_SYMBOL(blkdev_issue_discard);
 153
 154 /**
 155  * __blkdev_issue_write_same - generate number of bios with same page
 156  * @bdev:       target blockdev
 157  * @sector:     start sector
 158  * @nr_sects:   number of sectors to write
 159  * @gfp_mask:   memory allocation flags (for bio_alloc)
 160  * @page:       page containing data to write
 161  * @biop:       pointer to anchor bio
 162  *
 163  * Description:
 164  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 165  */
 166 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 167                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 168                 struct bio **biop)
 169 {
 170         struct request_queue *q = bdev_get_queue(bdev);
 171         unsigned int max_write_same_sectors;
 172         struct bio *bio = *biop;
 173         sector_t bs_mask;
 174
 175         if (!q)
 176                 return -ENXIO;
 177
 178         if (bdev_read_only(bdev))
 179                 return -EPERM;
 180
 181         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 182         if ((sector | nr_sects) & bs_mask)
 183                 return -EINVAL;
 184
 185         if (!bdev_write_same(bdev))
 186                 return -EOPNOTSUPP;
 187
 188         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 189         max_write_same_sectors = UINT_MAX >> 9;
 190
 191         while (nr_sects) {
 192                 bio = next_bio(bio, 1, gfp_mask);
 193                 bio->bi_iter.bi_sector = sector;
 194                 bio_set_dev(bio, bdev);
 195                 bio->bi_vcnt = 1;
 196                 bio->bi_io_vec->bv_page = page;
 197                 bio->bi_io_vec->bv_offset = 0;
 198                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 199                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 200
 201                 if (nr_sects > max_write_same_sectors) {
 202                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 203                         nr_sects -= max_write_same_sectors;
 204                         sector += max_write_same_sectors;
 205                 } else {
 206                         bio->bi_iter.bi_size = nr_sects << 9;
 207                         nr_sects = 0;
 208                 }
 209                 cond_resched();
 210         }
 211
 212         *biop = bio;
 213         return 0;
 214 }
 215
 216 /**
 217  * blkdev_issue_write_same - queue a write same operation
 218  * @bdev:       target blockdev
 219  * @sector:     start sector
 220  * @nr_sects:   number of sectors to write
 221  * @gfp_mask:   memory allocation flags (for bio_alloc)
 222  * @page:       page containing data
 223  *
 224  * Description:
 225  *    Issue a write same request for the sectors in question.
 226  */
 227 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 228                                 sector_t nr_sects, gfp_t gfp_mask,
 229                                 struct page *page)
 230 {
 231         struct bio *bio = NULL;
 232         struct blk_plug plug;
 233         int ret;
 234
 235         blk_start_plug(&plug);
 236         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 237                         &bio);
 238         if (ret == 0 && bio) {
 239                 ret = submit_bio_wait(bio);
 240                 bio_put(bio);
 241         }
 242         blk_finish_plug(&plug);
 243         return ret;
 244 }
 245 EXPORT_SYMBOL(blkdev_issue_write_same);
 246
 247 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 248                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 249                 struct bio **biop, unsigned flags)
 250 {
 251         struct bio *bio = *biop;
 252         unsigned int max_write_zeroes_sectors;
 253         struct request_queue *q = bdev_get_queue(bdev);
 254
 255         if (!q)
 256                 return -ENXIO;
 257
 258         if (bdev_read_only(bdev))
 259                 return -EPERM;
 260
 261         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 262         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 263
 264         if (max_write_zeroes_sectors == 0)
 265                 return -EOPNOTSUPP;
 266
 267         while (nr_sects) {
 268                 bio = next_bio(bio, 0, gfp_mask);
 269                 bio->bi_iter.bi_sector = sector;
 270                 bio_set_dev(bio, bdev);
 271                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 272                 if (flags & BLKDEV_ZERO_NOUNMAP)
 273                         bio->bi_opf |= REQ_NOUNMAP;
 274
 275                 if (nr_sects > max_write_zeroes_sectors) {
 276                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 277                         nr_sects -= max_write_zeroes_sectors;
 278                         sector += max_write_zeroes_sectors;
 279                 } else {
 280                         bio->bi_iter.bi_size = nr_sects << 9;
 281                         nr_sects = 0;
 282                 }
 283                 cond_resched();
 284         }
 285
 286         *biop = bio;
 287         return 0;
 288 }
 289
 290 /*
 291  * Convert a number of 512B sectors to a number of pages.
 292  * The result is limited to a number of pages that can fit into a BIO.
 293  * Also make sure that the result is always at least 1 (page) for the cases
 294  * where nr_sects is lower than the number of sectors in a page.
 295  */
 296 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 297 {
 298         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 299
 300         return min(pages, (sector_t)BIO_MAX_PAGES);
 301 }
 302
 303 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 304                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 305                 struct bio **biop)
 306 {
 307         struct request_queue *q = bdev_get_queue(bdev);
 308         struct bio *bio = *biop;
 309         int bi_size = 0;
 310         unsigned int sz;
 311
 312         if (!q)
 313                 return -ENXIO;
 314
 315         if (bdev_read_only(bdev))
 316                 return -EPERM;
 317
 318         while (nr_sects != 0) {
 319                 bio = next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
 320                                gfp_mask);
 321                 bio->bi_iter.bi_sector = sector;
 322                 bio_set_dev(bio, bdev);
 323                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 324
 325                 while (nr_sects != 0) {
 326                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 327                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 328                         nr_sects -= bi_size >> 9;
 329                         sector += bi_size >> 9;
 330                         if (bi_size < sz)
 331                                 break;
 332                 }
 333                 cond_resched();
 334         }
 335
 336         *biop = bio;
 337         return 0;
 338 }
 339
 340 /**
 341  * __blkdev_issue_zeroout - generate number of zero filed write bios
 342  * @bdev:       blockdev to issue
 343  * @sector:     start sector
 344  * @nr_sects:   number of sectors to write
 345  * @gfp_mask:   memory allocation flags (for bio_alloc)
 346  * @biop:       pointer to anchor bio
 347  * @flags:      controls detailed behavior
 348  *
 349  * Description:
 350  *  Zero-fill a block range, either using hardware offload or by explicitly
 351  *  writing zeroes to the device.
 352  *
 353  *  If a device is using logical block provisioning, the underlying space will
 354  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 355  *
 356  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 357  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 358  */
 359 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 360                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 361                 unsigned flags)
 362 {
 363         int ret;
 364         sector_t bs_mask;
 365
 366         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 367         if ((sector | nr_sects) & bs_mask)
 368                 return -EINVAL;
 369
 370         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 371                         biop, flags);
 372         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 373                 return ret;
 374
 375         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 376                                          biop);
 377 }
 378 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 379
 380 /**
 381  * blkdev_issue_zeroout - zero-fill a block range
 382  * @bdev:       blockdev to write
 383  * @sector:     start sector
 384  * @nr_sects:   number of sectors to write
 385  * @gfp_mask:   memory allocation flags (for bio_alloc)
 386  * @flags:      controls detailed behavior
 387  *
 388  * Description:
 389  *  Zero-fill a block range, either using hardware offload or by explicitly
 390  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 391  *  valid values for %flags.
 392  */
 393 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 394                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 395 {
 396         int ret = 0;
 397         sector_t bs_mask;
 398         struct bio *bio;
 399         struct blk_plug plug;
 400         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 401
 402         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 403         if ((sector | nr_sects) & bs_mask)
 404                 return -EINVAL;
 405
 406 retry:
 407         bio = NULL;
 408         blk_start_plug(&plug);
 409         if (try_write_zeroes) {
 410                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 411                                                   gfp_mask, &bio, flags);
 412         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 413                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 414                                                 gfp_mask, &bio);
 415         } else {
 416                 /* No zeroing offload support */
 417                 ret = -EOPNOTSUPP;
 418         }
 419         if (ret == 0 && bio) {
 420                 ret = submit_bio_wait(bio);
 421                 bio_put(bio);
 422         }
 423         blk_finish_plug(&plug);
 424         if (ret && try_write_zeroes) {
 425                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 426                         try_write_zeroes = false;
 427                         goto retry;
 428                 }
 429                 if (!bdev_write_zeroes_sectors(bdev)) {
 430                         /*
 431                          * Zeroing offload support was indicated, but the
 432                          * device reported ILLEGAL REQUEST (for some devices
 433                          * there is no non-destructive way to verify whether
 434                          * WRITE ZEROES is actually supported).
 435                          */
 436                         ret = -EOPNOTSUPP;
 437                 }
 438         }
 439
 440         return ret;
 441 }
 442 EXPORT_SYMBOL(blkdev_issue_zeroout);