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