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