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 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp)
  14 {
  15         struct bio *new = bio_alloc(gfp, nr_pages);
  16
  17         if (bio) {
  18                 bio_chain(bio, new);
  19                 submit_bio(bio);
  20         }
  21
  22         return new;
  23 }
  24
  25 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
  26                 sector_t nr_sects, gfp_t gfp_mask, int flags,
  27                 struct bio **biop)
  28 {
  29         struct request_queue *q = bdev_get_queue(bdev);
  30         struct bio *bio = *biop;
  31         unsigned int op;
  32         sector_t bs_mask;
  33
  34         if (!q)
  35                 return -ENXIO;
  36
  37         if (bdev_read_only(bdev))
  38                 return -EPERM;
  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         if (!nr_sects)
  55                 return -EINVAL;
  56
  57         while (nr_sects) {
  58                 unsigned int req_sects = min_t(unsigned int, nr_sects,
  59                                 bio_allowed_max_sectors(q));
  60
  61                 bio = blk_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         if (bdev_read_only(bdev))
 142                 return -EPERM;
 143
 144         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 145         if ((sector | nr_sects) & bs_mask)
 146                 return -EINVAL;
 147
 148         if (!bdev_write_same(bdev))
 149                 return -EOPNOTSUPP;
 150
 151         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 152         max_write_same_sectors = bio_allowed_max_sectors(q);
 153
 154         while (nr_sects) {
 155                 bio = blk_next_bio(bio, 1, gfp_mask);
 156                 bio->bi_iter.bi_sector = sector;
 157                 bio_set_dev(bio, bdev);
 158                 bio->bi_vcnt = 1;
 159                 bio->bi_io_vec->bv_page = page;
 160                 bio->bi_io_vec->bv_offset = 0;
 161                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 162                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 163
 164                 if (nr_sects > max_write_same_sectors) {
 165                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 166                         nr_sects -= max_write_same_sectors;
 167                         sector += max_write_same_sectors;
 168                 } else {
 169                         bio->bi_iter.bi_size = nr_sects << 9;
 170                         nr_sects = 0;
 171                 }
 172                 cond_resched();
 173         }
 174
 175         *biop = bio;
 176         return 0;
 177 }
 178
 179 /**
 180  * blkdev_issue_write_same - queue a write same operation
 181  * @bdev:       target blockdev
 182  * @sector:     start sector
 183  * @nr_sects:   number of sectors to write
 184  * @gfp_mask:   memory allocation flags (for bio_alloc)
 185  * @page:       page containing data
 186  *
 187  * Description:
 188  *    Issue a write same request for the sectors in question.
 189  */
 190 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 191                                 sector_t nr_sects, gfp_t gfp_mask,
 192                                 struct page *page)
 193 {
 194         struct bio *bio = NULL;
 195         struct blk_plug plug;
 196         int ret;
 197
 198         blk_start_plug(&plug);
 199         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 200                         &bio);
 201         if (ret == 0 && bio) {
 202                 ret = submit_bio_wait(bio);
 203                 bio_put(bio);
 204         }
 205         blk_finish_plug(&plug);
 206         return ret;
 207 }
 208 EXPORT_SYMBOL(blkdev_issue_write_same);
 209
 210 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 211                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 212                 struct bio **biop, unsigned flags)
 213 {
 214         struct bio *bio = *biop;
 215         unsigned int max_write_zeroes_sectors;
 216         struct request_queue *q = bdev_get_queue(bdev);
 217
 218         if (!q)
 219                 return -ENXIO;
 220
 221         if (bdev_read_only(bdev))
 222                 return -EPERM;
 223
 224         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 225         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 226
 227         if (max_write_zeroes_sectors == 0)
 228                 return -EOPNOTSUPP;
 229
 230         while (nr_sects) {
 231                 bio = blk_next_bio(bio, 0, gfp_mask);
 232                 bio->bi_iter.bi_sector = sector;
 233                 bio_set_dev(bio, bdev);
 234                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 235                 if (flags & BLKDEV_ZERO_NOUNMAP)
 236                         bio->bi_opf |= REQ_NOUNMAP;
 237
 238                 if (nr_sects > max_write_zeroes_sectors) {
 239                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 240                         nr_sects -= max_write_zeroes_sectors;
 241                         sector += max_write_zeroes_sectors;
 242                 } else {
 243                         bio->bi_iter.bi_size = nr_sects << 9;
 244                         nr_sects = 0;
 245                 }
 246                 cond_resched();
 247         }
 248
 249         *biop = bio;
 250         return 0;
 251 }
 252
 253 /*
 254  * Convert a number of 512B sectors to a number of pages.
 255  * The result is limited to a number of pages that can fit into a BIO.
 256  * Also make sure that the result is always at least 1 (page) for the cases
 257  * where nr_sects is lower than the number of sectors in a page.
 258  */
 259 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 260 {
 261         sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512);
 262
 263         return min(pages, (sector_t)BIO_MAX_PAGES);
 264 }
 265
 266 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 267                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 268                 struct bio **biop)
 269 {
 270         struct request_queue *q = bdev_get_queue(bdev);
 271         struct bio *bio = *biop;
 272         int bi_size = 0;
 273         unsigned int sz;
 274
 275         if (!q)
 276                 return -ENXIO;
 277
 278         if (bdev_read_only(bdev))
 279                 return -EPERM;
 280
 281         while (nr_sects != 0) {
 282                 bio = blk_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);