block/blk-lib.c

   1 /*
   2  * Functions related to generic helpers functions
   3  */
   4 #include <linux/kernel.h>
   5 #include <linux/module.h>
   6 #include <linux/bio.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/scatterlist.h>
   9
  10 #include "blk.h"
  11
  12 static struct bio *next_bio(struct bio *bio, unsigned int nr_pages,
  13                 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 granularity;
  32         unsigned int op;
  33         int alignment;
  34         sector_t bs_mask;
  35
  36         if (!q)
  37                 return -ENXIO;
  38
  39         if (flags & BLKDEV_DISCARD_SECURE) {
  40                 if (!blk_queue_secure_erase(q))
  41                         return -EOPNOTSUPP;
  42                 op = REQ_OP_SECURE_ERASE;
  43         } else {
  44                 if (!blk_queue_discard(q))
  45                         return -EOPNOTSUPP;
  46                 op = REQ_OP_DISCARD;
  47         }
  48
  49         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
  50         if ((sector | nr_sects) & bs_mask)
  51                 return -EINVAL;
  52
  53         /* Zero-sector (unknown) and one-sector granularities are the same.  */
  54         granularity = max(q->limits.discard_granularity >> 9, 1U);
  55         alignment = (bdev_discard_alignment(bdev) >> 9) % granularity;
  56
  57         while (nr_sects) {
  58                 unsigned int req_sects;
  59                 sector_t end_sect, tmp;
  60
  61                 /* Make sure bi_size doesn't overflow */
  62                 req_sects = min_t(sector_t, nr_sects, UINT_MAX >> 9);
  63
  64                 /**
  65                  * If splitting a request, and the next starting sector would be
  66                  * misaligned, stop the discard at the previous aligned sector.
  67                  */
  68                 end_sect = sector + req_sects;
  69                 tmp = end_sect;
  70                 if (req_sects < nr_sects &&
  71                     sector_div(tmp, granularity) != alignment) {
  72                         end_sect = end_sect - alignment;
  73                         sector_div(end_sect, granularity);
  74                         end_sect = end_sect * granularity + alignment;
  75                         req_sects = end_sect - sector;
  76                 }
  77
  78                 bio = next_bio(bio, 0, gfp_mask);
  79                 bio->bi_iter.bi_sector = sector;
  80                 bio->bi_bdev = bdev;
  81                 bio_set_op_attrs(bio, op, 0);
  82
  83                 bio->bi_iter.bi_size = req_sects << 9;
  84                 nr_sects -= req_sects;
  85                 sector = end_sect;
  86
  87                 /*
  88                  * We can loop for a long time in here, if someone does
  89                  * full device discards (like mkfs). Be nice and allow
  90                  * us to schedule out to avoid softlocking if preempt
  91                  * is disabled.
  92                  */
  93                 cond_resched();
  94         }
  95
  96         *biop = bio;
  97         return 0;
  98 }
  99 EXPORT_SYMBOL(__blkdev_issue_discard);
 100
 101 /**
 102  * blkdev_issue_discard - queue a discard
 103  * @bdev:       blockdev to issue discard for
 104  * @sector:     start sector
 105  * @nr_sects:   number of sectors to discard
 106  * @gfp_mask:   memory allocation flags (for bio_alloc)
 107  * @flags:      BLKDEV_DISCARD_* flags to control behaviour
 108  *
 109  * Description:
 110  *    Issue a discard request for the sectors in question.
 111  */
 112 int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 113                 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
 114 {
 115         struct bio *bio = NULL;
 116         struct blk_plug plug;
 117         int ret;
 118
 119         blk_start_plug(&plug);
 120         ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
 121                         &bio);
 122         if (!ret && bio) {
 123                 ret = submit_bio_wait(bio);
 124                 if (ret == -EOPNOTSUPP)
 125                         ret = 0;
 126                 bio_put(bio);
 127         }
 128         blk_finish_plug(&plug);
 129
 130         return ret;
 131 }
 132 EXPORT_SYMBOL(blkdev_issue_discard);
 133
 134 /**
 135  * __blkdev_issue_write_same - generate number of bios with same page
 136  * @bdev:       target blockdev
 137  * @sector:     start sector
 138  * @nr_sects:   number of sectors to write
 139  * @gfp_mask:   memory allocation flags (for bio_alloc)
 140  * @page:       page containing data to write
 141  * @biop:       pointer to anchor bio
 142  *
 143  * Description:
 144  *  Generate and issue number of bios(REQ_OP_WRITE_SAME) with same page.
 145  */
 146 static int __blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 147                 sector_t nr_sects, gfp_t gfp_mask, struct page *page,
 148                 struct bio **biop)
 149 {
 150         struct request_queue *q = bdev_get_queue(bdev);
 151         unsigned int max_write_same_sectors;
 152         struct bio *bio = *biop;
 153         sector_t bs_mask;
 154
 155         if (!q)
 156                 return -ENXIO;
 157
 158         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 159         if ((sector | nr_sects) & bs_mask)
 160                 return -EINVAL;
 161
 162         if (!bdev_write_same(bdev))
 163                 return -EOPNOTSUPP;
 164
 165         /* Ensure that max_write_same_sectors doesn't overflow bi_size */
 166         max_write_same_sectors = UINT_MAX >> 9;
 167
 168         while (nr_sects) {
 169                 bio = next_bio(bio, 1, gfp_mask);
 170                 bio->bi_iter.bi_sector = sector;
 171                 bio->bi_bdev = bdev;
 172                 bio->bi_vcnt = 1;
 173                 bio->bi_io_vec->bv_page = page;
 174                 bio->bi_io_vec->bv_offset = 0;
 175                 bio->bi_io_vec->bv_len = bdev_logical_block_size(bdev);
 176                 bio_set_op_attrs(bio, REQ_OP_WRITE_SAME, 0);
 177
 178                 if (nr_sects > max_write_same_sectors) {
 179                         bio->bi_iter.bi_size = max_write_same_sectors << 9;
 180                         nr_sects -= max_write_same_sectors;
 181                         sector += max_write_same_sectors;
 182                 } else {
 183                         bio->bi_iter.bi_size = nr_sects << 9;
 184                         nr_sects = 0;
 185                 }
 186                 cond_resched();
 187         }
 188
 189         *biop = bio;
 190         return 0;
 191 }
 192
 193 /**
 194  * blkdev_issue_write_same - queue a write same operation
 195  * @bdev:       target blockdev
 196  * @sector:     start sector
 197  * @nr_sects:   number of sectors to write
 198  * @gfp_mask:   memory allocation flags (for bio_alloc)
 199  * @page:       page containing data
 200  *
 201  * Description:
 202  *    Issue a write same request for the sectors in question.
 203  */
 204 int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 205                                 sector_t nr_sects, gfp_t gfp_mask,
 206                                 struct page *page)
 207 {
 208         struct bio *bio = NULL;
 209         struct blk_plug plug;
 210         int ret;
 211
 212         blk_start_plug(&plug);
 213         ret = __blkdev_issue_write_same(bdev, sector, nr_sects, gfp_mask, page,
 214                         &bio);
 215         if (ret == 0 && bio) {
 216                 ret = submit_bio_wait(bio);
 217                 bio_put(bio);
 218         }
 219         blk_finish_plug(&plug);
 220         return ret;
 221 }
 222 EXPORT_SYMBOL(blkdev_issue_write_same);
 223
 224 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 225                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 226                 struct bio **biop, unsigned flags)
 227 {
 228         struct bio *bio = *biop;
 229         unsigned int max_write_zeroes_sectors;
 230         struct request_queue *q = bdev_get_queue(bdev);
 231
 232         if (!q)
 233                 return -ENXIO;
 234
 235         /* Ensure that max_write_zeroes_sectors doesn't overflow bi_size */
 236         max_write_zeroes_sectors = bdev_write_zeroes_sectors(bdev);
 237
 238         if (max_write_zeroes_sectors == 0)
 239                 return -EOPNOTSUPP;
 240
 241         while (nr_sects) {
 242                 bio = next_bio(bio, 0, gfp_mask);
 243                 bio->bi_iter.bi_sector = sector;
 244                 bio->bi_bdev = bdev;
 245                 bio->bi_opf = REQ_OP_WRITE_ZEROES;
 246                 if (flags & BLKDEV_ZERO_NOUNMAP)
 247                         bio->bi_opf |= REQ_NOUNMAP;
 248
 249                 if (nr_sects > max_write_zeroes_sectors) {
 250                         bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
 251                         nr_sects -= max_write_zeroes_sectors;
 252                         sector += max_write_zeroes_sectors;
 253                 } else {
 254                         bio->bi_iter.bi_size = nr_sects << 9;
 255                         nr_sects = 0;
 256                 }
 257                 cond_resched();
 258         }
 259
 260         *biop = bio;
 261         return 0;
 262 }
 263
 264 /*
 265  * Convert a number of 512B sectors to a number of pages.
 266  * The result is limited to a number of pages that can fit into a BIO.
 267  * Also make sure that the result is always at least 1 (page) for the cases
 268  * where nr_sects is lower than the number of sectors in a page.
 269  */
 270 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects)
 271 {
 272         sector_t bytes = (nr_sects << 9) + PAGE_SIZE - 1;
 273
 274         return min(bytes >> PAGE_SHIFT, (sector_t)BIO_MAX_PAGES);
 275 }
 276
 277 static int __blkdev_issue_zero_pages(struct block_device *bdev,
 278                 sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 279                 struct bio **biop)
 280 {
 281         struct request_queue *q = bdev_get_queue(bdev);
 282         struct bio *bio = *biop;
 283         int bi_size = 0;
 284         unsigned int sz;
 285
 286         if (!q)
 287                 return -ENXIO;
 288
 289         while (nr_sects != 0) {
 290                 bio = next_bio(bio, __blkdev_sectors_to_bio_pages(nr_sects),
 291                                gfp_mask);
 292                 bio->bi_iter.bi_sector = sector;
 293                 bio->bi_bdev = bdev;
 294                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 295
 296                 while (nr_sects != 0) {
 297                         sz = min((sector_t) PAGE_SIZE, nr_sects << 9);
 298                         bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0);
 299                         nr_sects -= bi_size >> 9;
 300                         sector += bi_size >> 9;
 301                         if (bi_size < sz)
 302                                 break;
 303                 }
 304                 cond_resched();
 305         }
 306
 307         *biop = bio;
 308         return 0;
 309 }
 310
 311 /**
 312  * __blkdev_issue_zeroout - generate number of zero filed write bios
 313  * @bdev:       blockdev to issue
 314  * @sector:     start sector
 315  * @nr_sects:   number of sectors to write
 316  * @gfp_mask:   memory allocation flags (for bio_alloc)
 317  * @biop:       pointer to anchor bio
 318  * @flags:      controls detailed behavior
 319  *
 320  * Description:
 321  *  Zero-fill a block range, either using hardware offload or by explicitly
 322  *  writing zeroes to the device.
 323  *
 324  *  If a device is using logical block provisioning, the underlying space will
 325  *  not be released if %flags contains BLKDEV_ZERO_NOUNMAP.
 326  *
 327  *  If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return
 328  *  -EOPNOTSUPP if no explicit hardware offload for zeroing is provided.
 329  */
 330 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 331                 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop,
 332                 unsigned flags)
 333 {
 334         int ret;
 335         sector_t bs_mask;
 336
 337         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 338         if ((sector | nr_sects) & bs_mask)
 339                 return -EINVAL;
 340
 341         ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask,
 342                         biop, flags);
 343         if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK))
 344                 return ret;
 345
 346         return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask,
 347                                          biop);
 348 }
 349 EXPORT_SYMBOL(__blkdev_issue_zeroout);
 350
 351 /**
 352  * blkdev_issue_zeroout - zero-fill a block range
 353  * @bdev:       blockdev to write
 354  * @sector:     start sector
 355  * @nr_sects:   number of sectors to write
 356  * @gfp_mask:   memory allocation flags (for bio_alloc)
 357  * @flags:      controls detailed behavior
 358  *
 359  * Description:
 360  *  Zero-fill a block range, either using hardware offload or by explicitly
 361  *  writing zeroes to the device.  See __blkdev_issue_zeroout() for the
 362  *  valid values for %flags.
 363  */
 364 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 365                 sector_t nr_sects, gfp_t gfp_mask, unsigned flags)
 366 {
 367         int ret = 0;
 368         sector_t bs_mask;
 369         struct bio *bio;
 370         struct blk_plug plug;
 371         bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev);
 372
 373         bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
 374         if ((sector | nr_sects) & bs_mask)
 375                 return -EINVAL;
 376
 377 retry:
 378         bio = NULL;
 379         blk_start_plug(&plug);
 380         if (try_write_zeroes) {
 381                 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects,
 382                                                   gfp_mask, &bio, flags);
 383         } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 384                 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects,
 385                                                 gfp_mask, &bio);
 386         } else {
 387                 /* No zeroing offload support */
 388                 ret = -EOPNOTSUPP;
 389         }
 390         if (ret == 0 && bio) {
 391                 ret = submit_bio_wait(bio);
 392                 bio_put(bio);
 393         }
 394         blk_finish_plug(&plug);
 395         if (ret && try_write_zeroes) {
 396                 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) {
 397                         try_write_zeroes = false;
 398                         goto retry;
 399                 }
 400                 if (!bdev_write_zeroes_sectors(bdev)) {
 401                         /*
 402                          * Zeroing offload support was indicated, but the
 403                          * device reported ILLEGAL REQUEST (for some devices
 404                          * there is no non-destructive way to verify whether
 405                          * WRITE ZEROES is actually supported).
 406                          */
 407                         ret = -EOPNOTSUPP;
 408                 }
 409         }
 410
 411         return ret;
 412 }
 413 EXPORT_SYMBOL(blkdev_issue_zeroout);