block/bio-integrity.c

   1 /*
   2  * bio-integrity.c - bio data integrity extensions
   3  *
   4  * Copyright (C) 2007, 2008, 2009 Oracle Corporation
   5  * Written by: Martin K. Petersen <martin.petersen@oracle.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; see the file COPYING.  If not, write to
  18  * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
  19  * USA.
  20  *
  21  */
  22
  23 #include <linux/blkdev.h>
  24 #include <linux/mempool.h>
  25 #include <linux/export.h>
  26 #include <linux/bio.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/slab.h>
  29
  30 #define BIP_INLINE_VECS 4
  31
  32 static struct kmem_cache *bip_slab;
  33 static struct workqueue_struct *kintegrityd_wq;
  34
  35 /**
  36  * bio_integrity_alloc - Allocate integrity payload and attach it to bio
  37  * @bio:        bio to attach integrity metadata to
  38  * @gfp_mask:   Memory allocation mask
  39  * @nr_vecs:    Number of integrity metadata scatter-gather elements
  40  *
  41  * Description: This function prepares a bio for attaching integrity
  42  * metadata.  nr_vecs specifies the maximum number of pages containing
  43  * integrity metadata that can be attached.
  44  */
  45 struct bio_integrity_payload *bio_integrity_alloc(struct bio *bio,
  46                                                   gfp_t gfp_mask,
  47                                                   unsigned int nr_vecs)
  48 {
  49         struct bio_integrity_payload *bip;
  50         struct bio_set *bs = bio->bi_pool;
  51         unsigned long idx = BIO_POOL_NONE;
  52         unsigned inline_vecs;
  53
  54         if (!bs) {
  55                 bip = kmalloc(sizeof(struct bio_integrity_payload) +
  56                               sizeof(struct bio_vec) * nr_vecs, gfp_mask);
  57                 inline_vecs = nr_vecs;
  58         } else {
  59                 bip = mempool_alloc(bs->bio_integrity_pool, gfp_mask);
  60                 inline_vecs = BIP_INLINE_VECS;
  61         }
  62
  63         if (unlikely(!bip))
  64                 return NULL;
  65
  66         memset(bip, 0, sizeof(*bip));
  67
  68         if (nr_vecs > inline_vecs) {
  69                 bip->bip_vec = bvec_alloc(gfp_mask, nr_vecs, &idx,
  70                                           bs->bvec_integrity_pool);
  71                 if (!bip->bip_vec)
  72                         goto err;
  73                 bip->bip_max_vcnt = bvec_nr_vecs(idx);
  74         } else {
  75                 bip->bip_vec = bip->bip_inline_vecs;
  76                 bip->bip_max_vcnt = inline_vecs;
  77         }
  78
  79         bip->bip_slab = idx;
  80         bip->bip_bio = bio;
  81         bio->bi_integrity = bip;
  82         bio->bi_rw |= REQ_INTEGRITY;
  83
  84         return bip;
  85 err:
  86         mempool_free(bip, bs->bio_integrity_pool);
  87         return NULL;
  88 }
  89 EXPORT_SYMBOL(bio_integrity_alloc);
  90
  91 /**
  92  * bio_integrity_free - Free bio integrity payload
  93  * @bio:        bio containing bip to be freed
  94  *
  95  * Description: Used to free the integrity portion of a bio. Usually
  96  * called from bio_free().
  97  */
  98 void bio_integrity_free(struct bio *bio)
  99 {
 100         struct bio_integrity_payload *bip = bio_integrity(bio);
 101         struct bio_set *bs = bio->bi_pool;
 102
 103         if (bip->bip_owns_buf)
 104                 kfree(bip->bip_buf);
 105
 106         if (bs) {
 107                 if (bip->bip_slab != BIO_POOL_NONE)
 108                         bvec_free(bs->bvec_integrity_pool, bip->bip_vec,
 109                                   bip->bip_slab);
 110
 111                 mempool_free(bip, bs->bio_integrity_pool);
 112         } else {
 113                 kfree(bip);
 114         }
 115
 116         bio->bi_integrity = NULL;
 117 }
 118 EXPORT_SYMBOL(bio_integrity_free);
 119
 120 /**
 121  * bio_integrity_add_page - Attach integrity metadata
 122  * @bio:        bio to update
 123  * @page:       page containing integrity metadata
 124  * @len:        number of bytes of integrity metadata in page
 125  * @offset:     start offset within page
 126  *
 127  * Description: Attach a page containing integrity metadata to bio.
 128  */
 129 int bio_integrity_add_page(struct bio *bio, struct page *page,
 130                            unsigned int len, unsigned int offset)
 131 {
 132         struct bio_integrity_payload *bip = bio_integrity(bio);
 133         struct bio_vec *iv;
 134
 135         if (bip->bip_vcnt >= bip->bip_max_vcnt) {
 136                 printk(KERN_ERR "%s: bip_vec full\n", __func__);
 137                 return 0;
 138         }
 139
 140         iv = bip->bip_vec + bip->bip_vcnt;
 141
 142         iv->bv_page = page;
 143         iv->bv_len = len;
 144         iv->bv_offset = offset;
 145         bip->bip_vcnt++;
 146
 147         return len;
 148 }
 149 EXPORT_SYMBOL(bio_integrity_add_page);
 150
 151 /**
 152  * bio_integrity_enabled - Check whether integrity can be passed
 153  * @bio:        bio to check
 154  *
 155  * Description: Determines whether bio_integrity_prep() can be called
 156  * on this bio or not.  bio data direction and target device must be
 157  * set prior to calling.  The functions honors the write_generate and
 158  * read_verify flags in sysfs.
 159  */
 160 bool bio_integrity_enabled(struct bio *bio)
 161 {
 162         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 163
 164         if (!bio_is_rw(bio))
 165                 return false;
 166
 167         /* Already protected? */
 168         if (bio_integrity(bio))
 169                 return false;
 170
 171         if (bi == NULL)
 172                 return false;
 173
 174         if (bio_data_dir(bio) == READ && bi->verify_fn != NULL &&
 175             (bi->flags & INTEGRITY_FLAG_READ))
 176                 return true;
 177
 178         if (bio_data_dir(bio) == WRITE && bi->generate_fn != NULL &&
 179             (bi->flags & INTEGRITY_FLAG_WRITE))
 180                 return true;
 181
 182         return false;
 183 }
 184 EXPORT_SYMBOL(bio_integrity_enabled);
 185
 186 /**
 187  * bio_integrity_hw_sectors - Convert 512b sectors to hardware ditto
 188  * @bi:         blk_integrity profile for device
 189  * @sectors:    Number of 512 sectors to convert
 190  *
 191  * Description: The block layer calculates everything in 512 byte
 192  * sectors but integrity metadata is done in terms of the hardware
 193  * sector size of the storage device.  Convert the block layer sectors
 194  * to physical sectors.
 195  */
 196 static inline unsigned int bio_integrity_hw_sectors(struct blk_integrity *bi,
 197                                                     unsigned int sectors)
 198 {
 199         /* At this point there are only 512b or 4096b DIF/EPP devices */
 200         if (bi->sector_size == 4096)
 201                 return sectors >>= 3;
 202
 203         return sectors;
 204 }
 205
 206 static inline unsigned int bio_integrity_bytes(struct blk_integrity *bi,
 207                                                unsigned int sectors)
 208 {
 209         return bio_integrity_hw_sectors(bi, sectors) * bi->tuple_size;
 210 }
 211
 212 /**
 213  * bio_integrity_generate_verify - Generate/verify integrity metadata for a bio
 214  * @bio:        bio to generate/verify integrity metadata for
 215  * @operate:    operate number, 1 for generate, 0 for verify
 216  */
 217 static int bio_integrity_generate_verify(struct bio *bio, int operate)
 218 {
 219         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 220         struct blk_integrity_exchg bix;
 221         struct bio_vec *bv;
 222         sector_t sector;
 223         unsigned int sectors, ret = 0, i;
 224         void *prot_buf = bio_integrity(bio)->bip_buf;
 225
 226         if (operate)
 227                 sector = bio->bi_iter.bi_sector;
 228         else
 229                 sector = bio_integrity(bio)->bip_iter.bi_sector;
 230
 231         bix.disk_name = bio->bi_bdev->bd_disk->disk_name;
 232         bix.sector_size = bi->sector_size;
 233
 234         bio_for_each_segment_all(bv, bio, i) {
 235                 void *kaddr = kmap_atomic(bv->bv_page);
 236                 bix.data_buf = kaddr + bv->bv_offset;
 237                 bix.data_size = bv->bv_len;
 238                 bix.prot_buf = prot_buf;
 239                 bix.sector = sector;
 240
 241                 if (operate)
 242                         bi->generate_fn(&bix);
 243                 else {
 244                         ret = bi->verify_fn(&bix);
 245                         if (ret) {
 246                                 kunmap_atomic(kaddr);
 247                                 return ret;
 248                         }
 249                 }
 250
 251                 sectors = bv->bv_len / bi->sector_size;
 252                 sector += sectors;
 253                 prot_buf += sectors * bi->tuple_size;
 254
 255                 kunmap_atomic(kaddr);
 256         }
 257         return ret;
 258 }
 259
 260 /**
 261  * bio_integrity_generate - Generate integrity metadata for a bio
 262  * @bio:        bio to generate integrity metadata for
 263  *
 264  * Description: Generates integrity metadata for a bio by calling the
 265  * block device's generation callback function.  The bio must have a
 266  * bip attached with enough room to accommodate the generated
 267  * integrity metadata.
 268  */
 269 static void bio_integrity_generate(struct bio *bio)
 270 {
 271         bio_integrity_generate_verify(bio, 1);
 272 }
 273
 274 /**
 275  * bio_integrity_prep - Prepare bio for integrity I/O
 276  * @bio:        bio to prepare
 277  *
 278  * Description: Allocates a buffer for integrity metadata, maps the
 279  * pages and attaches them to a bio.  The bio must have data
 280  * direction, target device and start sector set priot to calling.  In
 281  * the WRITE case, integrity metadata will be generated using the
 282  * block device's integrity function.  In the READ case, the buffer
 283  * will be prepared for DMA and a suitable end_io handler set up.
 284  */
 285 int bio_integrity_prep(struct bio *bio)
 286 {
 287         struct bio_integrity_payload *bip;
 288         struct blk_integrity *bi;
 289         struct request_queue *q;
 290         void *buf;
 291         unsigned long start, end;
 292         unsigned int len, nr_pages;
 293         unsigned int bytes, offset, i;
 294         unsigned int sectors;
 295
 296         bi = bdev_get_integrity(bio->bi_bdev);
 297         q = bdev_get_queue(bio->bi_bdev);
 298         BUG_ON(bi == NULL);
 299         BUG_ON(bio_integrity(bio));
 300
 301         sectors = bio_integrity_hw_sectors(bi, bio_sectors(bio));
 302
 303         /* Allocate kernel buffer for protection data */
 304         len = sectors * bi->tuple_size;
 305         buf = kmalloc(len, GFP_NOIO | q->bounce_gfp);
 306         if (unlikely(buf == NULL)) {
 307                 printk(KERN_ERR "could not allocate integrity buffer\n");
 308                 return -ENOMEM;
 309         }
 310
 311         end = (((unsigned long) buf) + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 312         start = ((unsigned long) buf) >> PAGE_SHIFT;
 313         nr_pages = end - start;
 314
 315         /* Allocate bio integrity payload and integrity vectors */
 316         bip = bio_integrity_alloc(bio, GFP_NOIO, nr_pages);
 317         if (unlikely(bip == NULL)) {
 318                 printk(KERN_ERR "could not allocate data integrity bioset\n");
 319                 kfree(buf);
 320                 return -EIO;
 321         }
 322
 323         bip->bip_owns_buf = 1;
 324         bip->bip_buf = buf;
 325         bip->bip_iter.bi_size = len;
 326         bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
 327
 328         /* Map it */
 329         offset = offset_in_page(buf);
 330         for (i = 0 ; i < nr_pages ; i++) {
 331                 int ret;
 332                 bytes = PAGE_SIZE - offset;
 333
 334                 if (len <= 0)
 335                         break;
 336
 337                 if (bytes > len)
 338                         bytes = len;
 339
 340                 ret = bio_integrity_add_page(bio, virt_to_page(buf),
 341                                              bytes, offset);
 342
 343                 if (ret == 0)
 344                         return 0;
 345
 346                 if (ret < bytes)
 347                         break;
 348
 349                 buf += bytes;
 350                 len -= bytes;
 351                 offset = 0;
 352         }
 353
 354         /* Install custom I/O completion handler if read verify is enabled */
 355         if (bio_data_dir(bio) == READ) {
 356                 bip->bip_end_io = bio->bi_end_io;
 357                 bio->bi_end_io = bio_integrity_endio;
 358         }
 359
 360         /* Auto-generate integrity metadata if this is a write */
 361         if (bio_data_dir(bio) == WRITE)
 362                 bio_integrity_generate(bio);
 363
 364         return 0;
 365 }
 366 EXPORT_SYMBOL(bio_integrity_prep);
 367
 368 /**
 369  * bio_integrity_verify - Verify integrity metadata for a bio
 370  * @bio:        bio to verify
 371  *
 372  * Description: This function is called to verify the integrity of a
 373  * bio.  The data in the bio io_vec is compared to the integrity
 374  * metadata returned by the HBA.
 375  */
 376 static int bio_integrity_verify(struct bio *bio)
 377 {
 378         return bio_integrity_generate_verify(bio, 0);
 379 }
 380
 381 /**
 382  * bio_integrity_verify_fn - Integrity I/O completion worker
 383  * @work:       Work struct stored in bio to be verified
 384  *
 385  * Description: This workqueue function is called to complete a READ
 386  * request.  The function verifies the transferred integrity metadata
 387  * and then calls the original bio end_io function.
 388  */
 389 static void bio_integrity_verify_fn(struct work_struct *work)
 390 {
 391         struct bio_integrity_payload *bip =
 392                 container_of(work, struct bio_integrity_payload, bip_work);
 393         struct bio *bio = bip->bip_bio;
 394         int error;
 395
 396         error = bio_integrity_verify(bio);
 397
 398         /* Restore original bio completion handler */
 399         bio->bi_end_io = bip->bip_end_io;
 400         bio_endio_nodec(bio, error);
 401 }
 402
 403 /**
 404  * bio_integrity_endio - Integrity I/O completion function
 405  * @bio:        Protected bio
 406  * @error:      Pointer to errno
 407  *
 408  * Description: Completion for integrity I/O
 409  *
 410  * Normally I/O completion is done in interrupt context.  However,
 411  * verifying I/O integrity is a time-consuming task which must be run
 412  * in process context.  This function postpones completion
 413  * accordingly.
 414  */
 415 void bio_integrity_endio(struct bio *bio, int error)
 416 {
 417         struct bio_integrity_payload *bip = bio_integrity(bio);
 418
 419         BUG_ON(bip->bip_bio != bio);
 420
 421         /* In case of an I/O error there is no point in verifying the
 422          * integrity metadata.  Restore original bio end_io handler
 423          * and run it.
 424          */
 425         if (error) {
 426                 bio->bi_end_io = bip->bip_end_io;
 427                 bio_endio_nodec(bio, error);
 428
 429                 return;
 430         }
 431
 432         INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);
 433         queue_work(kintegrityd_wq, &bip->bip_work);
 434 }
 435 EXPORT_SYMBOL(bio_integrity_endio);
 436
 437 /**
 438  * bio_integrity_advance - Advance integrity vector
 439  * @bio:        bio whose integrity vector to update
 440  * @bytes_done: number of data bytes that have been completed
 441  *
 442  * Description: This function calculates how many integrity bytes the
 443  * number of completed data bytes correspond to and advances the
 444  * integrity vector accordingly.
 445  */
 446 void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)
 447 {
 448         struct bio_integrity_payload *bip = bio_integrity(bio);
 449         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 450         unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);
 451
 452         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);
 453 }
 454 EXPORT_SYMBOL(bio_integrity_advance);
 455
 456 /**
 457  * bio_integrity_trim - Trim integrity vector
 458  * @bio:        bio whose integrity vector to update
 459  * @offset:     offset to first data sector
 460  * @sectors:    number of data sectors
 461  *
 462  * Description: Used to trim the integrity vector in a cloned bio.
 463  * The ivec will be advanced corresponding to 'offset' data sectors
 464  * and the length will be truncated corresponding to 'len' data
 465  * sectors.
 466  */
 467 void bio_integrity_trim(struct bio *bio, unsigned int offset,
 468                         unsigned int sectors)
 469 {
 470         struct bio_integrity_payload *bip = bio_integrity(bio);
 471         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 472
 473         bio_integrity_advance(bio, offset << 9);
 474         bip->bip_iter.bi_size = bio_integrity_bytes(bi, sectors);
 475 }
 476 EXPORT_SYMBOL(bio_integrity_trim);
 477
 478 /**
 479  * bio_integrity_clone - Callback for cloning bios with integrity metadata
 480  * @bio:        New bio
 481  * @bio_src:    Original bio
 482  * @gfp_mask:   Memory allocation mask
 483  *
 484  * Description: Called to allocate a bip when cloning a bio
 485  */
 486 int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
 487                         gfp_t gfp_mask)
 488 {
 489         struct bio_integrity_payload *bip_src = bio_integrity(bio_src);
 490         struct bio_integrity_payload *bip;
 491
 492         BUG_ON(bip_src == NULL);
 493
 494         bip = bio_integrity_alloc(bio, gfp_mask, bip_src->bip_vcnt);
 495
 496         if (bip == NULL)
 497                 return -EIO;
 498
 499         memcpy(bip->bip_vec, bip_src->bip_vec,
 500                bip_src->bip_vcnt * sizeof(struct bio_vec));
 501
 502         bip->bip_vcnt = bip_src->bip_vcnt;
 503         bip->bip_iter = bip_src->bip_iter;
 504
 505         return 0;
 506 }
 507 EXPORT_SYMBOL(bio_integrity_clone);
 508
 509 int bioset_integrity_create(struct bio_set *bs, int pool_size)
 510 {
 511         if (bs->bio_integrity_pool)
 512                 return 0;
 513
 514         bs->bio_integrity_pool = mempool_create_slab_pool(pool_size, bip_slab);
 515         if (!bs->bio_integrity_pool)
 516                 return -1;
 517
 518         bs->bvec_integrity_pool = biovec_create_pool(pool_size);
 519         if (!bs->bvec_integrity_pool) {
 520                 mempool_destroy(bs->bio_integrity_pool);
 521                 return -1;
 522         }
 523
 524         return 0;
 525 }
 526 EXPORT_SYMBOL(bioset_integrity_create);
 527
 528 void bioset_integrity_free(struct bio_set *bs)
 529 {
 530         if (bs->bio_integrity_pool)
 531                 mempool_destroy(bs->bio_integrity_pool);
 532
 533         if (bs->bvec_integrity_pool)
 534                 mempool_destroy(bs->bvec_integrity_pool);
 535 }
 536 EXPORT_SYMBOL(bioset_integrity_free);
 537
 538 void __init bio_integrity_init(void)
 539 {
 540         /*
 541          * kintegrityd won't block much but may burn a lot of CPU cycles.
 542          * Make it highpri CPU intensive wq with max concurrency of 1.
 543          */
 544         kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |
 545                                          WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);
 546         if (!kintegrityd_wq)
 547                 panic("Failed to create kintegrityd\n");
 548
 549         bip_slab = kmem_cache_create("bio_integrity_payload",
 550                                      sizeof(struct bio_integrity_payload) +
 551                                      sizeof(struct bio_vec) * BIP_INLINE_VECS,
 552                                      0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 553         if (!bip_slab)
 554                 panic("Failed to create slab\n");
 555 }