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(page_address(bip->bip_vec->bv_page) +
 105                       bip->bip_vec->bv_offset);
 106
 107         if (bs) {
 108                 if (bip->bip_slab != BIO_POOL_NONE)
 109                         bvec_free(bs->bvec_integrity_pool, bip->bip_vec,
 110                                   bip->bip_slab);
 111
 112                 mempool_free(bip, bs->bio_integrity_pool);
 113         } else {
 114                 kfree(bip);
 115         }
 116
 117         bio->bi_integrity = NULL;
 118 }
 119 EXPORT_SYMBOL(bio_integrity_free);
 120
 121 /**
 122  * bio_integrity_add_page - Attach integrity metadata
 123  * @bio:        bio to update
 124  * @page:       page containing integrity metadata
 125  * @len:        number of bytes of integrity metadata in page
 126  * @offset:     start offset within page
 127  *
 128  * Description: Attach a page containing integrity metadata to bio.
 129  */
 130 int bio_integrity_add_page(struct bio *bio, struct page *page,
 131                            unsigned int len, unsigned int offset)
 132 {
 133         struct bio_integrity_payload *bip = bio_integrity(bio);
 134         struct bio_vec *iv;
 135
 136         if (bip->bip_vcnt >= bip->bip_max_vcnt) {
 137                 printk(KERN_ERR "%s: bip_vec full\n", __func__);
 138                 return 0;
 139         }
 140
 141         iv = bip->bip_vec + bip->bip_vcnt;
 142
 143         iv->bv_page = page;
 144         iv->bv_len = len;
 145         iv->bv_offset = offset;
 146         bip->bip_vcnt++;
 147
 148         return len;
 149 }
 150 EXPORT_SYMBOL(bio_integrity_add_page);
 151
 152 /**
 153  * bio_integrity_enabled - Check whether integrity can be passed
 154  * @bio:        bio to check
 155  *
 156  * Description: Determines whether bio_integrity_prep() can be called
 157  * on this bio or not.  bio data direction and target device must be
 158  * set prior to calling.  The functions honors the write_generate and
 159  * read_verify flags in sysfs.
 160  */
 161 bool bio_integrity_enabled(struct bio *bio)
 162 {
 163         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 164
 165         if (!bio_is_rw(bio))
 166                 return false;
 167
 168         /* Already protected? */
 169         if (bio_integrity(bio))
 170                 return false;
 171
 172         if (bi == NULL)
 173                 return false;
 174
 175         if (bio_data_dir(bio) == READ && bi->verify_fn != NULL &&
 176             (bi->flags & INTEGRITY_FLAG_READ))
 177                 return true;
 178
 179         if (bio_data_dir(bio) == WRITE && bi->generate_fn != NULL &&
 180             (bi->flags & INTEGRITY_FLAG_WRITE))
 181                 return true;
 182
 183         return false;
 184 }
 185 EXPORT_SYMBOL(bio_integrity_enabled);
 186
 187 /**
 188  * bio_integrity_hw_sectors - Convert 512b sectors to hardware ditto
 189  * @bi:         blk_integrity profile for device
 190  * @sectors:    Number of 512 sectors to convert
 191  *
 192  * Description: The block layer calculates everything in 512 byte
 193  * sectors but integrity metadata is done in terms of the hardware
 194  * sector size of the storage device.  Convert the block layer sectors
 195  * to physical sectors.
 196  */
 197 static inline unsigned int bio_integrity_hw_sectors(struct blk_integrity *bi,
 198                                                     unsigned int sectors)
 199 {
 200         /* At this point there are only 512b or 4096b DIF/EPP devices */
 201         if (bi->sector_size == 4096)
 202                 return sectors >>= 3;
 203
 204         return sectors;
 205 }
 206
 207 static inline unsigned int bio_integrity_bytes(struct blk_integrity *bi,
 208                                                unsigned int sectors)
 209 {
 210         return bio_integrity_hw_sectors(bi, sectors) * bi->tuple_size;
 211 }
 212
 213 /**
 214  * bio_integrity_generate_verify - Generate/verify integrity metadata for a bio
 215  * @bio:        bio to generate/verify integrity metadata for
 216  * @operate:    operate number, 1 for generate, 0 for verify
 217  */
 218 static int bio_integrity_generate_verify(struct bio *bio, int operate)
 219 {
 220         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 221         struct blk_integrity_exchg bix;
 222         struct bio_vec *bv;
 223         struct bio_integrity_payload *bip = bio_integrity(bio);
 224         sector_t sector;
 225         unsigned int sectors, ret = 0, i;
 226         void *prot_buf = page_address(bip->bip_vec->bv_page) +
 227                 bip->bip_vec->bv_offset;
 228
 229         if (operate)
 230                 sector = bio->bi_iter.bi_sector;
 231         else
 232                 sector = bip->bip_iter.bi_sector;
 233
 234         bix.disk_name = bio->bi_bdev->bd_disk->disk_name;
 235         bix.sector_size = bi->sector_size;
 236
 237         bio_for_each_segment_all(bv, bio, i) {
 238                 void *kaddr = kmap_atomic(bv->bv_page);
 239                 bix.data_buf = kaddr + bv->bv_offset;
 240                 bix.data_size = bv->bv_len;
 241                 bix.prot_buf = prot_buf;
 242                 bix.sector = sector;
 243
 244                 if (operate)
 245                         bi->generate_fn(&bix);
 246                 else {
 247                         ret = bi->verify_fn(&bix);
 248                         if (ret) {
 249                                 kunmap_atomic(kaddr);
 250                                 return ret;
 251                         }
 252                 }
 253
 254                 sectors = bv->bv_len / bi->sector_size;
 255                 sector += sectors;
 256                 prot_buf += sectors * bi->tuple_size;
 257
 258                 kunmap_atomic(kaddr);
 259         }
 260         return ret;
 261 }
 262
 263 /**
 264  * bio_integrity_generate - Generate integrity metadata for a bio
 265  * @bio:        bio to generate integrity metadata for
 266  *
 267  * Description: Generates integrity metadata for a bio by calling the
 268  * block device's generation callback function.  The bio must have a
 269  * bip attached with enough room to accommodate the generated
 270  * integrity metadata.
 271  */
 272 static void bio_integrity_generate(struct bio *bio)
 273 {
 274         bio_integrity_generate_verify(bio, 1);
 275 }
 276
 277 /**
 278  * bio_integrity_prep - Prepare bio for integrity I/O
 279  * @bio:        bio to prepare
 280  *
 281  * Description: Allocates a buffer for integrity metadata, maps the
 282  * pages and attaches them to a bio.  The bio must have data
 283  * direction, target device and start sector set priot to calling.  In
 284  * the WRITE case, integrity metadata will be generated using the
 285  * block device's integrity function.  In the READ case, the buffer
 286  * will be prepared for DMA and a suitable end_io handler set up.
 287  */
 288 int bio_integrity_prep(struct bio *bio)
 289 {
 290         struct bio_integrity_payload *bip;
 291         struct blk_integrity *bi;
 292         struct request_queue *q;
 293         void *buf;
 294         unsigned long start, end;
 295         unsigned int len, nr_pages;
 296         unsigned int bytes, offset, i;
 297         unsigned int sectors;
 298
 299         bi = bdev_get_integrity(bio->bi_bdev);
 300         q = bdev_get_queue(bio->bi_bdev);
 301         BUG_ON(bi == NULL);
 302         BUG_ON(bio_integrity(bio));
 303
 304         sectors = bio_integrity_hw_sectors(bi, bio_sectors(bio));
 305
 306         /* Allocate kernel buffer for protection data */
 307         len = sectors * bi->tuple_size;
 308         buf = kmalloc(len, GFP_NOIO | q->bounce_gfp);
 309         if (unlikely(buf == NULL)) {
 310                 printk(KERN_ERR "could not allocate integrity buffer\n");
 311                 return -ENOMEM;
 312         }
 313
 314         end = (((unsigned long) buf) + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 315         start = ((unsigned long) buf) >> PAGE_SHIFT;
 316         nr_pages = end - start;
 317
 318         /* Allocate bio integrity payload and integrity vectors */
 319         bip = bio_integrity_alloc(bio, GFP_NOIO, nr_pages);
 320         if (unlikely(bip == NULL)) {
 321                 printk(KERN_ERR "could not allocate data integrity bioset\n");
 322                 kfree(buf);
 323                 return -EIO;
 324         }
 325
 326         bip->bip_owns_buf = 1;
 327         bip->bip_iter.bi_size = len;
 328         bip->bip_iter.bi_sector = bio->bi_iter.bi_sector;
 329
 330         /* Map it */
 331         offset = offset_in_page(buf);
 332         for (i = 0 ; i < nr_pages ; i++) {
 333                 int ret;
 334                 bytes = PAGE_SIZE - offset;
 335
 336                 if (len <= 0)
 337                         break;
 338
 339                 if (bytes > len)
 340                         bytes = len;
 341
 342                 ret = bio_integrity_add_page(bio, virt_to_page(buf),
 343                                              bytes, offset);
 344
 345                 if (ret == 0)
 346                         return 0;
 347
 348                 if (ret < bytes)
 349                         break;
 350
 351                 buf += bytes;
 352                 len -= bytes;
 353                 offset = 0;
 354         }
 355
 356         /* Install custom I/O completion handler if read verify is enabled */
 357         if (bio_data_dir(bio) == READ) {
 358                 bip->bip_end_io = bio->bi_end_io;
 359                 bio->bi_end_io = bio_integrity_endio;
 360         }
 361
 362         /* Auto-generate integrity metadata if this is a write */
 363         if (bio_data_dir(bio) == WRITE)
 364                 bio_integrity_generate(bio);
 365
 366         return 0;
 367 }
 368 EXPORT_SYMBOL(bio_integrity_prep);
 369
 370 /**
 371  * bio_integrity_verify - Verify integrity metadata for a bio
 372  * @bio:        bio to verify
 373  *
 374  * Description: This function is called to verify the integrity of a
 375  * bio.  The data in the bio io_vec is compared to the integrity
 376  * metadata returned by the HBA.
 377  */
 378 static int bio_integrity_verify(struct bio *bio)
 379 {
 380         return bio_integrity_generate_verify(bio, 0);
 381 }
 382
 383 /**
 384  * bio_integrity_verify_fn - Integrity I/O completion worker
 385  * @work:       Work struct stored in bio to be verified
 386  *
 387  * Description: This workqueue function is called to complete a READ
 388  * request.  The function verifies the transferred integrity metadata
 389  * and then calls the original bio end_io function.
 390  */
 391 static void bio_integrity_verify_fn(struct work_struct *work)
 392 {
 393         struct bio_integrity_payload *bip =
 394                 container_of(work, struct bio_integrity_payload, bip_work);
 395         struct bio *bio = bip->bip_bio;
 396         int error;
 397
 398         error = bio_integrity_verify(bio);
 399
 400         /* Restore original bio completion handler */
 401         bio->bi_end_io = bip->bip_end_io;
 402         bio_endio_nodec(bio, error);
 403 }
 404
 405 /**
 406  * bio_integrity_endio - Integrity I/O completion function
 407  * @bio:        Protected bio
 408  * @error:      Pointer to errno
 409  *
 410  * Description: Completion for integrity I/O
 411  *
 412  * Normally I/O completion is done in interrupt context.  However,
 413  * verifying I/O integrity is a time-consuming task which must be run
 414  * in process context.  This function postpones completion
 415  * accordingly.
 416  */
 417 void bio_integrity_endio(struct bio *bio, int error)
 418 {
 419         struct bio_integrity_payload *bip = bio_integrity(bio);
 420
 421         BUG_ON(bip->bip_bio != bio);
 422
 423         /* In case of an I/O error there is no point in verifying the
 424          * integrity metadata.  Restore original bio end_io handler
 425          * and run it.
 426          */
 427         if (error) {
 428                 bio->bi_end_io = bip->bip_end_io;
 429                 bio_endio_nodec(bio, error);
 430
 431                 return;
 432         }
 433
 434         INIT_WORK(&bip->bip_work, bio_integrity_verify_fn);
 435         queue_work(kintegrityd_wq, &bip->bip_work);
 436 }
 437 EXPORT_SYMBOL(bio_integrity_endio);
 438
 439 /**
 440  * bio_integrity_advance - Advance integrity vector
 441  * @bio:        bio whose integrity vector to update
 442  * @bytes_done: number of data bytes that have been completed
 443  *
 444  * Description: This function calculates how many integrity bytes the
 445  * number of completed data bytes correspond to and advances the
 446  * integrity vector accordingly.
 447  */
 448 void bio_integrity_advance(struct bio *bio, unsigned int bytes_done)
 449 {
 450         struct bio_integrity_payload *bip = bio_integrity(bio);
 451         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 452         unsigned bytes = bio_integrity_bytes(bi, bytes_done >> 9);
 453
 454         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, bytes);
 455 }
 456 EXPORT_SYMBOL(bio_integrity_advance);
 457
 458 /**
 459  * bio_integrity_trim - Trim integrity vector
 460  * @bio:        bio whose integrity vector to update
 461  * @offset:     offset to first data sector
 462  * @sectors:    number of data sectors
 463  *
 464  * Description: Used to trim the integrity vector in a cloned bio.
 465  * The ivec will be advanced corresponding to 'offset' data sectors
 466  * and the length will be truncated corresponding to 'len' data
 467  * sectors.
 468  */
 469 void bio_integrity_trim(struct bio *bio, unsigned int offset,
 470                         unsigned int sectors)
 471 {
 472         struct bio_integrity_payload *bip = bio_integrity(bio);
 473         struct blk_integrity *bi = bdev_get_integrity(bio->bi_bdev);
 474
 475         bio_integrity_advance(bio, offset << 9);
 476         bip->bip_iter.bi_size = bio_integrity_bytes(bi, sectors);
 477 }
 478 EXPORT_SYMBOL(bio_integrity_trim);
 479
 480 /**
 481  * bio_integrity_clone - Callback for cloning bios with integrity metadata
 482  * @bio:        New bio
 483  * @bio_src:    Original bio
 484  * @gfp_mask:   Memory allocation mask
 485  *
 486  * Description: Called to allocate a bip when cloning a bio
 487  */
 488 int bio_integrity_clone(struct bio *bio, struct bio *bio_src,
 489                         gfp_t gfp_mask)
 490 {
 491         struct bio_integrity_payload *bip_src = bio_integrity(bio_src);
 492         struct bio_integrity_payload *bip;
 493
 494         BUG_ON(bip_src == NULL);
 495
 496         bip = bio_integrity_alloc(bio, gfp_mask, bip_src->bip_vcnt);
 497
 498         if (bip == NULL)
 499                 return -EIO;
 500
 501         memcpy(bip->bip_vec, bip_src->bip_vec,
 502                bip_src->bip_vcnt * sizeof(struct bio_vec));
 503
 504         bip->bip_vcnt = bip_src->bip_vcnt;
 505         bip->bip_iter = bip_src->bip_iter;
 506
 507         return 0;
 508 }
 509 EXPORT_SYMBOL(bio_integrity_clone);
 510
 511 int bioset_integrity_create(struct bio_set *bs, int pool_size)
 512 {
 513         if (bs->bio_integrity_pool)
 514                 return 0;
 515
 516         bs->bio_integrity_pool = mempool_create_slab_pool(pool_size, bip_slab);
 517         if (!bs->bio_integrity_pool)
 518                 return -1;
 519
 520         bs->bvec_integrity_pool = biovec_create_pool(pool_size);
 521         if (!bs->bvec_integrity_pool) {
 522                 mempool_destroy(bs->bio_integrity_pool);
 523                 return -1;
 524         }
 525
 526         return 0;
 527 }
 528 EXPORT_SYMBOL(bioset_integrity_create);
 529
 530 void bioset_integrity_free(struct bio_set *bs)
 531 {
 532         if (bs->bio_integrity_pool)
 533                 mempool_destroy(bs->bio_integrity_pool);
 534
 535         if (bs->bvec_integrity_pool)
 536                 mempool_destroy(bs->bvec_integrity_pool);
 537 }
 538 EXPORT_SYMBOL(bioset_integrity_free);
 539
 540 void __init bio_integrity_init(void)
 541 {
 542         /*
 543          * kintegrityd won't block much but may burn a lot of CPU cycles.
 544          * Make it highpri CPU intensive wq with max concurrency of 1.
 545          */
 546         kintegrityd_wq = alloc_workqueue("kintegrityd", WQ_MEM_RECLAIM |
 547                                          WQ_HIGHPRI | WQ_CPU_INTENSIVE, 1);
 548         if (!kintegrityd_wq)
 549                 panic("Failed to create kintegrityd\n");
 550
 551         bip_slab = kmem_cache_create("bio_integrity_payload",
 552                                      sizeof(struct bio_integrity_payload) +
 553                                      sizeof(struct bio_vec) * BIP_INLINE_VECS,
 554                                      0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 555         if (!bip_slab)
 556                 panic("Failed to create slab\n");
 557 }