module/zfs/zio_checksum.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
  24  * Copyright 2013 Saso Kiselkov. All rights reserved.
  25  */
  26
  27 #include <sys/zfs_context.h>
  28 #include <sys/spa.h>
  29 #include <sys/spa_impl.h>
  30 #include <sys/zio.h>
  31 #include <sys/zio_checksum.h>
  32 #include <sys/zil.h>
  33 #include <sys/abd.h>
  34 #include <zfs_fletcher.h>
  35
  36 /*
  37  * Checksum vectors.
  38  *
  39  * In the SPA, everything is checksummed.  We support checksum vectors
  40  * for three distinct reasons:
  41  *
  42  *   1. Different kinds of data need different levels of protection.
  43  *      For SPA metadata, we always want a very strong checksum.
  44  *      For user data, we let users make the trade-off between speed
  45  *      and checksum strength.
  46  *
  47  *   2. Cryptographic hash and MAC algorithms are an area of active research.
  48  *      It is likely that in future hash functions will be at least as strong
  49  *      as current best-of-breed, and may be substantially faster as well.
  50  *      We want the ability to take advantage of these new hashes as soon as
  51  *      they become available.
  52  *
  53  *   3. If someone develops hardware that can compute a strong hash quickly,
  54  *      we want the ability to take advantage of that hardware.
  55  *
  56  * Of course, we don't want a checksum upgrade to invalidate existing
  57  * data, so we store the checksum *function* in eight bits of the bp.
  58  * This gives us room for up to 256 different checksum functions.
  59  *
  60  * When writing a block, we always checksum it with the latest-and-greatest
  61  * checksum function of the appropriate strength.  When reading a block,
  62  * we compare the expected checksum against the actual checksum, which we
  63  * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
  64  *
  65  * SALTED CHECKSUMS
  66  *
  67  * To enable the use of less secure hash algorithms with dedup, we
  68  * introduce the notion of salted checksums (MACs, really).  A salted
  69  * checksum is fed both a random 256-bit value (the salt) and the data
  70  * to be checksummed.  This salt is kept secret (stored on the pool, but
  71  * never shown to the user).  Thus even if an attacker knew of collision
  72  * weaknesses in the hash algorithm, they won't be able to mount a known
  73  * plaintext attack on the DDT, since the actual hash value cannot be
  74  * known ahead of time.  How the salt is used is algorithm-specific
  75  * (some might simply prefix it to the data block, others might need to
  76  * utilize a full-blown HMAC).  On disk the salt is stored in a ZAP
  77  * object in the MOS (DMU_POOL_CHECKSUM_SALT).
  78  *
  79  * CONTEXT TEMPLATES
  80  *
  81  * Some hashing algorithms need to perform a substantial amount of
  82  * initialization work (e.g. salted checksums above may need to pre-hash
  83  * the salt) before being able to process data.  Performing this
  84  * redundant work for each block would be wasteful, so we instead allow
  85  * a checksum algorithm to do the work once (the first time it's used)
  86  * and then keep this pre-initialized context as a template inside the
  87  * spa_t (spa_cksum_tmpls).  If the zio_checksum_info_t contains
  88  * non-NULL ci_tmpl_init and ci_tmpl_free callbacks, they are used to
  89  * construct and destruct the pre-initialized checksum context.  The
  90  * pre-initialized context is then reused during each checksum
  91  * invocation and passed to the checksum function.
  92  */
  93
  94 /*ARGSUSED*/
  95 static void
  96 abd_checksum_off(abd_t *abd, uint64_t size,
  97     const void *ctx_template, zio_cksum_t *zcp)
  98 {
  99         ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
 100 }
 101
 102 /*ARGSUSED*/
 103 void
 104 abd_fletcher_2_native(abd_t *abd, uint64_t size,
 105     const void *ctx_template, zio_cksum_t *zcp)
 106 {
 107         fletcher_init(zcp);
 108         (void) abd_iterate_func(abd, 0, size,
 109             fletcher_2_incremental_native, zcp);
 110 }
 111
 112 /*ARGSUSED*/
 113 void
 114 abd_fletcher_2_byteswap(abd_t *abd, uint64_t size,
 115     const void *ctx_template, zio_cksum_t *zcp)
 116 {
 117         fletcher_init(zcp);
 118         (void) abd_iterate_func(abd, 0, size,
 119             fletcher_2_incremental_byteswap, zcp);
 120 }
 121
 122 /*ARGSUSED*/
 123 void
 124 abd_fletcher_4_native(abd_t *abd, uint64_t size,
 125     const void *ctx_template, zio_cksum_t *zcp)
 126 {
 127         fletcher_init(zcp);
 128         (void) abd_iterate_func(abd, 0, size,
 129             fletcher_4_incremental_native, zcp);
 130 }
 131
 132 /*ARGSUSED*/
 133 void
 134 abd_fletcher_4_byteswap(abd_t *abd, uint64_t size,
 135     const void *ctx_template, zio_cksum_t *zcp)
 136 {
 137         fletcher_init(zcp);
 138         (void) abd_iterate_func(abd, 0, size,
 139             fletcher_4_incremental_byteswap, zcp);
 140 }
 141
 142 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
 143         {{NULL, NULL}, NULL, NULL, 0, "inherit"},
 144         {{NULL, NULL}, NULL, NULL, 0, "on"},
 145         {{abd_checksum_off,             abd_checksum_off},
 146             NULL, NULL, 0, "off"},
 147         {{abd_checksum_SHA256,          abd_checksum_SHA256},
 148             NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
 149             "label"},
 150         {{abd_checksum_SHA256,          abd_checksum_SHA256},
 151             NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
 152             "gang_header"},
 153         {{abd_fletcher_2_native,        abd_fletcher_2_byteswap},
 154             NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog"},
 155         {{abd_fletcher_2_native,        abd_fletcher_2_byteswap},
 156             NULL, NULL, 0, "fletcher2"},
 157         {{abd_fletcher_4_native,        abd_fletcher_4_byteswap},
 158             NULL, NULL, ZCHECKSUM_FLAG_METADATA, "fletcher4"},
 159         {{abd_checksum_SHA256,          abd_checksum_SHA256},
 160             NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
 161             ZCHECKSUM_FLAG_NOPWRITE, "sha256"},
 162         {{abd_fletcher_4_native,        abd_fletcher_4_byteswap},
 163             NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog2"},
 164         {{abd_checksum_off,             abd_checksum_off},
 165             NULL, NULL, 0, "noparity"},
 166         {{abd_checksum_SHA512_native,   abd_checksum_SHA512_byteswap},
 167             NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
 168             ZCHECKSUM_FLAG_NOPWRITE, "sha512"},
 169         {{abd_checksum_skein_native,    abd_checksum_skein_byteswap},
 170             abd_checksum_skein_tmpl_init, abd_checksum_skein_tmpl_free,
 171             ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
 172             ZCHECKSUM_FLAG_SALTED | ZCHECKSUM_FLAG_NOPWRITE, "skein"},
 173         {{abd_checksum_edonr_native,    abd_checksum_edonr_byteswap},
 174             abd_checksum_edonr_tmpl_init, abd_checksum_edonr_tmpl_free,
 175             ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_SALTED |
 176             ZCHECKSUM_FLAG_NOPWRITE, "edonr"},
 177 };
 178
 179 /*
 180  * The flag corresponding to the "verify" in dedup=[checksum,]verify
 181  * must be cleared first, so callers should use ZIO_CHECKSUM_MASK.
 182  */
 183 spa_feature_t
 184 zio_checksum_to_feature(enum zio_checksum cksum)
 185 {
 186         VERIFY((cksum & ~ZIO_CHECKSUM_MASK) == 0);
 187
 188         switch (cksum) {
 189         case ZIO_CHECKSUM_SHA512:
 190                 return (SPA_FEATURE_SHA512);
 191         case ZIO_CHECKSUM_SKEIN:
 192                 return (SPA_FEATURE_SKEIN);
 193         case ZIO_CHECKSUM_EDONR:
 194                 return (SPA_FEATURE_EDONR);
 195         default:
 196                 return (SPA_FEATURE_NONE);
 197         }
 198 }
 199
 200 enum zio_checksum
 201 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
 202 {
 203         ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
 204         ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
 205         ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
 206
 207         if (child == ZIO_CHECKSUM_INHERIT)
 208                 return (parent);
 209
 210         if (child == ZIO_CHECKSUM_ON)
 211                 return (ZIO_CHECKSUM_ON_VALUE);
 212
 213         return (child);
 214 }
 215
 216 enum zio_checksum
 217 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
 218     enum zio_checksum parent)
 219 {
 220         ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
 221         ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
 222         ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
 223
 224         if (child == ZIO_CHECKSUM_INHERIT)
 225                 return (parent);
 226
 227         if (child == ZIO_CHECKSUM_ON)
 228                 return (spa_dedup_checksum(spa));
 229
 230         if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
 231                 return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
 232
 233         ASSERT((zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_flags &
 234             ZCHECKSUM_FLAG_DEDUP) ||
 235             (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
 236
 237         return (child);
 238 }
 239
 240 /*
 241  * Set the external verifier for a gang block based on <vdev, offset, txg>,
 242  * a tuple which is guaranteed to be unique for the life of the pool.
 243  */
 244 static void
 245 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
 246 {
 247         const dva_t *dva = BP_IDENTITY(bp);
 248         uint64_t txg = BP_PHYSICAL_BIRTH(bp);
 249
 250         ASSERT(BP_IS_GANG(bp));
 251
 252         ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
 253 }
 254
 255 /*
 256  * Set the external verifier for a label block based on its offset.
 257  * The vdev is implicit, and the txg is unknowable at pool open time --
 258  * hence the logic in vdev_uberblock_load() to find the most recent copy.
 259  */
 260 static void
 261 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
 262 {
 263         ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
 264 }
 265
 266 /*
 267  * Calls the template init function of a checksum which supports context
 268  * templates and installs the template into the spa_t.
 269  */
 270 static void
 271 zio_checksum_template_init(enum zio_checksum checksum, spa_t *spa)
 272 {
 273         zio_checksum_info_t *ci = &zio_checksum_table[checksum];
 274
 275         if (ci->ci_tmpl_init == NULL)
 276                 return;
 277         if (spa->spa_cksum_tmpls[checksum] != NULL)
 278                 return;
 279
 280         VERIFY(ci->ci_tmpl_free != NULL);
 281         mutex_enter(&spa->spa_cksum_tmpls_lock);
 282         if (spa->spa_cksum_tmpls[checksum] == NULL) {
 283                 spa->spa_cksum_tmpls[checksum] =
 284                     ci->ci_tmpl_init(&spa->spa_cksum_salt);
 285                 VERIFY(spa->spa_cksum_tmpls[checksum] != NULL);
 286         }
 287         mutex_exit(&spa->spa_cksum_tmpls_lock);
 288 }
 289
 290 /*
 291  * Generate the checksum.
 292  */
 293 void
 294 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
 295     abd_t *abd, uint64_t size)
 296 {
 297         blkptr_t *bp = zio->io_bp;
 298         uint64_t offset = zio->io_offset;
 299         zio_checksum_info_t *ci = &zio_checksum_table[checksum];
 300         zio_cksum_t cksum;
 301         spa_t *spa = zio->io_spa;
 302
 303         ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
 304         ASSERT(ci->ci_func[0] != NULL);
 305
 306         zio_checksum_template_init(checksum, spa);
 307
 308         if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
 309                 zio_eck_t *eck;
 310                 void *data = abd_to_buf(abd);
 311
 312                 if (checksum == ZIO_CHECKSUM_ZILOG2) {
 313                         zil_chain_t *zilc = data;
 314
 315                         size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
 316                             uint64_t);
 317                         eck = &zilc->zc_eck;
 318                 } else {
 319                         eck = (zio_eck_t *)((char *)data + size) - 1;
 320                 }
 321                 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
 322                         zio_checksum_gang_verifier(&eck->zec_cksum, bp);
 323                 else if (checksum == ZIO_CHECKSUM_LABEL)
 324                         zio_checksum_label_verifier(&eck->zec_cksum, offset);
 325                 else
 326                         bp->blk_cksum = eck->zec_cksum;
 327                 eck->zec_magic = ZEC_MAGIC;
 328                 ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
 329                     &cksum);
 330                 eck->zec_cksum = cksum;
 331         } else {
 332                 ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
 333                     &bp->blk_cksum);
 334         }
 335 }
 336
 337 int
 338 zio_checksum_error_impl(spa_t *spa, blkptr_t *bp, enum zio_checksum checksum,
 339     abd_t *abd, uint64_t size, uint64_t offset, zio_bad_cksum_t *info)
 340 {
 341         zio_checksum_info_t *ci = &zio_checksum_table[checksum];
 342         int byteswap;
 343         zio_cksum_t actual_cksum, expected_cksum;
 344
 345         if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
 346                 return (SET_ERROR(EINVAL));
 347
 348         zio_checksum_template_init(checksum, spa);
 349
 350         if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
 351                 zio_eck_t *eck;
 352                 zio_cksum_t verifier;
 353                 size_t eck_offset;
 354                 uint64_t data_size = size;
 355                 void *data = abd_borrow_buf_copy(abd, data_size);
 356
 357                 if (checksum == ZIO_CHECKSUM_ZILOG2) {
 358                         zil_chain_t *zilc = data;
 359                         uint64_t nused;
 360
 361                         eck = &zilc->zc_eck;
 362                         if (eck->zec_magic == ZEC_MAGIC) {
 363                                 nused = zilc->zc_nused;
 364                         } else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC)) {
 365                                 nused = BSWAP_64(zilc->zc_nused);
 366                         } else {
 367                                 abd_return_buf(abd, data, data_size);
 368                                 return (SET_ERROR(ECKSUM));
 369                         }
 370
 371                         if (nused > data_size) {
 372                                 abd_return_buf(abd, data, data_size);
 373                                 return (SET_ERROR(ECKSUM));
 374                         }
 375
 376                         size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
 377                 } else {
 378                         eck = (zio_eck_t *)((char *)data + data_size) - 1;
 379                 }
 380
 381                 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
 382                         zio_checksum_gang_verifier(&verifier, bp);
 383                 else if (checksum == ZIO_CHECKSUM_LABEL)
 384                         zio_checksum_label_verifier(&verifier, offset);
 385                 else
 386                         verifier = bp->blk_cksum;
 387
 388                 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
 389
 390                 if (byteswap)
 391                         byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
 392
 393                 eck_offset = (size_t)(&eck->zec_cksum) - (size_t)data;
 394                 expected_cksum = eck->zec_cksum;
 395                 eck->zec_cksum = verifier;
 396                 abd_return_buf_copy(abd, data, data_size);
 397
 398                 ci->ci_func[byteswap](abd, size,
 399                     spa->spa_cksum_tmpls[checksum], &actual_cksum);
 400                 abd_copy_from_buf_off(abd, &expected_cksum,
 401                     eck_offset, sizeof (zio_cksum_t));
 402
 403                 if (byteswap) {
 404                         byteswap_uint64_array(&expected_cksum,
 405                             sizeof (zio_cksum_t));
 406                 }
 407         } else {
 408                 byteswap = BP_SHOULD_BYTESWAP(bp);
 409                 expected_cksum = bp->blk_cksum;
 410                 ci->ci_func[byteswap](abd, size,
 411                     spa->spa_cksum_tmpls[checksum], &actual_cksum);
 412         }
 413
 414         if (info != NULL) {
 415                 info->zbc_expected = expected_cksum;
 416                 info->zbc_actual = actual_cksum;
 417                 info->zbc_checksum_name = ci->ci_name;
 418                 info->zbc_byteswapped = byteswap;
 419                 info->zbc_injected = 0;
 420                 info->zbc_has_cksum = 1;
 421         }
 422
 423         if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
 424                 return (SET_ERROR(ECKSUM));
 425
 426         return (0);
 427 }
 428
 429 int
 430 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
 431 {
 432         blkptr_t *bp = zio->io_bp;
 433         uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
 434             (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
 435         int error;
 436         uint64_t size = (bp == NULL ? zio->io_size :
 437             (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
 438         uint64_t offset = zio->io_offset;
 439         abd_t *data = zio->io_abd;
 440         spa_t *spa = zio->io_spa;
 441
 442         error = zio_checksum_error_impl(spa, bp, checksum, data, size,
 443             offset, info);
 444
 445         if (zio_injection_enabled && error == 0 && zio->io_error == 0) {
 446                 error = zio_handle_fault_injection(zio, ECKSUM);
 447                 if (error != 0)
 448                         info->zbc_injected = 1;
 449         }
 450
 451         return (error);
 452 }
 453
 454 /*
 455  * Called by a spa_t that's about to be deallocated. This steps through
 456  * all of the checksum context templates and deallocates any that were
 457  * initialized using the algorithm-specific template init function.
 458  */
 459 void
 460 zio_checksum_templates_free(spa_t *spa)
 461 {
 462         enum zio_checksum checksum;
 463         for (checksum = 0; checksum < ZIO_CHECKSUM_FUNCTIONS;
 464             checksum++) {
 465                 if (spa->spa_cksum_tmpls[checksum] != NULL) {
 466                         zio_checksum_info_t *ci = &zio_checksum_table[checksum];
 467
 468                         VERIFY(ci->ci_tmpl_free != NULL);
 469                         ci->ci_tmpl_free(spa->spa_cksum_tmpls[checksum]);
 470                         spa->spa_cksum_tmpls[checksum] = NULL;
 471                 }
 472         }
 473 }