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.
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.
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]
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 (c) 2013, Joyent, Inc. All rights reserved.
25 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
32 #include <sys/zio_checksum.h>
35 #include <zfs_fletcher.h>
40 * In the SPA, everything is checksummed. We support checksum vectors
41 * for three distinct reasons:
43 * 1. Different kinds of data need different levels of protection.
44 * For SPA metadata, we always want a very strong checksum.
45 * For user data, we let users make the trade-off between speed
46 * and checksum strength.
48 * 2. Cryptographic hash and MAC algorithms are an area of active research.
49 * It is likely that in future hash functions will be at least as strong
50 * as current best-of-breed, and may be substantially faster as well.
51 * We want the ability to take advantage of these new hashes as soon as
52 * they become available.
54 * 3. If someone develops hardware that can compute a strong hash quickly,
55 * we want the ability to take advantage of that hardware.
57 * Of course, we don't want a checksum upgrade to invalidate existing
58 * data, so we store the checksum *function* in eight bits of the bp.
59 * This gives us room for up to 256 different checksum functions.
61 * When writing a block, we always checksum it with the latest-and-greatest
62 * checksum function of the appropriate strength. When reading a block,
63 * we compare the expected checksum against the actual checksum, which we
64 * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
68 * To enable the use of less secure hash algorithms with dedup, we
69 * introduce the notion of salted checksums (MACs, really). A salted
70 * checksum is fed both a random 256-bit value (the salt) and the data
71 * to be checksummed. This salt is kept secret (stored on the pool, but
72 * never shown to the user). Thus even if an attacker knew of collision
73 * weaknesses in the hash algorithm, they won't be able to mount a known
74 * plaintext attack on the DDT, since the actual hash value cannot be
75 * known ahead of time. How the salt is used is algorithm-specific
76 * (some might simply prefix it to the data block, others might need to
77 * utilize a full-blown HMAC). On disk the salt is stored in a ZAP
78 * object in the MOS (DMU_POOL_CHECKSUM_SALT).
82 * Some hashing algorithms need to perform a substantial amount of
83 * initialization work (e.g. salted checksums above may need to pre-hash
84 * the salt) before being able to process data. Performing this
85 * redundant work for each block would be wasteful, so we instead allow
86 * a checksum algorithm to do the work once (the first time it's used)
87 * and then keep this pre-initialized context as a template inside the
88 * spa_t (spa_cksum_tmpls). If the zio_checksum_info_t contains
89 * non-NULL ci_tmpl_init and ci_tmpl_free callbacks, they are used to
90 * construct and destruct the pre-initialized checksum context. The
91 * pre-initialized context is then reused during each checksum
92 * invocation and passed to the checksum function.
97 abd_checksum_off(abd_t *abd, uint64_t size,
98 const void *ctx_template, zio_cksum_t *zcp)
100 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
105 abd_fletcher_2_native(abd_t *abd, uint64_t size,
106 const void *ctx_template, zio_cksum_t *zcp)
109 (void) abd_iterate_func(abd, 0, size,
110 fletcher_2_incremental_native, zcp);
115 abd_fletcher_2_byteswap(abd_t *abd, uint64_t size,
116 const void *ctx_template, zio_cksum_t *zcp)
119 (void) abd_iterate_func(abd, 0, size,
120 fletcher_2_incremental_byteswap, zcp);
125 abd_fletcher_4_native(abd_t *abd, uint64_t size,
126 const void *ctx_template, zio_cksum_t *zcp)
129 (void) abd_iterate_func(abd, 0, size,
130 fletcher_4_incremental_native, zcp);
135 abd_fletcher_4_byteswap(abd_t *abd, uint64_t size,
136 const void *ctx_template, zio_cksum_t *zcp)
139 (void) abd_iterate_func(abd, 0, size,
140 fletcher_4_incremental_byteswap, zcp);
143 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
144 {{NULL, NULL}, NULL, NULL, 0, "inherit"},
145 {{NULL, NULL}, NULL, NULL, 0, "on"},
146 {{abd_checksum_off, abd_checksum_off},
147 NULL, NULL, 0, "off"},
148 {{abd_checksum_SHA256, abd_checksum_SHA256},
149 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
151 {{abd_checksum_SHA256, abd_checksum_SHA256},
152 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
154 {{abd_fletcher_2_native, abd_fletcher_2_byteswap},
155 NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog"},
156 {{abd_fletcher_2_native, abd_fletcher_2_byteswap},
157 NULL, NULL, 0, "fletcher2"},
158 {{abd_fletcher_4_native, abd_fletcher_4_byteswap},
159 NULL, NULL, ZCHECKSUM_FLAG_METADATA, "fletcher4"},
160 {{abd_checksum_SHA256, abd_checksum_SHA256},
161 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
162 ZCHECKSUM_FLAG_NOPWRITE, "sha256"},
163 {{abd_fletcher_4_native, abd_fletcher_4_byteswap},
164 NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog2"},
165 {{abd_checksum_off, abd_checksum_off},
166 NULL, NULL, 0, "noparity"},
167 {{abd_checksum_SHA512_native, abd_checksum_SHA512_byteswap},
168 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
169 ZCHECKSUM_FLAG_NOPWRITE, "sha512"},
170 {{abd_checksum_skein_native, abd_checksum_skein_byteswap},
171 abd_checksum_skein_tmpl_init, abd_checksum_skein_tmpl_free,
172 ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
173 ZCHECKSUM_FLAG_SALTED | ZCHECKSUM_FLAG_NOPWRITE, "skein"},
175 {{abd_checksum_edonr_native, abd_checksum_edonr_byteswap},
176 abd_checksum_edonr_tmpl_init, abd_checksum_edonr_tmpl_free,
177 ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_SALTED |
178 ZCHECKSUM_FLAG_NOPWRITE, "edonr"},
183 * The flag corresponding to the "verify" in dedup=[checksum,]verify
184 * must be cleared first, so callers should use ZIO_CHECKSUM_MASK.
187 zio_checksum_to_feature(enum zio_checksum cksum)
189 VERIFY((cksum & ~ZIO_CHECKSUM_MASK) == 0);
192 case ZIO_CHECKSUM_SHA512:
193 return (SPA_FEATURE_SHA512);
194 case ZIO_CHECKSUM_SKEIN:
195 return (SPA_FEATURE_SKEIN);
197 case ZIO_CHECKSUM_EDONR:
198 return (SPA_FEATURE_EDONR);
201 return (SPA_FEATURE_NONE);
205 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
207 ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
208 ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
209 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
211 if (child == ZIO_CHECKSUM_INHERIT)
214 if (child == ZIO_CHECKSUM_ON)
215 return (ZIO_CHECKSUM_ON_VALUE);
221 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
222 enum zio_checksum parent)
224 ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
225 ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
226 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
228 if (child == ZIO_CHECKSUM_INHERIT)
231 if (child == ZIO_CHECKSUM_ON)
232 return (spa_dedup_checksum(spa));
234 if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
235 return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
237 ASSERT((zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_flags &
238 ZCHECKSUM_FLAG_DEDUP) ||
239 (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
245 * Set the external verifier for a gang block based on <vdev, offset, txg>,
246 * a tuple which is guaranteed to be unique for the life of the pool.
249 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
251 dva_t *dva = BP_IDENTITY(bp);
252 uint64_t txg = BP_PHYSICAL_BIRTH(bp);
254 ASSERT(BP_IS_GANG(bp));
256 ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
260 * Set the external verifier for a label block based on its offset.
261 * The vdev is implicit, and the txg is unknowable at pool open time --
262 * hence the logic in vdev_uberblock_load() to find the most recent copy.
265 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
267 ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
271 * Calls the template init function of a checksum which supports context
272 * templates and installs the template into the spa_t.
275 zio_checksum_template_init(enum zio_checksum checksum, spa_t *spa)
277 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
279 if (ci->ci_tmpl_init == NULL)
281 if (spa->spa_cksum_tmpls[checksum] != NULL)
284 VERIFY(ci->ci_tmpl_free != NULL);
285 mutex_enter(&spa->spa_cksum_tmpls_lock);
286 if (spa->spa_cksum_tmpls[checksum] == NULL) {
287 spa->spa_cksum_tmpls[checksum] =
288 ci->ci_tmpl_init(&spa->spa_cksum_salt);
289 VERIFY(spa->spa_cksum_tmpls[checksum] != NULL);
291 mutex_exit(&spa->spa_cksum_tmpls_lock);
295 * Generate the checksum.
298 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
299 abd_t *abd, uint64_t size)
301 blkptr_t *bp = zio->io_bp;
302 uint64_t offset = zio->io_offset;
303 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
305 spa_t *spa = zio->io_spa;
307 ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
308 ASSERT(ci->ci_func[0] != NULL);
310 zio_checksum_template_init(checksum, spa);
312 if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
314 void *data = abd_to_buf(abd);
316 if (checksum == ZIO_CHECKSUM_ZILOG2) {
317 zil_chain_t *zilc = data;
319 size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
323 eck = (zio_eck_t *)((char *)data + size) - 1;
325 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
326 zio_checksum_gang_verifier(&eck->zec_cksum, bp);
327 else if (checksum == ZIO_CHECKSUM_LABEL)
328 zio_checksum_label_verifier(&eck->zec_cksum, offset);
330 bp->blk_cksum = eck->zec_cksum;
331 eck->zec_magic = ZEC_MAGIC;
332 ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
334 eck->zec_cksum = cksum;
336 ci->ci_func[0](abd, size, spa->spa_cksum_tmpls[checksum],
342 zio_checksum_error_impl(spa_t *spa, blkptr_t *bp, enum zio_checksum checksum,
343 abd_t *abd, uint64_t size, uint64_t offset, zio_bad_cksum_t *info)
345 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
346 zio_cksum_t actual_cksum, expected_cksum;
349 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
350 return (SET_ERROR(EINVAL));
352 zio_checksum_template_init(checksum, spa);
354 if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
356 zio_cksum_t verifier;
357 uint64_t data_size = size;
358 void *data = abd_borrow_buf_copy(abd, data_size);
360 if (checksum == ZIO_CHECKSUM_ZILOG2) {
361 zil_chain_t *zilc = data;
365 if (eck->zec_magic == ZEC_MAGIC) {
366 nused = zilc->zc_nused;
367 } else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC)) {
368 nused = BSWAP_64(zilc->zc_nused);
370 abd_return_buf(abd, data, data_size);
371 return (SET_ERROR(ECKSUM));
374 if (nused > data_size) {
375 abd_return_buf(abd, data, data_size);
376 return (SET_ERROR(ECKSUM));
379 size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
381 eck = (zio_eck_t *)((char *)data + data_size) - 1;
384 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
385 zio_checksum_gang_verifier(&verifier, bp);
386 else if (checksum == ZIO_CHECKSUM_LABEL)
387 zio_checksum_label_verifier(&verifier, offset);
389 verifier = bp->blk_cksum;
391 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
394 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
396 size_t eck_offset = (size_t)(&eck->zec_cksum) - (size_t)data;
397 expected_cksum = eck->zec_cksum;
398 eck->zec_cksum = verifier;
399 abd_return_buf_copy(abd, data, data_size);
401 ci->ci_func[byteswap](abd, size,
402 spa->spa_cksum_tmpls[checksum], &actual_cksum);
403 abd_copy_from_buf_off(abd, &expected_cksum,
404 eck_offset, sizeof (zio_cksum_t));
407 byteswap_uint64_array(&expected_cksum,
408 sizeof (zio_cksum_t));
411 byteswap = BP_SHOULD_BYTESWAP(bp);
412 expected_cksum = bp->blk_cksum;
413 ci->ci_func[byteswap](abd, size,
414 spa->spa_cksum_tmpls[checksum], &actual_cksum);
418 info->zbc_expected = expected_cksum;
419 info->zbc_actual = actual_cksum;
420 info->zbc_checksum_name = ci->ci_name;
421 info->zbc_byteswapped = byteswap;
422 info->zbc_injected = 0;
423 info->zbc_has_cksum = 1;
426 if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
427 return (SET_ERROR(ECKSUM));
433 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
435 blkptr_t *bp = zio->io_bp;
436 uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
437 (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
439 uint64_t size = (bp == NULL ? zio->io_size :
440 (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
441 uint64_t offset = zio->io_offset;
442 abd_t *data = zio->io_abd;
443 spa_t *spa = zio->io_spa;
445 error = zio_checksum_error_impl(spa, bp, checksum, data, size,
448 if (zio_injection_enabled && error == 0 && zio->io_error == 0) {
449 error = zio_handle_fault_injection(zio, ECKSUM);
451 info->zbc_injected = 1;
458 * Called by a spa_t that's about to be deallocated. This steps through
459 * all of the checksum context templates and deallocates any that were
460 * initialized using the algorithm-specific template init function.
463 zio_checksum_templates_free(spa_t *spa)
465 for (enum zio_checksum checksum = 0;
466 checksum < ZIO_CHECKSUM_FUNCTIONS; checksum++) {
467 if (spa->spa_cksum_tmpls[checksum] != NULL) {
468 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
470 VERIFY(ci->ci_tmpl_free != NULL);
471 ci->ci_tmpl_free(spa->spa_cksum_tmpls[checksum]);
472 spa->spa_cksum_tmpls[checksum] = NULL;