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, 2015 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>
34 #include <zfs_fletcher.h>
39 * In the SPA, everything is checksummed. We support checksum vectors
40 * for three distinct reasons:
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.
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.
53 * 3. If someone develops hardware that can compute a strong hash quickly,
54 * we want the ability to take advantage of that hardware.
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.
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).
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).
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.
96 zio_checksum_off(const void *buf, uint64_t size,
97 const void *ctx_template, zio_cksum_t *zcp)
99 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
102 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
103 {{NULL, NULL}, NULL, NULL, 0, "inherit"},
104 {{NULL, NULL}, NULL, NULL, 0, "on"},
105 {{zio_checksum_off, zio_checksum_off},
106 NULL, NULL, 0, "off"},
107 {{zio_checksum_SHA256, zio_checksum_SHA256},
108 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
110 {{zio_checksum_SHA256, zio_checksum_SHA256},
111 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_EMBEDDED,
113 {{fletcher_2_native, fletcher_2_byteswap},
114 NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog"},
115 {{fletcher_2_native, fletcher_2_byteswap},
116 NULL, NULL, 0, "fletcher2"},
117 {{fletcher_4_native, fletcher_4_byteswap},
118 NULL, NULL, ZCHECKSUM_FLAG_METADATA, "fletcher4"},
119 {{zio_checksum_SHA256, zio_checksum_SHA256},
120 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
121 ZCHECKSUM_FLAG_NOPWRITE, "sha256"},
122 {{fletcher_4_native, fletcher_4_byteswap},
123 NULL, NULL, ZCHECKSUM_FLAG_EMBEDDED, "zilog2"},
124 {{zio_checksum_off, zio_checksum_off},
125 NULL, NULL, 0, "noparity"},
127 {{zio_checksum_SHA512_native, zio_checksum_SHA512_byteswap},
128 NULL, NULL, ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
129 ZCHECKSUM_FLAG_NOPWRITE, "sha512"},
130 {{zio_checksum_skein_native, zio_checksum_skein_byteswap},
131 zio_checksum_skein_tmpl_init, zio_checksum_skein_tmpl_free,
132 ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_DEDUP |
133 ZCHECKSUM_FLAG_SALTED | ZCHECKSUM_FLAG_NOPWRITE, "skein"},
134 {{zio_checksum_edonr_native, zio_checksum_edonr_byteswap},
135 zio_checksum_edonr_tmpl_init, zio_checksum_edonr_tmpl_free,
136 ZCHECKSUM_FLAG_METADATA | ZCHECKSUM_FLAG_SALTED |
137 ZCHECKSUM_FLAG_NOPWRITE, "edonr"},
142 * The flag corresponding to the "verify" in dedup=[checksum,]verify
143 * must be cleared first, so callers should use ZIO_CHECKSUM_MASK.
146 zio_checksum_to_feature(enum zio_checksum cksum)
149 VERIFY((cksum & ~ZIO_CHECKSUM_MASK) == 0);
152 case ZIO_CHECKSUM_SHA512:
153 return (SPA_FEATURE_SHA512);
154 case ZIO_CHECKSUM_SKEIN:
155 return (SPA_FEATURE_SKEIN);
156 case ZIO_CHECKSUM_EDONR:
157 return (SPA_FEATURE_EDONR);
160 return (SPA_FEATURE_NONE);
164 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
166 ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
167 ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
168 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
170 if (child == ZIO_CHECKSUM_INHERIT)
173 if (child == ZIO_CHECKSUM_ON)
174 return (ZIO_CHECKSUM_ON_VALUE);
180 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
181 enum zio_checksum parent)
183 ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
184 ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
185 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
187 if (child == ZIO_CHECKSUM_INHERIT)
190 if (child == ZIO_CHECKSUM_ON)
191 return (spa_dedup_checksum(spa));
193 if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
194 return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
196 ASSERT((zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_flags &
197 ZCHECKSUM_FLAG_DEDUP) ||
198 (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
204 * Set the external verifier for a gang block based on <vdev, offset, txg>,
205 * a tuple which is guaranteed to be unique for the life of the pool.
208 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
210 dva_t *dva = BP_IDENTITY(bp);
211 uint64_t txg = BP_PHYSICAL_BIRTH(bp);
213 ASSERT(BP_IS_GANG(bp));
215 ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
219 * Set the external verifier for a label block based on its offset.
220 * The vdev is implicit, and the txg is unknowable at pool open time --
221 * hence the logic in vdev_uberblock_load() to find the most recent copy.
224 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
226 ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
230 * Calls the template init function of a checksum which supports context
231 * templates and installs the template into the spa_t.
234 zio_checksum_template_init(enum zio_checksum checksum, spa_t *spa)
236 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
238 if (ci->ci_tmpl_init == NULL)
240 if (spa->spa_cksum_tmpls[checksum] != NULL)
243 VERIFY(ci->ci_tmpl_free != NULL);
244 mutex_enter(&spa->spa_cksum_tmpls_lock);
245 if (spa->spa_cksum_tmpls[checksum] == NULL) {
246 spa->spa_cksum_tmpls[checksum] =
247 ci->ci_tmpl_init(&spa->spa_cksum_salt);
248 VERIFY(spa->spa_cksum_tmpls[checksum] != NULL);
250 mutex_exit(&spa->spa_cksum_tmpls_lock);
254 * Generate the checksum.
257 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
258 void *data, uint64_t size)
260 blkptr_t *bp = zio->io_bp;
261 uint64_t offset = zio->io_offset;
262 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
264 spa_t *spa = zio->io_spa;
266 ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
267 ASSERT(ci->ci_func[0] != NULL);
269 zio_checksum_template_init(checksum, spa);
271 if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
274 if (checksum == ZIO_CHECKSUM_ZILOG2) {
275 zil_chain_t *zilc = data;
277 size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
281 eck = (zio_eck_t *)((char *)data + size) - 1;
283 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
284 zio_checksum_gang_verifier(&eck->zec_cksum, bp);
285 else if (checksum == ZIO_CHECKSUM_LABEL)
286 zio_checksum_label_verifier(&eck->zec_cksum, offset);
288 bp->blk_cksum = eck->zec_cksum;
289 eck->zec_magic = ZEC_MAGIC;
290 ci->ci_func[0](data, size, spa->spa_cksum_tmpls[checksum],
292 eck->zec_cksum = cksum;
294 ci->ci_func[0](data, size, spa->spa_cksum_tmpls[checksum],
300 zio_checksum_error_impl(spa_t *spa, blkptr_t *bp, enum zio_checksum checksum,
301 void *data, uint64_t size, uint64_t offset, zio_bad_cksum_t *info)
303 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
304 zio_cksum_t actual_cksum, expected_cksum;
307 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
308 return (SET_ERROR(EINVAL));
310 zio_checksum_template_init(checksum, spa);
312 if (ci->ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
314 zio_cksum_t verifier;
316 if (checksum == ZIO_CHECKSUM_ZILOG2) {
317 zil_chain_t *zilc = data;
321 if (eck->zec_magic == ZEC_MAGIC)
322 nused = zilc->zc_nused;
323 else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC))
324 nused = BSWAP_64(zilc->zc_nused);
326 return (SET_ERROR(ECKSUM));
329 return (SET_ERROR(ECKSUM));
331 size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
333 eck = (zio_eck_t *)((char *)data + size) - 1;
336 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
337 zio_checksum_gang_verifier(&verifier, bp);
338 else if (checksum == ZIO_CHECKSUM_LABEL)
339 zio_checksum_label_verifier(&verifier, offset);
341 verifier = bp->blk_cksum;
343 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
346 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
348 expected_cksum = eck->zec_cksum;
349 eck->zec_cksum = verifier;
350 ci->ci_func[byteswap](data, size,
351 spa->spa_cksum_tmpls[checksum], &actual_cksum);
352 eck->zec_cksum = expected_cksum;
355 byteswap_uint64_array(&expected_cksum,
356 sizeof (zio_cksum_t));
359 byteswap = BP_SHOULD_BYTESWAP(bp);
360 expected_cksum = bp->blk_cksum;
361 ci->ci_func[byteswap](data, size,
362 spa->spa_cksum_tmpls[checksum], &actual_cksum);
366 info->zbc_expected = expected_cksum;
367 info->zbc_actual = actual_cksum;
368 info->zbc_checksum_name = ci->ci_name;
369 info->zbc_byteswapped = byteswap;
370 info->zbc_injected = 0;
371 info->zbc_has_cksum = 1;
374 if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
375 return (SET_ERROR(ECKSUM));
381 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
383 blkptr_t *bp = zio->io_bp;
384 uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
385 (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
387 uint64_t size = (bp == NULL ? zio->io_size :
388 (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
389 uint64_t offset = zio->io_offset;
390 void *data = zio->io_data;
391 spa_t *spa = zio->io_spa;
393 error = zio_checksum_error_impl(spa, bp, checksum, data, size,
395 if (error != 0 && zio_injection_enabled && !zio->io_error &&
396 (error = zio_handle_fault_injection(zio, ECKSUM)) != 0) {
398 info->zbc_injected = 1;
405 * Called by a spa_t that's about to be deallocated. This steps through
406 * all of the checksum context templates and deallocates any that were
407 * initialized using the algorithm-specific template init function.
410 zio_checksum_templates_free(spa_t *spa)
412 for (enum zio_checksum checksum = 0;
413 checksum < ZIO_CHECKSUM_FUNCTIONS; checksum++) {
414 if (spa->spa_cksum_tmpls[checksum] != NULL) {
415 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
417 VERIFY(ci->ci_tmpl_free != NULL);
418 ci->ci_tmpl_free(spa->spa_cksum_tmpls[checksum]);
419 spa->spa_cksum_tmpls[checksum] = NULL;