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 */
24
25 #include <sys/zfs_context.h>
26 #include <sys/spa.h>
27 #include <sys/zio.h>
28 #include <sys/zio_checksum.h>
29 #include <sys/zil.h>
30 #include <zfs_fletcher.h>
31
32 /*
33 * Checksum vectors.
34 *
35 * In the SPA, everything is checksummed. We support checksum vectors
36 * for three distinct reasons:
37 *
38 * 1. Different kinds of data need different levels of protection.
39 * For SPA metadata, we always want a very strong checksum.
40 * For user data, we let users make the trade-off between speed
41 * and checksum strength.
42 *
43 * 2. Cryptographic hash and MAC algorithms are an area of active research.
44 * It is likely that in future hash functions will be at least as strong
45 * as current best-of-breed, and may be substantially faster as well.
46 * We want the ability to take advantage of these new hashes as soon as
47 * they become available.
48 *
49 * 3. If someone develops hardware that can compute a strong hash quickly,
50 * we want the ability to take advantage of that hardware.
51 *
52 * Of course, we don't want a checksum upgrade to invalidate existing
53 * data, so we store the checksum *function* in eight bits of the bp.
54 * This gives us room for up to 256 different checksum functions.
55 *
56 * When writing a block, we always checksum it with the latest-and-greatest
57 * checksum function of the appropriate strength. When reading a block,
58 * we compare the expected checksum against the actual checksum, which we
59 * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
60 */
61
62 /*ARGSUSED*/
63 static void
64 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp)
65 {
66 ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
67 }
68
69 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
70 {{NULL, NULL}, 0, 0, 0, "inherit"},
71 {{NULL, NULL}, 0, 0, 0, "on"},
72 {{zio_checksum_off, zio_checksum_off}, 0, 0, 0, "off"},
73 {{zio_checksum_SHA256, zio_checksum_SHA256}, 1, 1, 0, "label"},
74 {{zio_checksum_SHA256, zio_checksum_SHA256}, 1, 1, 0, "gang_header"},
75 {{fletcher_2_native, fletcher_2_byteswap}, 0, 1, 0, "zilog"},
76 {{fletcher_2_native, fletcher_2_byteswap}, 0, 0, 0, "fletcher2"},
77 {{fletcher_4_native, fletcher_4_byteswap}, 1, 0, 0, "fletcher4"},
78 {{zio_checksum_SHA256, zio_checksum_SHA256}, 1, 0, 1, "sha256"},
79 {{fletcher_4_native, fletcher_4_byteswap}, 0, 1, 0, "zilog2"},
80 };
81
82 enum zio_checksum
83 zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
84 {
85 ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
86 ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
87 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
88
89 if (child == ZIO_CHECKSUM_INHERIT)
90 return (parent);
91
92 if (child == ZIO_CHECKSUM_ON)
93 return (ZIO_CHECKSUM_ON_VALUE);
94
95 return (child);
96 }
97
98 enum zio_checksum
99 zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
100 enum zio_checksum parent)
101 {
102 ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
103 ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
104 ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
105
106 if (child == ZIO_CHECKSUM_INHERIT)
107 return (parent);
108
109 if (child == ZIO_CHECKSUM_ON)
110 return (spa_dedup_checksum(spa));
111
112 if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
113 return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
114
115 ASSERT(zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_dedup ||
116 (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
117
118 return (child);
119 }
120
121 /*
122 * Set the external verifier for a gang block based on <vdev, offset, txg>,
123 * a tuple which is guaranteed to be unique for the life of the pool.
124 */
125 static void
126 zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
127 {
128 dva_t *dva = BP_IDENTITY(bp);
129 uint64_t txg = BP_PHYSICAL_BIRTH(bp);
130
131 ASSERT(BP_IS_GANG(bp));
132
133 ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
134 }
135
136 /*
137 * Set the external verifier for a label block based on its offset.
138 * The vdev is implicit, and the txg is unknowable at pool open time --
139 * hence the logic in vdev_uberblock_load() to find the most recent copy.
140 */
141 static void
142 zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
143 {
144 ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
145 }
146
147 /*
148 * Generate the checksum.
149 */
150 void
151 zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
152 void *data, uint64_t size)
153 {
154 blkptr_t *bp = zio->io_bp;
155 uint64_t offset = zio->io_offset;
156 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
157 zio_cksum_t cksum;
158
159 ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
160 ASSERT(ci->ci_func[0] != NULL);
161
162 if (ci->ci_eck) {
163 zio_eck_t *eck;
164
165 if (checksum == ZIO_CHECKSUM_ZILOG2) {
166 zil_chain_t *zilc = data;
167
168 size = P2ROUNDUP_TYPED(zilc->zc_nused, ZIL_MIN_BLKSZ,
169 uint64_t);
170 eck = &zilc->zc_eck;
171 } else {
172 eck = (zio_eck_t *)((char *)data + size) - 1;
173 }
174 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
175 zio_checksum_gang_verifier(&eck->zec_cksum, bp);
176 else if (checksum == ZIO_CHECKSUM_LABEL)
177 zio_checksum_label_verifier(&eck->zec_cksum, offset);
178 else
179 bp->blk_cksum = eck->zec_cksum;
180 eck->zec_magic = ZEC_MAGIC;
181 ci->ci_func[0](data, size, &cksum);
182 eck->zec_cksum = cksum;
183 } else {
184 ci->ci_func[0](data, size, &bp->blk_cksum);
185 }
186 }
187
188 int
189 zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
190 {
191 blkptr_t *bp = zio->io_bp;
192 uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
193 (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
194 int byteswap;
195 int error;
196 uint64_t size = (bp == NULL ? zio->io_size :
197 (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
198 uint64_t offset = zio->io_offset;
199 void *data = zio->io_data;
200 zio_checksum_info_t *ci = &zio_checksum_table[checksum];
201 zio_cksum_t actual_cksum, expected_cksum, verifier;
202
203 if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
204 return (EINVAL);
205
206 if (ci->ci_eck) {
207 zio_eck_t *eck;
208
209 if (checksum == ZIO_CHECKSUM_ZILOG2) {
210 zil_chain_t *zilc = data;
211 uint64_t nused;
212
213 eck = &zilc->zc_eck;
214 if (eck->zec_magic == ZEC_MAGIC)
215 nused = zilc->zc_nused;
216 else if (eck->zec_magic == BSWAP_64(ZEC_MAGIC))
217 nused = BSWAP_64(zilc->zc_nused);
218 else
219 return (ECKSUM);
220
221 if (nused > size)
222 return (ECKSUM);
223
224 size = P2ROUNDUP_TYPED(nused, ZIL_MIN_BLKSZ, uint64_t);
225 } else {
226 eck = (zio_eck_t *)((char *)data + size) - 1;
227 }
228
229 if (checksum == ZIO_CHECKSUM_GANG_HEADER)
230 zio_checksum_gang_verifier(&verifier, bp);
231 else if (checksum == ZIO_CHECKSUM_LABEL)
232 zio_checksum_label_verifier(&verifier, offset);
233 else
234 verifier = bp->blk_cksum;
235
236 byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
237
238 if (byteswap)
239 byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
240
241 expected_cksum = eck->zec_cksum;
242 eck->zec_cksum = verifier;
243 ci->ci_func[byteswap](data, size, &actual_cksum);
244 eck->zec_cksum = expected_cksum;
245
246 if (byteswap)
247 byteswap_uint64_array(&expected_cksum,
248 sizeof (zio_cksum_t));
249 } else {
250 ASSERT(!BP_IS_GANG(bp));
251 byteswap = BP_SHOULD_BYTESWAP(bp);
252 expected_cksum = bp->blk_cksum;
253 ci->ci_func[byteswap](data, size, &actual_cksum);
254 }
255
256 info->zbc_expected = expected_cksum;
257 info->zbc_actual = actual_cksum;
258 info->zbc_checksum_name = ci->ci_name;
259 info->zbc_byteswapped = byteswap;
260 info->zbc_injected = 0;
261 info->zbc_has_cksum = 1;
262
263 if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
264 return (ECKSUM);
265
266 if (zio_injection_enabled && !zio->io_error &&
267 (error = zio_handle_fault_injection(zio, ECKSUM)) != 0) {
268
269 info->zbc_injected = 1;
270 return (error);
271 }
272
273 return (0);
274 }