xref: /titanic_51/usr/src/uts/common/fs/zfs/zio_checksum.c (revision 749f21d359d8fbd020c974a1a5227316221bfc9c)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/zfs_context.h>
30 #include <sys/spa.h>
31 #include <sys/zio.h>
32 #include <sys/zio_checksum.h>
33 
34 /*
35  * Checksum vectors.
36  *
37  * In the SPA, everything is checksummed.  We support checksum vectors
38  * for three distinct reasons:
39  *
40  *   1. Different kinds of data need different levels of protection.
41  *	For SPA metadata, we always want a very strong checksum.
42  *	For user data, we let users make the trade-off between speed
43  *	and checksum strength.
44  *
45  *   2. Cryptographic hash and MAC algorithms are an area of active research.
46  *	It is likely that in future hash functions will be at least as strong
47  *	as current best-of-breed, and may be substantially faster as well.
48  *	We want the ability to take advantage of these new hashes as soon as
49  *	they become available.
50  *
51  *   3. If someone develops hardware that can compute a strong hash quickly,
52  *	we want the ability to take advantage of that hardware.
53  *
54  * Of course, we don't want a checksum upgrade to invalidate existing
55  * data, so we store the checksum *function* in five bits of the DVA.
56  * This gives us room for up to 32 different checksum functions.
57  *
58  * When writing a block, we always checksum it with the latest-and-greatest
59  * checksum function of the appropriate strength.  When reading a block,
60  * we compare the expected checksum against the actual checksum, which we
61  * compute via the checksum function specified in the DVA encoding.
62  */
63 
64 /*ARGSUSED*/
65 static void
66 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp)
67 {
68 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
69 }
70 
71 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
72 	NULL,			NULL,			0, 0,	"inherit",
73 	NULL,			NULL,			0, 0,	"on",
74 	zio_checksum_off,	zio_checksum_off,	0, 0,	"off",
75 	zio_checksum_SHA256,	zio_checksum_SHA256,	1, 1,	"label",
76 	zio_checksum_SHA256,	zio_checksum_SHA256,	1, 1,	"gang_header",
77 	fletcher_2_native,	fletcher_2_byteswap,	0, 1,	"zilog",
78 	fletcher_2_native,	fletcher_2_byteswap,	0, 0,	"fletcher2",
79 	fletcher_4_native,	fletcher_4_byteswap,	1, 0,	"fletcher4",
80 	zio_checksum_SHA256,	zio_checksum_SHA256,	1, 0,	"SHA256",
81 };
82 
83 uint8_t
84 zio_checksum_select(uint8_t child, uint8_t parent)
85 {
86 	ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
87 	ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
88 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
89 
90 	if (child == ZIO_CHECKSUM_INHERIT)
91 		return (parent);
92 
93 	if (child == ZIO_CHECKSUM_ON)
94 		return (ZIO_CHECKSUM_ON_VALUE);
95 
96 	return (child);
97 }
98 
99 /*
100  * Generate the checksum.
101  */
102 void
103 zio_checksum(uint_t checksum, zio_cksum_t *zcp, void *data, uint64_t size)
104 {
105 	zio_block_tail_t *zbt = (zio_block_tail_t *)((char *)data + size) - 1;
106 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
107 	zio_cksum_t zbt_cksum;
108 
109 	ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
110 	ASSERT(ci->ci_func[0] != NULL);
111 
112 	if (ci->ci_zbt) {
113 		*zcp = zbt->zbt_cksum;
114 		zbt->zbt_magic = ZBT_MAGIC;
115 		ci->ci_func[0](data, size, &zbt_cksum);
116 		zbt->zbt_cksum = zbt_cksum;
117 	} else {
118 		ci->ci_func[0](data, size, zcp);
119 	}
120 }
121 
122 int
123 zio_checksum_error(zio_t *zio)
124 {
125 	blkptr_t *bp = zio->io_bp;
126 	dva_t *dva = ZIO_GET_DVA(zio);
127 	zio_cksum_t zc = bp->blk_cksum;
128 	uint_t checksum = DVA_GET_GANG(dva) ? ZIO_CHECKSUM_GANG_HEADER :
129 	    BP_GET_CHECKSUM(bp);
130 	int byteswap = BP_SHOULD_BYTESWAP(bp);
131 	void *data = zio->io_data;
132 	uint64_t size = zio->io_size;
133 	zio_block_tail_t *zbt = (zio_block_tail_t *)((char *)data + size) - 1;
134 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
135 	zio_cksum_t actual_cksum, expected_cksum;
136 
137 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
138 		return (EINVAL);
139 
140 	if (ci->ci_zbt) {
141 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
142 			zio_set_gang_verifier(zio, &zc);
143 
144 		if (zbt->zbt_magic == BSWAP_64(ZBT_MAGIC)) {
145 			expected_cksum = zbt->zbt_cksum;
146 			byteswap_uint64_array(&expected_cksum,
147 			    sizeof (zio_cksum_t));
148 			zbt->zbt_cksum = zc;
149 			byteswap_uint64_array(&zbt->zbt_cksum,
150 			    sizeof (zio_cksum_t));
151 			ci->ci_func[1](data, size, &actual_cksum);
152 			zbt->zbt_cksum = expected_cksum;
153 			byteswap_uint64_array(&zbt->zbt_cksum,
154 			    sizeof (zio_cksum_t));
155 		} else {
156 			expected_cksum = zbt->zbt_cksum;
157 			zbt->zbt_cksum = zc;
158 			ci->ci_func[0](data, size, &actual_cksum);
159 			zbt->zbt_cksum = expected_cksum;
160 		}
161 		zc = expected_cksum;
162 	} else {
163 		ASSERT(!DVA_GET_GANG(dva));
164 		ci->ci_func[byteswap](data, size, &actual_cksum);
165 	}
166 
167 	if ((actual_cksum.zc_word[0] - zc.zc_word[0]) |
168 	    (actual_cksum.zc_word[1] - zc.zc_word[1]) |
169 	    (actual_cksum.zc_word[2] - zc.zc_word[2]) |
170 	    (actual_cksum.zc_word[3] - zc.zc_word[3]))
171 		return (ECKSUM);
172 
173 	return (0);
174 }
175