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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
33 * ZFS's 2nd and 4th order Fletcher checksums are defined by the following
34 * recurrence relations:
42 * c = c + b (fletcher-4 only)
45 * d = d + c (fletcher-4 only)
49 * a_0 = b_0 = c_0 = d_0 = 0
51 * f_0 .. f_(n-1) are the input data.
53 * Using standard techniques, these translate into the following series:
58 * n /___| n - i n /___| n - i
63 * \ | i*(i+1) \ | i*(i+1)*(i+2)
64 * c = > ------- f d = > ------------- f
65 * n /___| 2 n - i n /___| 6 n - i
68 * For fletcher-2, the f_is are 64-bit, and [ab]_i are 64-bit accumulators.
69 * Since the additions are done mod (2^64), errors in the high bits may not
70 * be noticed. For this reason, fletcher-2 is deprecated.
72 * For fletcher-4, the f_is are 32-bit, and [abcd]_i are 64-bit accumulators.
73 * A conservative estimate of how big the buffer can get before we overflow
74 * can be estimated using f_i = 0xffffffff for all i:
77 * f=2^32-1;d=0; for (i = 1; d<2^64; i++) { d += f*i*(i+1)*(i+2)/6 }; (i-1)*4
82 * So blocks of up to 2k will not overflow. Our largest block size is
83 * 128k, which has 32k 4-byte words, so we can compute the largest possible
84 * accumulators, then divide by 2^64 to figure the max amount of overflow:
87 * a=b=c=d=0; f=2^32-1; for (i=1; i<=32*1024; i++) { a+=f; b+=a; c+=b; d+=c }
88 * a/2^64;b/2^64;c/2^64;d/2^64
96 * So a and b cannot overflow. To make sure each bit of input has some
97 * effect on the contents of c and d, we can look at what the factors of
98 * the coefficients in the equations for c_n and d_n are. The number of 2s
99 * in the factors determines the lowest set bit in the multiplier. Running
100 * through the cases for n*(n+1)/2 reveals that the highest power of 2 is
101 * 2^14, and for n*(n+1)*(n+2)/6 it is 2^15. So while some data may overflow
102 * the 64-bit accumulators, every bit of every f_i effects every accumulator,
103 * even for 128k blocks.
105 * If we wanted to make a stronger version of fletcher4 (fletcher4c?),
106 * we could do our calculations mod (2^32 - 1) by adding in the carries
107 * periodically, and store the number of carries in the top 32-bits.
109 * --------------------
110 * Checksum Performance
111 * --------------------
113 * There are two interesting components to checksum performance: cached and
114 * uncached performance. With cached data, fletcher-2 is about four times
115 * faster than fletcher-4. With uncached data, the performance difference is
116 * negligible, since the cost of a cache fill dominates the processing time.
117 * Even though fletcher-4 is slower than fletcher-2, it is still a pretty
118 * efficient pass over the data.
120 * In normal operation, the data which is being checksummed is in a buffer
121 * which has been filled either by:
123 * 1. a compression step, which will be mostly cached, or
124 * 2. a bcopy() or copyin(), which will be uncached (because the
125 * copy is cache-bypassing).
127 * For both cached and uncached data, both fletcher checksums are much faster
128 * than sha-256, and slower than 'off', which doesn't touch the data at all.
131 #include <sys/types.h>
132 #include <sys/sysmacros.h>
133 #include <sys/byteorder.h>
139 fletcher_2_native(const void *buf, uint64_t size,
140 const void *ctx_template, zio_cksum_t *zcp)
142 const uint64_t *ip = buf;
143 const uint64_t *ipend = ip + (size / sizeof (uint64_t));
144 uint64_t a0, b0, a1, b1;
146 for (a0 = b0 = a1 = b1 = 0; ip < ipend; ip += 2) {
153 ZIO_SET_CHECKSUM(zcp, a0, a1, b0, b1);
158 fletcher_2_byteswap(const void *buf, uint64_t size,
159 const void *ctx_template, zio_cksum_t *zcp)
161 const uint64_t *ip = buf;
162 const uint64_t *ipend = ip + (size / sizeof (uint64_t));
163 uint64_t a0, b0, a1, b1;
165 for (a0 = b0 = a1 = b1 = 0; ip < ipend; ip += 2) {
166 a0 += BSWAP_64(ip[0]);
167 a1 += BSWAP_64(ip[1]);
172 ZIO_SET_CHECKSUM(zcp, a0, a1, b0, b1);
177 fletcher_4_native(const void *buf, uint64_t size,
178 const void *ctx_template, zio_cksum_t *zcp)
180 const uint32_t *ip = buf;
181 const uint32_t *ipend = ip + (size / sizeof (uint32_t));
184 for (a = b = c = d = 0; ip < ipend; ip++) {
191 ZIO_SET_CHECKSUM(zcp, a, b, c, d);
196 fletcher_4_byteswap(const void *buf, uint64_t size,
197 const void *ctx_template, zio_cksum_t *zcp)
199 const uint32_t *ip = buf;
200 const uint32_t *ipend = ip + (size / sizeof (uint32_t));
203 for (a = b = c = d = 0; ip < ipend; ip++) {
204 a += BSWAP_32(ip[0]);
210 ZIO_SET_CHECKSUM(zcp, a, b, c, d);
214 fletcher_4_incremental_native(const void *buf, uint64_t size,
217 const uint32_t *ip = buf;
218 const uint32_t *ipend = ip + (size / sizeof (uint32_t));
226 for (; ip < ipend; ip++) {
233 ZIO_SET_CHECKSUM(zcp, a, b, c, d);
237 fletcher_4_incremental_byteswap(const void *buf, uint64_t size,
240 const uint32_t *ip = buf;
241 const uint32_t *ipend = ip + (size / sizeof (uint32_t));
249 for (; ip < ipend; ip++) {
250 a += BSWAP_32(ip[0]);
256 ZIO_SET_CHECKSUM(zcp, a, b, c, d);