How to get CRC64 distributed calculation (use its linearity property)?
I need hash over pretty large files which is stored on distributed FS. I'm able to process parts of file with much more better performance than whole file so I'd like to be able to calculate hash over parts and then sum it.
I'm thinking about CRC64 as hashing algorithm but I have no clue how to use its theoretical 'linear function' property so I can sum CRC over parts of file. Any recommendation? Anything I missed here?
Additional notes why I'm looking at CRC64:
- I can control file blocks but because of application nature they need to have different size (up to 1 byte, no any fixed blocks are possible).
- I know about
CRC32implementation (zlib) which includes way to sum CRC over parts but I'd like something more wider. 8 bytes look nice for me. - I know CRC is pretty fast. I'd like to get profit from this as file can be really huge (up to few Gb).
Answer
Decided that this was generally useful enough to write and make available:
/* crc64.c -- compute CRC-64
* Copyright (C) 2013 Mark Adler
* Version 1.4 16 Dec 2013 Mark Adler
*/
/*
This software is provided 'as-is', without any express or implied
warranty. In no event will the author be held liable for any damages
arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it
freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not
claim that you wrote the original software. If you use this software
in a product, an acknowledgment in the product documentation would be
appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
Mark Adler
[email protected]
*/
/* Compute CRC-64 in the manner of xz, using the ECMA-182 polynomial,
bit-reversed, with one's complement pre and post processing. Provide a
means to combine separately computed CRC-64's. */
/* Version history:
1.0 13 Dec 2013 First version
1.1 13 Dec 2013 Fix comments in test code
1.2 14 Dec 2013 Determine endianess at run time
1.3 15 Dec 2013 Add eight-byte processing for big endian as well
Make use of the pthread library optional
1.4 16 Dec 2013 Make once variable volatile for limited thread protection
*/
#include <stdio.h>
#include <inttypes.h>
#include <assert.h>
/* The include of pthread.h below can be commented out in order to not use the
pthread library for table initialization. In that case, the initialization
will not be thread-safe. That's fine, so long as it can be assured that
there is only one thread using crc64(). */
#include <pthread.h> /* link with -lpthread */
/* 64-bit CRC polynomial with these coefficients, but reversed:
64, 62, 57, 55, 54, 53, 52, 47, 46, 45, 40, 39, 38, 37, 35, 33, 32,
31, 29, 27, 24, 23, 22, 21, 19, 17, 13, 12, 10, 9, 7, 4, 1, 0 */
#define POLY UINT64_C(0xc96c5795d7870f42)
/* Tables for CRC calculation -- filled in by initialization functions that are
called once. These could be replaced by constant tables generated in the
same way. There are two tables, one for each endianess. Since these are
static, i.e. local, one should be compiled out of existence if the compiler
can evaluate the endianess check in crc64() at compile time. */
static uint64_t crc64_little_table[8][256];
static uint64_t crc64_big_table[8][256];
/* Fill in the CRC-64 constants table. */
static void crc64_init(uint64_t table[][256])
{
unsigned n, k;
uint64_t crc;
/* generate CRC-64's for all single byte sequences */
for (n = 0; n < 256; n++) {
crc = n;
for (k = 0; k < 8; k++)
crc = crc & 1 ? POLY ^ (crc >> 1) : crc >> 1;
table[0][n] = crc;
}
/* generate CRC-64's for those followed by 1 to 7 zeros */
for (n = 0; n < 256; n++) {
crc = table[0][n];
for (k = 1; k < 8; k++) {
crc = table[0][crc & 0xff] ^ (crc >> 8);
table[k][n] = crc;
}
}
}
/* This function is called once to initialize the CRC-64 table for use on a
little-endian architecture. */
static void crc64_little_init(void)
{
crc64_init(crc64_little_table);
}
/* Reverse the bytes in a 64-bit word. */
static inline uint64_t rev8(uint64_t a)
{
uint64_t m;
m = UINT64_C(0xff00ff00ff00ff);
a = ((a >> 8) & m) | (a & m) << 8;
m = UINT64_C(0xffff0000ffff);
a = ((a >> 16) & m) | (a & m) << 16;
return a >> 32 | a << 32;
}
/* This function is called once to initialize the CRC-64 table for use on a
big-endian architecture. */
static void crc64_big_init(void)
{
unsigned k, n;
crc64_init(crc64_big_table);
for (k = 0; k < 8; k++)
for (n = 0; n < 256; n++)
crc64_big_table[k][n] = rev8(crc64_big_table[k][n]);
}
/* Run the init() function exactly once. If pthread.h is not included, then
this macro will use a simple static state variable for the purpose, which is
not thread-safe. The init function must be of the type void init(void). */
#ifdef PTHREAD_ONCE_INIT
# define ONCE(init) \
do { \
static pthread_once_t once = PTHREAD_ONCE_INIT; \
pthread_once(&once, init); \
} while (0)
#else
# define ONCE(init) \
do { \
static volatile int once = 1; \
if (once) { \
if (once++ == 1) { \
init(); \
once = 0; \
} \
else \
while (once) \
; \
} \
} while (0)
#endif
/* Calculate a CRC-64 eight bytes at a time on a little-endian architecture. */
static inline uint64_t crc64_little(uint64_t crc, void *buf, size_t len)
{
unsigned char *next = buf;
ONCE(crc64_little_init);
crc = ~crc;
while (len && ((uintptr_t)next & 7) != 0) {
crc = crc64_little_table[0][(crc ^ *next++) & 0xff] ^ (crc >> 8);
len--;
}
while (len >= 8) {
crc ^= *(uint64_t *)next;
crc = crc64_little_table[7][crc & 0xff] ^
crc64_little_table[6][(crc >> 8) & 0xff] ^
crc64_little_table[5][(crc >> 16) & 0xff] ^
crc64_little_table[4][(crc >> 24) & 0xff] ^
crc64_little_table[3][(crc >> 32) & 0xff] ^
crc64_little_table[2][(crc >> 40) & 0xff] ^
crc64_little_table[1][(crc >> 48) & 0xff] ^
crc64_little_table[0][crc >> 56];
next += 8;
len -= 8;
}
while (len) {
crc = crc64_little_table[0][(crc ^ *next++) & 0xff] ^ (crc >> 8);
len--;
}
return ~crc;
}
/* Calculate a CRC-64 eight bytes at a time on a big-endian architecture. */
static inline uint64_t crc64_big(uint64_t crc, void *buf, size_t len)
{
unsigned char *next = buf;
ONCE(crc64_big_init);
crc = ~rev8(crc);
while (len && ((uintptr_t)next & 7) != 0) {
crc = crc64_big_table[0][(crc >> 56) ^ *next++] ^ (crc << 8);
len--;
}
while (len >= 8) {
crc ^= *(uint64_t *)next;
crc = crc64_big_table[0][crc & 0xff] ^
crc64_big_table[1][(crc >> 8) & 0xff] ^
crc64_big_table[2][(crc >> 16) & 0xff] ^
crc64_big_table[3][(crc >> 24) & 0xff] ^
crc64_big_table[4][(crc >> 32) & 0xff] ^
crc64_big_table[5][(crc >> 40) & 0xff] ^
crc64_big_table[6][(crc >> 48) & 0xff] ^
crc64_big_table[7][crc >> 56];
next += 8;
len -= 8;
}
while (len) {
crc = crc64_big_table[0][(crc >> 56) ^ *next++] ^ (crc << 8);
len--;
}
return ~rev8(crc);
}
/* Return the CRC-64 of buf[0..len-1] with initial crc, processing eight bytes
at a time. This selects one of two routines depending on the endianess of
the architecture. A good optimizing compiler will determine the endianess
at compile time if it can, and get rid of the unused code and table. If the
endianess can be changed at run time, then this code will handle that as
well, initializing and using two tables, if called upon to do so. */
uint64_t crc64(uint64_t crc, void *buf, size_t len)
{
uint64_t n = 1;
return *(char *)&n ? crc64_little(crc, buf, len) :
crc64_big(crc, buf, len);
}
#define GF2_DIM 64 /* dimension of GF(2) vectors (length of CRC) */
static uint64_t gf2_matrix_times(uint64_t *mat, uint64_t vec)
{
uint64_t sum;
sum = 0;
while (vec) {
if (vec & 1)
sum ^= *mat;
vec >>= 1;
mat++;
}
return sum;
}
static void gf2_matrix_square(uint64_t *square, uint64_t *mat)
{
unsigned n;
for (n = 0; n < GF2_DIM; n++)
square[n] = gf2_matrix_times(mat, mat[n]);
}
/* Return the CRC-64 of two sequential blocks, where crc1 is the CRC-64 of the
first block, crc2 is the CRC-64 of the second block, and len2 is the length
of the second block. */
uint64_t crc64_combine(uint64_t crc1, uint64_t crc2, uintmax_t len2)
{
unsigned n;
uint64_t row;
uint64_t even[GF2_DIM]; /* even-power-of-two zeros operator */
uint64_t odd[GF2_DIM]; /* odd-power-of-two zeros operator */
/* degenerate case */
if (len2 == 0)
return crc1;
/* put operator for one zero bit in odd */
odd[0] = POLY; /* CRC-64 polynomial */
row = 1;
for (n = 1; n < GF2_DIM; n++) {
odd[n] = row;
row <<= 1;
}
/* put operator for two zero bits in even */
gf2_matrix_square(even, odd);
/* put operator for four zero bits in odd */
gf2_matrix_square(odd, even);
/* apply len2 zeros to crc1 (first square will put the operator for one
zero byte, eight zero bits, in even) */
do {
/* apply zeros operator for this bit of len2 */
gf2_matrix_square(even, odd);
if (len2 & 1)
crc1 = gf2_matrix_times(even, crc1);
len2 >>= 1;
/* if no more bits set, then done */
if (len2 == 0)
break;
/* another iteration of the loop with odd and even swapped */
gf2_matrix_square(odd, even);
if (len2 & 1)
crc1 = gf2_matrix_times(odd, crc1);
len2 >>= 1;
/* if no more bits set, then done */
} while (len2 != 0);
/* return combined crc */
crc1 ^= crc2;
return crc1;
}
/* Test crc64() on vector[0..len-1] which should have CRC-64 crc. Also test
crc64_combine() on vector[] split in two. */
static void crc64_test(void *vector, size_t len, uint64_t crc)
{
uint64_t crc1, crc2;
/* test crc64() */
crc1 = crc64(0, vector, len);
if (crc1 ^ crc)
printf("mismatch: %" PRIx64 ", should be %" PRIx64 "\n", crc1, crc);
/* test crc64_combine() */
crc1 = crc64(0, vector, (len + 1) >> 1);
crc2 = crc64(0, vector + ((len + 1) >> 1), len >> 1);
crc1 = crc64_combine(crc1, crc2, len >> 1);
if (crc1 ^ crc)
printf("mismatch: %" PRIx64 ", should be %" PRIx64 "\n", crc1, crc);
}
/* Test vectors. */
#define TEST1 "123456789"
#define TESTLEN1 9
#define TESTCRC1 UINT64_C(0x995dc9bbdf1939fa)
#define TEST2 "This is a test of the emergency broadcast system."
#define TESTLEN2 49
#define TESTCRC2 UINT64_C(0x27db187fc15bbc72)
int main(void)
{
crc64_test(TEST1, TESTLEN1, TESTCRC1);
crc64_test(TEST2, TESTLEN2, TESTCRC2);
return 0;
}