关于SQL SERVER数据页checksum校验算法整理

Posted 2022-06-17

SQL SERVER数据页checksum校验算法

在SQL SERVER2005以上版本中，数据页默认开启checksum，标识为m_flagBits & 0x200 == True，其值m_tornBits位于页头0x3C，4字节。
其算法概述如下：

读8KB 进BUF
将BUF头部 CHECKSUM的4字节值清0
uint32 checksum = 0 //初始checksumfor i in range(0,15):
         //每扇区的初始checksum
         overall = 0;
        for ii in range(0,127):
                 //对当前扇区的每个4字节做累加异或
                overall = overall ^ BUF[i][ii];
                //对每扇区的checksum进行移位，方法为向左移位15-i位，
                //左边移出的15-i位补到最低位。
                checksum = checksum ^ rol(overall, 15- i); 
return checksum; //Gets checksum

c源码如下：

//***CODE***//#include <stdio.h>#include <stdlib.h>
#define seed 15 //Initial seed(for first sector)#define CHAR_BIT 8 
//***PROTOTYPES***//unsigned int page_checksum(int page_id, unsigned int *ondisk);unsigned int rol(unsigned int value, unsigned int rotation);
int main(int argc, char *argv[]) 
    unsigned int computed_checksum; //Var to retrieve calculated checksum
    unsigned int ondisk_checksum; //Var to retrieve checksum on disk
        computed_checksum = page_checksum(152, &ondisk_checksum); //page_checksum call to retrieve stored and calculated checksum for page 152
        //***PRINTS***//
        printf("Calculated checksum: 0x%08x\n", computed_checksum);
        printf("On disk checksum: 0x%08x\n", ondisk_checksum);

unsigned int page_checksum(int page_id, unsigned int *ondisk)
    FILE *fileptr; 
    unsigned int i; 
    unsigned int j;
    unsigned int checksum;
    unsigned int overall;
    unsigned int *pagebuf[16][128]; //A pointer to describe 2d array [sector][element]
    fileptr = fopen("C:\\Users\\andre\\Desktop\\teste.mdf", "r+b"); //Open dummy data file for binary read
    fseek(fileptr, page_id * 8192, SEEK_SET); //Calculate page address on data file and points to it
    fread(pagebuf, 4, 2048, fileptr); //Read page buffer
    fclose(fileptr);
    checksum = 0;
    overall = 0;
    *ondisk = pagebuf[0][15]; //This means that torn bits is stored on first sector in 15th element, Internals researches understand this
    pagebuf[0][15] = 0x00000000; //Fill checksum field with zeroes (this field will be discarded in algorithm)

    for (i = 0; i < 16; i++) //Loop through sectors
    
        overall = 0; //Reset overall sum for sectors
        for (j = 0; j < 128; j++) //Loop through elements in sector i
        
            overall = overall ^ (unsigned int)pagebuf[i][j]; //XOR operation between sector i elements
        
        checksum = checksum ^ rol(overall, seed - i); //Current checksum is overall for sector i circular shifted by seed (15 - i)
    
    return checksum; //Gets checksum

unsigned int rol(unsigned int value, unsigned int rotation)
    return (value) << (rotation) | (value) >> (sizeof(int) * CHAR_BIT - rotation) & ( (1 << rotation) -1);