SQL SERVER数据页checksum校验算法

愿太阳照亮我

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：

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

复制代码

c源码如下：

1. //***CODE***//
   
2. #include <stdio.h>
   
3. #include <stdlib.h>
   
4. 
   
5. #define seed 15 //Initial seed(for first sector)
   
6. #define CHAR_BIT 8
   
7. 
   
8. //***PROTOTYPES***//
   
9. unsigned int page_checksum(int page_id, unsigned int *ondisk);
   
10. unsigned int rol(unsigned int value, unsigned int rotation);
    
11. 
    
12. int main(int argc, char *argv[]) {
    
13. 
    
14.     unsigned int computed_checksum; //Var to retrieve calculated checksum
    
15.     unsigned int ondisk_checksum; //Var to retrieve checksum on disk
    
16. 
    
17.         computed_checksum = page_checksum(152, &ondisk_checksum); //page_checksum call to retrieve stored and calculated checksum for page 152
    
18. 
    
19.         //***PRINTS***//
    
20.         printf("Calculated checksum: 0x%08x\n", computed_checksum);
    
21.         printf("On disk checksum: 0x%08x\n", ondisk_checksum);
    
22. 
    
23. }
    
24. 
    
25. unsigned int page_checksum(int page_id, unsigned int *ondisk)
    
26. {
    
27. 
    
28.     FILE *fileptr;
    
29.     unsigned int i;
    
30.     unsigned int j;
    
31.     unsigned int checksum;
    
32.     unsigned int overall;
    
33.     unsigned int *pagebuf[16][128]; //A pointer to describe 2d array [sector][element]
    
34. 
    
35.     fileptr = fopen("C:\\Users\\andre\\Desktop\\teste.mdf", "r+b"); //Open dummy data file for binary read
    
36. 
    
37.     fseek(fileptr, page_id * 8192, SEEK_SET); //Calculate page address on data file and points to it
    
38. 
    
39.     fread(pagebuf, 4, 2048, fileptr); //Read page buffer
    
40. 
    
41.     fclose(fileptr);
    
42. 
    
43.     checksum = 0;
    
44.     overall = 0;
    
45. 
    
46.     *ondisk = pagebuf[0][15]; //This means that torn bits is stored on first sector in 15th element, Internals researches understand this
    
47. 
    
48.     pagebuf[0][15] = 0x00000000; //Fill checksum field with zeroes (this field will be discarded in algorithm)
    
49. 
    
50.     for (i = 0; i < 16; i++) //Loop through sectors
    
51.     {
    
52. 
    
53.         overall = 0; //Reset overall sum for sectors
    
54. 
    
55.         for (j = 0; j < 128; j++) //Loop through elements in sector i
    
56.         {
    
57.             overall = overall ^ (unsigned int)pagebuf[i][j]; //XOR operation between sector i elements
    
58.         }
    
59. 
    
60.         checksum = checksum ^ rol(overall, seed - i); //Current checksum is overall for sector i circular shifted by seed (15 - i)
    
61.     }
    
62. 
    
63.     return checksum; //Gets checksum
    
64. 
    
65. }
    
66. 
    
67. unsigned int rol(unsigned int value, unsigned int rotation)
    
68. {
    
69.     return (value) << (rotation) | (value) >> (sizeof(int) * CHAR_BIT - rotation) & ( (1 << rotation) -1);
    
70. }

复制代码