Self-Modifying Code

To write a self-modifying code, we'd usually need to run in the real-mode using the CS register (Code segment), so to avoid that restriction, we'll work around it by writing the code in the data instead of in the code segment, though the basic idea remains the same: overwriting specific bytes to change the instructions. To do this, we'll use a C program for calling the functions as well as printing.


#include "stdio.h"
extern int foo(void);
extern int change_code(void);

int main(void){
    printf("%d\n", foo());
    printf("changed %d\n", change_code());
    printf("%d\n", foo());
    return 0;


segment .data
    global foo
    global change_code

    mov eax, 10

    mov eax, [foo+1]
    mov ecx, eax
    neg ecx
    mov dword [foo+1], ecx

What happens here is that change_code takes the 4 bytes value (the 10) attached to the 1 byte µop to move the following 4 bytes into EAX, then it negates it and overwrites those 4 bytes with the new value, the -10. So until change_code is called once again, foo will always return -10.

$ nasm -f elf modifying.asm -o modifying.o
$ gcc -m32 main.c modifying.o -o main
$ ./main
changed 10

The important matter for achieving this effect is to know how many bytes of size which µops are and how many bytes the new instructiopn should take. If there is any difference in the size, it's importaant to add NOPs (no operation) to fill the gap. In this simple case, we only changed the 4 bytes value to be written on EAX, so this was not neeccesary. To see the corresponding opcodes and the sizes, you may first write on the code segment and use objdump on the resulting object file.