Implementing A Return Oriented Programming (ROP) Attack: A How-To Guide

1. Contents

• What is ROP?
• Implement a real ROP attack!
• Environment
• Steps to spawn a shell
• Summary

2. What is ROP?

A gadget is a sequence of assembly code that ends with a jump instruction, for example, “pop rax; ret;”. Jump instructions include “ret”, “jmp”, “call”, etc. If you use the last jump instruction of each gadget to execute many gadgets one by one, that’s return-oriented programming(ROP): gadget1 -(jump)> gadget2 -(jump)> gadget3 -(jump)> … Gadgets extensively exist in the vulnerable binary executable. You need to scan the binary executable, find its gadgets, exploit a vulnerability to execute some useful gadgets, and eventually finish your attack.

3. Implement a real ROP attack!

3.1 Environment

Download the necessary files from here. bug is the vulnerable binary executable. exploit_gen.c generates a binary data file called “exploit”. The data file is the input of bug. exploit_gen.c may not be able to exploit the bug on your machine. Follow the steps in the next section. Do your experiments and modify exploit_gen.c.

Here is the source code of bug:

In function fun, we write whatever in the file f into the buffer buf until the end of the file. This is definitely a stack overflow bug.

Here is how we will exploit the bug: First, we search for gadgets within bug. Second, we search for one gadgets within the libc file it uses. Third, we use the bug to overwrite the return address of fun. The program will be redirected to execute the gadgets.

The gadgets will prepare the memory and satisfy the constraints of the one gadgets. Eventually, the last gadget will direct the program to execute a one gadget.

And we will have a shell.

The above exploitation process can be summarized by this workflow: bug -(overwrite)> return address -(jump)> gadgets -(jump)> one gadget.

3.2 Steps to spawn a shell

Find the gadgets within bug by running “ROPgadget –binary bug”. ROPgadget is a tool for discovering gadgets within binary executables. Install it from the Internet. The following picture shows part of the output. Each line is a gadget. They all end with a jump instruction.

Find the libc file name the bug program uses. Use the following commands and you will find out the libc file name it uses: 1) “gdb bug”. 2) “r”. This command should be run within gdb. 3) “ctrl + c”. Press the two keys together on your keyboard. 4) “vmmap”. To use this command, you need to install a plugin such as gef to gdb. You will find out the libc name from the output of this command:

Find one gadgets within the libc file. Open your terminal and run “one_gadget /usr/lib/x86_64-linux-gnu/libc-2.31.so”. Remember to replace the libc file name with yours. You will get an output like the following. This is a one gadget. Its offset is 0xcbcda.

Choose gadgets that satisfy the constraints of the one gadget. Clearly, the above one gadget requires r12 and r13 to be 0. So you need to choose some gadgets to set their values before calling the one gadget.

Look at what gadgets we have now. You will find that the first gadget is perfect to set r12 and r13: pop rbp ; pop r12 ; pop r13 ; pop r14 ; pop r15 ; ret. And it’s offset within the bug program is 0x1273.Inspect the stack. Run bug within gdb. Put a breakpoint at function fun. Observe the stack:

There are 3 variables in function fun. Their addresses are given below.

From the above two pictures, you can clearly tell the structure of the stack:

Prepare the stack. We want to overwrite the return address, and jump to execute the selected gadget, and eventually jump to execute the selected one gadget. So this is how we overwrite the stack:

The data structure in the exploit array is the same as the stack structure we observed. We will use buf to overwrite c, i, rbp, return address and more.

At the first line of the array, we use three characters to overwrite buf and c. It doesn’t matter what the characters are because we don’t use them in the exploitation.

They are just placeholders. Next, we use 0x03 to overwrite i. This is crucial. We use i to access buf and overwrite the stack byte by byte. So we don’t want to change i. When we use buf to overwrite i, i is exactly 0x03. At the second line, the data will overwrite the 8 bytes at address 0x7fffffffdf60. We don’t care about what the data is, because we don’t use it in the exploitation.

The third line overwrites the return address so that the program will jump to execute our selected gadget. We need to calculate the address of the gadget. It equals to “bug’s code segment base address + the gadget offset”. And the base address equals to “address of fun – fun’s offset”. Let’s see how we get the values.

If you look at the source code of bug, you will see that the address of fun is given by the program. It’s 0x555555555155 for me. And we have already acquired the offset of the chosen gadget. It’s 0x1273. fun is a function symbol in bug. Its offset is recorded in bug’s symbol table.

Run “objdump -t bug | grep fun” and you will find the offset of fun in the output: 0x1155. So the gadget’s address = fun’s address – fun’s offset + the gadget’s offset = 0x555555555155 – 0x1155 + 0x1273 = 0x555555555273. This is exactly our third line.

The ret instruction in function fun will return to 0x555555555273 and execute our gadget: pop rbp ; pop r12 ; pop r13 ; pop r14 ; pop r15 ; ret. The gadget takes data from the stack. So we need to prepare the stack accordingly.

Beginning from the fourth line in the exploit array, each line provides a value to an instruction in the gadget. You see that the values set r12 and r13 to 0. It satisfies the requirements of our one gadget. Eventually, we use the last ret instruction in the gadget to jump to execute our one gadget. So we need to give the ret instruction our one gadget’s address.

One gadget’s address = libc base address + one gadget’s offset = printf’s address – printf’s offset + one gadget’s offset. printf’s address is given by the bug program: 0x7ffff7e3bcb0. The one gadget’s offset is acquired in the above steps: 0xcbcda. printf is a function symbol that exists in libc’s symbol table.

Observe libc’s symbol table with this command: objdump -T /usr/lib/x86_64-linux-gnu/libc-2.31.so | grep ” printf”, and you will find the offset of printf is 0x56cb0. Remember to replace the path to libc with yours. So the one gadget’s address = 0x7ffff7e3bcb0 – 0x56cb0 + 0xcbcda = 0x7ffff7eb0cda. This value is exactly the last line of the exploit array. The last ret instruction in the gadget will use this value and return to execute our one gadget.

Observe the exploitation. After correctly setting the exploit array in exploit_gen.c, compile and run it. A data file named “exploit” will be generated. Run bug within gdb. Put a breakpoint at line 14 (the return instruction of fun). Run the program. Now the stack becomes this:

The first column is address. The second is the value at the address. The third is the variable or instruction that uses the value. Now if we use the command “n” to run the program line by line, we will see how the instructions and values lead us to a shell. After the ret instruction of fun is executed, the program goes to execute at address 0x555555555273:

This is our gadget! The gadget takes data from the stack and puts them to the registers. If we execute until the last ret instruction, we will see from rbp to r15 all become 0. After we execute the ret instruction, the program goes to execute our one gadget:

r12 and r13 are arguments of execve. Now they are correctly set to 0. Run the 4 instruction and we will have a shell:

Spawn a shell without gdb. Our exploit cannot bypass ASLR. So you need to shut down ASLR. First, use “su” to get into root mode. Second, shut down ASLR with “echo 0 > /proc/sys/kernel/randomize_va_space”. Now if you execute bug, you will see a shell is spawned:

4. Summary

The post used a concrete example showing you how to implement a ROP attack. Implementing a ROP attack is to execute gadgets one by one that are scattered in the code space of the target program. The execution from one gadget to another is connected by the last jump instruction of each gadget.

The jump instructions include jmp, ret, call, etc. Dear readers, if you find this post helpful, please help me share it on your social media. Thank you so much!