ROP Emporium - ret2win

Intro

After discovering how much I enjoyed the pwn CTF category, I started looking into ways to improve my skills. The first thing that was immediately clear was just how expansive the category of binary exploitation is. From buffer overflows to heap exploitation, and everything in between, it was quite overwhelming at first. Not to mention I had very little familiarity with how computer architecture worked in the first place. As I worked to get my bearings, I discovered https://ropemporium.com. This website contains eight challenges designed to teach the basics of a fundamental binary exploitation technique, Return Oriented Programming or ROP. Before I get into the first challenge, I would like to give a shoutout to CryptoCat on YouTube. His videos on these challenges helped me extensively, whether it was debugging a solution or further understanding why something worked the way it did, their help was invaluable. Without further ado, lets have a look at the first challenge, called ret2win.

Recon

When we click on the challenge on the website, we are presented with a little introduction and some download links for various versions of the challenge. We will solving the x86-64 bit version in this writeup. Once the file is downloaded, we can unzip it and do some recon on the binary which is called ret2win. We will use a tool called checksec, which will analyze the binary for us to view some of its protections.

$ checksec --file=ret2win

Now we know what protections are enabled on the binary. We have Partial RELRO, No Stack Canary, NX Enabled, and No PIE. No canary and no PIE will make overflowing the buffer pretty easy since there is nothing stopping us from smashing the stack and addresses will remain constant. Because NX is enabled we will not be able to execute off the stack which means no shellcode off the stack. RELRO will not come into play in this challenge. According to the challenge description, in order to get the flag read to us, we will want to find a way to either get a shell to read it ourselves, or find a way for the program to display it out to us. The description also hints at a suspiciously named function called ret2win(). This is a hint that perhaps by calling this function we can have the flag printed out to us.

In order to verify if that is the case, lets have a look at this program in GDB. I will be using a GDB plugin called pwndbg which adds some extra features. It can be checked out here. Once we load the binary into GDB, we can have a look at the functions with info functions.

Functions

Notable functions include a main() function, a pwnme() function, and a ret2win() function. Let’s have a look inside these functions using GDB’s disassemble functionality to try and get a good gauge of how this program is layed out. Let’s start with the main() function. To disassemble a function in GDB we can use disassemble main.

main disassembly

Looking at the main() function, we can see that it calls several functions, namely setvbuf(), puts(), and the pwnme() function. Nothing seems overtly exploitable here, but the notable thing is that this main function is calling pwnme(). This must be where our vulnerability is, as the name of the function would suggest. Let’s have a look at that function next.

pwnme disassembly

Several things are happening here, most notably there is a call to read(). Now we can step back a bit and make a few infrences. When we run the program, it tells us the following:

running program

This must mean that the read() function is allowing us to read in 56 bytes while only allocating 32 bytes of stack space which would allow us to smash the stack and gain control of the program execution. So, now that we know we have control over the execution of the program, what next? Well we know we have to have the flag printed out to us somehow so let’s hop back into GDB and have a look at that ret2win() function.

ret2win disassembly.png

We can see that the ret2win() function is calling system with an argument located at 0x400943. Remember as per x86-64 calling conventions the first argument for functions called is stored in the rdi register and we can see an argument being moved into the edi register just before system is called, which is the lower order version of the rdi. What could that argument be? Judging from the name of the challenge and the description, it could be something like cat flag.txt or /bin/sh. Either of these would allow us to read the flag.

So, now we know that we have control over the programs execution via a buffer overflow and we need to call ret2win() to (hopefully) print out the flag.

Exploitation

The goal of our exploit will be to overflow the buffer just enough to gain control of the rip, which is the register that holds the address of the current instruction the program is executing. If we can overwrite rip with, lets say the address of the ret2win() function, we can control program execution to call ret2win() which was never meant to be called. We will use the principle called ROP (duh) which means we will leverage the control of return addresses to redirect the execution of the program to where we want it. So now we know two things; we will need to overflow the buffer enough to control rip and we will need the address of ret2win(). Now, lets solve this challenge!

To find out how many characters we will need to gain control of the rip, we can use pwndbg to generate a De Bruijn sequence with the command:

cyclic 50

The nature of the De Bruijn sequence means that we can find out how many characters there are up to a certain point in the sequence. This means we will be able to pass this into the program, watch it segfault, and then pinpoint exactly how many characters of input it takes to reach the rip. Let’s have a look at this in GDB.

After entering in our cyclic pattern, we can see that the program crashed! We can see GDB’s output below.

Crashed program.png

From our cyclic pattern we can see that part of our pattern was left in the rip! This is exactly what we wanted. To find out how many characters it took, we can do:

$ cyclic -l faaaaaaa

This returns 40, which means that it took 40 characters to overflow the buffer and leak into rip. Ultimately we will need to send in 40 characters to overflow the buffer right up to we get to rip, and then pass in the address of ret2win() so that we can successfully redirect program execution to that function. We can now start crafting an exploit script. I will be using a popular python library for pwn challenges called pwntools. It comes with some amazing functionality to exploit these types of challenges. This challenge will only be scratching the surface.

First we will import the pwn tools library and define our binary. We will use context.binary so that pwntools knows its an x86-64 binary and it take care of things like endianness for us. We will use an ELF() object so that pwntools can access all of the symbols and such for our binary.

from pwn import *

elf = context.binary = ELF("./ret2win")

Next we will define a couple more things. We will make our binary a process so that pwntools can run it and send input in, and we will define our offset.

p = elf.process()

offset = 40

Then we can make note of the address of our ret2win() function. This address can be obtained from GDB, just like we were doing before.

ret2win = 0x400756

Now we can now construct our payload. We can send in 40 a’s to smash the stack just enough and then send in the address of our ret2win() function so that rip will point to the ret2win function. To do this we will use the flat() function so that our payload will be converted into bytes. Since we specified the context of the binary, pwntools will automatically send our payload in the proper endianness and pad to the proper address size.

payload = flat([
	offset * 'a',
	ret2win
])

Then we can send this to the program/process using the line:

p.sendline(payload)

Finally we can use interactive mode to view any output that the program spit back at us. Like our flag!

p.interactive()

Our final payload will look something like this:

from pwn import *

elf = context.binary = ELF("./ret2win")

p = elf.process()

offset = 40

ret2win = 0x400756

payload = flat([
	offset * b'A',
	ret2win_func
	])

p.sendlineafter(payload)

p.interactive()

Once we run this script, we should get the flag!

final output

And we do!

This was the first of the ROP Emporium challenges. These challenges were very fun to solve and really helped me learn more about binary exploitation. Hope you enjoyed!