Last month I gave a workshop at Montrehack in which I presented a basic overview of what return oriented programming (ROP) is and how it is used in modern exploitation. A lot more people than I had anticipated turned up, which was very appreciated. Being that this was my first time presenting at this kind of event and also my first time designing challenges, I was worried that my challenges might be too easy or too difficult, but I think they turned out just right.
In this post I will go over the first of the three challenges, called
Stay tuned for the write ups for the other two challenges.
The code, binaries and full solutions can be found here for people who would like to follow along.
Without any further ado, let’s get started.
The first step in most binary exploitation challenges is usually to figure out what kind of program you’re dealing with, what kinds of protections are in place, and what the program does.
Before even running the program, it’s a good idea to identify interesting information about the binary file:
Great, so it’s a statically linked binary with no PIE, only partial RELRO and debugging symbols embedded in it. this should make it easy to reverse. The canaries being present would be a bad thing, but it’s very likely not the case and just an artifact of static linking (LIBC has some canaries enabled in )
This can be confirmed while reversing the binary and developing the exploit.
There are a few ways to go about finding the vulnerability. The most straightforward one is to play with the program dynamically and see what it does to get a feel for where the bug might be. This approach is not always feasible when the binary comes from untrusted sources or is malware. In the case of an organized CTF problem like this, it’s usually relatively safe to run it.
This small program has 2 interesting features:
- Read user-controlled data and display it to the screen (Option 1)
- Write data to a buffer somewhere (Option 2)
In nearly all binary challenges, the goal is to take control of the execution
$eip in x86 or
$rip in x64) to execute arbitrary code. This
challenge is no exception.
Digging into the disassembly reveals two interesting functions that seem to
correspond to the features identified above
The big red flag comes from the use of
gets when reading the new message of
the day. The
gets function is dangerous because it does not perform any
bound-check when reading user input. Because the
motd buffer is on the stack,
it is possible to override the stack until the beginning of the
stack frame, thereby overwriting where
read_motd will return to.
Sending a rather large amount of data to
gets results in a swift crash. This
makes it possible to overwrite
$rip to any location.
As seen above, the function will return to
corresponds to the sent buffer of
A characters. The next step is to identify
exactly where in the buffer this pattern occurs. The general idea is to send a
predictable pattern that will make it possible to retrieve the offset of the
part that ends up in
$rip after the function
read_motd returns. A good tool
to do this comes with the Metasploit Framework and is called
pattern_create.rb and it can be used like this:
After feeding this pattern to the binary (as opposed to only
As), the picture
becomes a bit clearer:
In other words, the return address is at
0x6a41396941386941. Putting that
number of the accompanying
pattern_offset.rb tool gives the offset within the
This output means that the value that ends up being in
returns is at offset 264 in the buffer. The only piece of information missing
now is how to use
$rip to gain code execution.
A common technique that is used in Return Oriented Programming is called
ret2libc and the goal of this challenge was to introduce the concept in a
practical way. Thankfully, the challenge is kind enough to call
system at the
beginning of the
main function. That gadget is very powerful because it makes
it easy to convert data into executable code through the power of the shell.
A few obstacles are in the way though:
- How to move execution to
- How to pass parameters to the function call?
- What is the address of
The first question is easy to answer: Put the address of
system on the stack
where it will end up in
$rip, that is, offset 264 of the payload.
The second question, however, is a little trickier: The calling convention in x64
does not make use of the stack for parameters 1 through 4, instead using
registers for better efficiency. Parameter 1 is always passed in
$rdi, and the
return value is always in
Upon inspection of the register state after the crash, a careful observer will
$rdi points to the stack, at the following data:
is the pattern that was sent to get the crash, meaning it already points to the
stack… lucky! This offset also happens to be 1 byte inside the buffer, meaning
that the argument for system can be stored there.
And finally, onto the last question: Where is
system? There are many answers
to this one since the binary is statically linked and has no address space
- Grab the address of
- Grab the real address of
- Grab the GOT offset of the real address of
readelf method looks like this:
Which means that the address of
0x4092b0 (line 5).
At last, all the parts can be put together:
There are a few tricky details that require explaining in the above code. First,
the address needs to be converted to little endian, since that is how bytes are
stored in memory on the Intel x64 architecture. This is what the
does. Secondly, the
\x00 byte in the payload is necessary to prevent
system() from attempting to evaluate the entire buffer, which would result in
command not found error or otherwise weird behavior. Lastly, if you are
reproducing this locally, you will need to change the
p = remote(...) line to
spawn the process locally, as the challenges have since been taken offline.
The purpose of this challenge was to introduce participants to the basics of
Return Oriented Programming, specifically the
ret2libc technique and provide a
gentle transition to people already familiar with classical buffer overflow
exploits. It introduced a few common tools to deal with reverse engineering and
binary exploitation, without too much complexity.
Challenges 2 and 3 were designed to take the concepts a little bit further and will be linked here once the writeups are available.