| 5 min read
In past blog entries, we’ve written a good amount of articles dealing with Windows exploit development, most of them attacking Vulnserver, a vulnerable-by-design (VbD
) server that, as can be easily guessed, is designed for such a noble endeavor. We also wrote a couple of articles creating an exploit for QuickZIP and MiTec Net Scanner. All of those exploits relied on the ability to execute instructions written on the stack of the process.
However, modern CPUs have a mechanism that allow the OS to prevent that.
In this article, we will introduce that protection and in forthcoming articles we will check a way to bypass it, called ROP
(Return-Oriented Programming).
No-Execute bit
The protection on the CPUs is known as the NX (No-Execute) bit. The OS will use such capability to mark some memory areas (remarkably, the stack) as non-executable and thus, prevent common buffer overflow exploits like the ones we’ve used so far. Let’s clarify that.
In x86
architecture, when a function is called, a function frame is created on the stack. This is a common function stack frame distribution on memory:
Common function stack frame.
.____________._____________._____________._____________._________________.
vuln_var Saved EBP Saved EIP Func args Rest of stack
On a simple buffer overflow, when we write past the vuln_var
size, we can overwrite anything that’s below the stack, including the Saved EBP
and Saved EIP
. When the vulnerable function returns, it will get the Saved EIP
value back from the stack and use it as the next instruction pointer. That’s why we usually overwrite the Saved EIP
with a pointer to a JMP ESP
instruction that allows us to redirect execution back to the stack on where we put the shellcode.
Example overflowed vuln_var.
AAAAAAAAAAAA AAAAAAAAAAAAA <pointer to JMP ESP> Shellcode
.____________._____________.____________________._______________.
vuln_var Saved EBP Saved EIP Rest of stack
For example, let’s take a look a this exploit:
#!/usr/bin/env python3
#
## Simple DEP check
import socket
import struct
HOST = '192.168.0.20'
PORT = 9999
PAYLOAD = (
b'TRUN .' +
b'A' * 2006 +
# 625011AF . FFE4 JMP ESP
struct.pack('<L', 0x625011AF) +
b'\x31\xc0' + # xor eax,eax
b'\x04\x08' + # add al,0x8
b'\x90' + # nop
b'C' * 990
)
with socket.create_connection((HOST, PORT)) as fd:
fd.sendall(PAYLOAD)
This is a simple exploit that will take advantage of a buffer overflow vulnerability of the Vulnserver TRUN
command. Here you can see the full writeup of how to find that vulnerability using fuzzing and here using reverse engineering.
This version of the exploit will overflow the vulnerable variable this way:
Example overflowed vuln_var.
AAAAAAAAAAAA AAAAAAAAAAAAA \xaf\x11\x50\x62 \x31\xc0\x04\x08\x90 CCCCCCCCCCC
.____________._____________._________________.____________________.___________.
vuln_var Saved EBP Saved EIP Shellcode Fill buffer
Where:
-
2,006
A
s are added to trigger the overflow. -
0x625011AF
is a pointer to aJMP ESP
instruction and will be placed onSaved EIP
. -
When the vulnerable function returns, it will execute the instruction pointed by
Saved EIP
which holds theJMP ESP
instruction. -
With that, the execution flow is now redirected to the stack where the shellcode was placed.
-
The shellcode, in this case will execute three arbitrary instructions:
-
xor eax eax
→ Zero-outsEAX
register -
add al,0x8
→ MakesEAX
=0x00000008
-
nop
→ Does nothing
-
Let’s see it in action:
As you can see, we were able to execute the instructions on our shellcode that we placed on the stack, as expected.
Enabling DEP
On modern Windows versions, the NX bit of the CPU can be leveraged by using a feature called Data Execution Prevention or DEP
. An application can be compiled with the /NXCOMPAT
flag to enable DEP
for that application. Also, you can use editbin.exe /NXCOMPAT
over an .exe
file to enable it on an already compiled file.
In a debugger, we can check if an executable has that flag enabled:
You can also enable DEP
system-wide, which will force DEP
to all applications, including those compiled without /NXCOMPAT
. To do that, you can use the following instructions:
-
Press the
Windows
key and search forView advanced system settings
. -
In the resulting window, click on tab
Advanced
:
-
Then in
Performance
click onSettings
. -
Move to the tab
Data Execution Prevention
:
-
The default setting is
Turn on DEP for essential Windows programs…
, but to turn it on for every application, you must selectTurn on DEP for all programs…
. -
Apply and restart the PC.
WARNING: When you change this value and you have Bitlocker
enabled, you will be asked to enter the Bitlocker
recovery key after the reboot. If you don’t have that information, please don’t change the DEP
value or your system will become unusable.
With that in place, we can check our exploit again to see if DEP
really prevents the execution of the instructions of our shellcode.
NOTE: We will talk about Hardware-based DEP
, which uses the NX bit of the CPU to mark memory regions as non-executable. Software-based DEP
will only prevent SEH-based overflows, and that’s not in the scope of this article. You can get more information on SEH-based exploits here.
Executing shellcode with DEP enabled
Now, after enabling DEP
system-wide, let’s execute our exploit again:
Several things have happened:
-
The overflow is performed.
-
The
Saved EIP
value was overwritten successfully with the pointer toJMP ESP
. -
The
JMP ESP
instruction is performed and execution flow is redirected to the stack where our shellcode is placed. -
However, when it tries to execute the first instruction on the shellcode (
xor eax,eax
), anAccess violation
exception is triggered, which means that it was trying to execute code on a memory region marked as non-executable.DEP
worked.
Bypassing DEP
Now, we cannot execute instructions placed on the stack, but we control the execution flow of the application. However, the stack is a place where the application (and therefore, the exploit) can read and write data and by controlling both (the execution flow and the stack), we can do wonders.
In the previous example, we couldn’t execute the instructions on the shellcode, but we were able to execute a single instruction: JMP ESP
. We did that by placing the pointer to the instruction in the right place.
We can use that to run arbitrary code, without executing a single instruction on the stack. Let’s welcome Return-Oriented Programming.
Conclusions
This article shows a mechanism created to prevent the exploitation of buffer overflow vulnerabilities. DEP
surely renders common exploits unsuccessful. However, in the next article we will see how to bypass DEP
using Return-Oriented Programming and later we can create a fully working exploit that triggers a reverse TCP shell on a DEP-enabled application.
At Fluid Attacks, we probe security measures to find weaknesses. We do that in research but also perform red teaming operations for software development companies. Right now we offer a 21-day free trial of our vulnerability scanning. You can upgrade at any time to include manual security testing, such as penetration testing, by our hackers.
Share
Recommended blog posts
You might be interested in the following related posts.
Consequential data breaches in the financial sector
Data breaches that left their mark on time
Our pick of the hardest challenges for ethical hackers
Beware of insecure-by-default libraries!
An OffSec Exploitation Expert review
An interview with members of our hacking team
What is invisible to some hackers is visible to others