Polymorphism (AES encryption)
Metamorphism (logic and constants changing)
Platform independent (Linux/BSD/Windows)
IDPS stealthing (the total number of possible signatures is more the number of atoms in the universe for one given code)
Sandbox evasion (special assembly instructions)
Realism (no null bytes)
Can produce executables (malwares)
Input code can have arbitrary length
Dependencies for the morpher:
Python 2.7 – main engine
Python Crypto 2.6 – for encryption
Dependencies for the code execution:
64-bit Intel AES-NI – for decryption
Nonetheless, there are some limitations (aka white-hat aspects):
Metamorphism is not very robust and can be detected using regular expressions (but can be improved pretty easily)
Unicode null bytes might still work (but who cares?)
It will only work on 64-bit Intel processors with AES-NI support, but since all the user’s PCs (like Pentium, Celeron, i3, i5, i7) and the industry’s servers (like Xeon) have it, it’s more a specification, rather than a limitation, thus a 32-bit implementation is unpractical
Almost any shellcode is guarantee to work however, an arbitrary code (malware) doesn’t
Windows/BSD PoC and executables are in progress…
How it works
Shellcode padding with NOPs (since AES is a block cipher)
Shellcode encryption with a random key using AES-128-ECB (not the best, but the simplest) – polymorphism
Constants randomization, logic changes, instructions modification and rewriting – metamorphism
sudo apt-get install python python-crypto
Execute the Pyhton script and enter your shellcode or nothing for a default Linux shell. You can specify your own execution address as well.
It is possible to build and execute on Windows/BSD/Mac as well, but I’m still testing it.
You can also use the Linux PoC in assembly:
as shellcode.s -o shellcode.old shellcode.o -o shellcode./shellcode
Every file is commented and explainedTests
At this point, it should be pretty obvious that, the hashes would be different every time, but let’s compare SSDEEPes of 2 Linux executables of the same shellcode:
Well, there’s something in common, but globally those are 2 different signatures, now what about the shellcode it-self:
Almost totally different signatures for the same morphed shellcode!
At the publication date, the executable was detected as a shellcode only by 2 out of 53 antiviruses (AVG and Ikarus) on virustotal , but now, it just fails to analyze.malwr’s with cuckoo2 doesn’t see anything suspicious.
On the reverser’s perspective, IDA won’t see anything either.
Radare2 would show the real instructions only if assembled by the assembler it-self however, it doesn’t detects any crypto or suspicious activity for the executable.
Althrough, I didn’t test it personally, I think that FortiSandbox, Sophos Sandstorm, Blue Coat, GateWatcher and their derivatives might fail badly…To put it in the nutshell
Basically, it can transform a script-kid’s code (or a known-one ) into a zero-day.
IDPS will fail because, it’s almost impossible to make a signature and difficult to make a regular expression or heuristic analysis.
Most of the sandboxes doesn’t use Intel’s AES-NI instructions directly, so they will not execute the code, so “everything is fine” for them, whereas it’s not.
The only way to defeat this type of shellcode/malware is to use an appropriate sandboxing or/and an AI.
Notice that, the whole execution is done by a pure assembly, no Python (or shitty OpenSSL) is needed for the shellcode’s/malware’s execution since, I use built-in assembly instructions only, thus it’s system-independent (surely, you will have to assemble it for each-one by adapting the instructions/opcodes, but they are still same).Notes
This is still a work in progress, I will implement Windows and BSD/Mac engines and PoCs ASAP.
IDPSes and sanboxes suck.
“Tradition becomes our security, and when the mind is secure it is in decay.”