Hacks Programming

Reverse Engineering a PyInstaller Executable on Linux

I was working on a reversing challenge recently and I was given an odd binary file.

I say odd because it had a prompt that said “enter username” and such but the string username was not to be found in the binary. Something was up.

So I looked at it from a few angles:

First, I knew it was an ELF(64) (file told me so) so I opened it in Ghidra & Cutter-radare2 w/ the Ghidra plugin(thanks NSA for the free stuff!). The code made little sense but it did give me a hint, it was setting an env var _MEIPASS2.

Next, I ran it against ltrace and noticed it was carving pieces of itself and extracting them in a folder in /tmp. OK, so it was one of those self extracting binaries. Towards the end it fork()ed and launched a new binary. Now this binary was not found in the /tmp folder, that only contained support files: they were python libs, specifically python3.6 libs(this is relevant).

fopen("/home/cristian/Desktop/Workspace"..., "rb")                                                                                                                                                   = 0x1601340
fseek(0x1601340, 0x424e96, 0, 15)                                                                                                                                                                    = 0
malloc(8307)                                                                                                                                                                                         = 0x1602bc0
fread(0x1602bc0, 8307, 1, 0x1601340)                                                                                                                                                                 = 1
zlibVersion(0x1601420, 0x1601520, 0, 0x7f9a220b6ab2)                                                                                                                                                 = 0x7f9a221a2034
malloc(24968)                                                                                                                                                                                        = 0x1604c40
inflateInit_(0x7ffd58a73c80, 0x40544a, 112, 0x160adc0)                                                                                                                                               = 0
inflate(0x7ffd58a73c80, 4, 0x160add0, 0x160b328)                                                                                                                                                     = 1
inflateEnd(0x7ffd58a73c80, 0, 0x3f50, 0xe800)                                                                                                                                                        = 0
free(0x1602bc0)                                                                                                                                                                                      = <void>
fclose(0x1601340)                                                                                                                                                                                    = 0
strncpy(0x7ffd58a71d00, "/tmp/_MEIpqGisz", 4096)                                                                                                                                                     = 0x7ffd58a71d00
strncpy(0x7ffd58a72d00, "termios.cpython-36m-x86_64-linux"..., 4096)                                                                                                                                 = 0x7ffd58a72d00
strtok("termios.cpython-36m-x86_64-linux"..., "/")                                                                                                                                                   = "termios.cpython-36m-x86_64-linux"...
strlen("termios.cpython-36m-x86_64-linux"...)                                                                                                                                                        = 39
__strcpy_chk(0x7ffd58a71d10, 0x7ffd58a72d00, 4096, 0x7ffd58a71d0f)                                                                                                                                   = 0x7ffd58a71d10
strtok(nil, "/")                                                                                                                                                                                     = nil
__xstat(1, "/tmp/_MEIpqGisz/termios.cpython-"..., 0x7ffd58a71c70)                                                                                                                                    = -1
fopen("/tmp/_MEIpqGisz/termios.cpython-"..., "wb")                                                                                                                                                   = 0x1601340
fwrite("\177ELF\002\001\001", 24968, 1, 0x1601340)                                                                                                                                                   = 1
fileno(0x1601340)                                                                                                                                                                                    = 3
fchmod(3, 0700)                                                                                                                                                                                      = 0
fclose(0x1601340)                                                                                                                                                                                    = 0

A few fun facts: it opens itself and fseeks at different offsets from which it slices off the different bits, and also manually adds the ELF magic bytes at the beginning of files(this is another important point: binary headers are missing)

Now, I’ll be honest, I googled _MEIPASS2 and found this: https://medium.com/@alexskalozub/solving-the-malwarebytes-crackme-2-6ba23a7c5b56

This was the final hint that this is in fact a PyInstaller packed binary.

For Windows there is a tool to extract PyInstaller Windows executable files but for Linux things are trickier.

Initially I did not know this was built with python3.6 so I fumbled a bit.

The first step is to get pyinstaller installed and use it to get files from the binary:

pip install pyinstaller # this is installs it for python3 on my system, on older distros you might need to use pip3

Then use it to view and extract files:

pyi-archive_viewer ./my-sneaky-binary

You will be prompted with a list of files and offsets. I initially went on a false track and extracted the PYZ-00.pyz file which is just a bundle of python runtime files. The file you need will be marked as ‘s’ -> for script. Extract that to a file called code.pyc or something similar.

Now, the header is stripped so you need to append it back. After I knew that this was built for python3.6 I ran a docker with python3.6(I have python 3.7 on my system) to be able to decompile the .pyc file. I spent a lot of time fumbling with this. Simply put, bytecode generated for a specific version of python is not really forwards-compatible or python 3.7/3.8/3.9 is not binary backwards compatible. And why would it be, you are not supposed to distribute .pyc compiled files.

docker run -v `pwd`:/data --rm -it python:3.6-buster /bin/bash

Now, we need to get a sane header, so just make some dummy python file and compile to get a valid header:

python -m compileall test.py # result is put in __pycache__/test.cpython-36.pyc

Slice off the first 16 bytes from the resulting file:

And add them to beginning of the previously extracted file code.pyc.

Now, still in the docker we need to get uncompyle6(for python3.6) so do:

pip install uncompyle6

If you have problems then you are using the wrong python version.

And then just uncompile it!

uncompyle6 -o . code.pyc

Aaaand done, silly Python, do you call THAT compiling?

This was a fun challenge. I did my best not to share any details that would help others identify what challenge I was solving, I just wanted to talk about the tools to address this on Linux.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.