Configuration-oriented encryption toolkit to make secure config files simple
by Naomi Most (@nthmost)
A simple solution to the often annoying problem of protecting config files.
secureconfig makes keeping your secrets secure on servers and source control repositories easy by restricting your choices on the matter, defaulting to a “medium-high paranoia” set of operations.
Those choices, specifically:
Use AES-128 CBC via Fernet (see https://cryptography.io/en/latest/fernet/ ) Store keys in environment variables, or in files protected by the system. Provide an easy way to overwrite sensitive string data left behind by with zeroes.
This library has undergone an overhaul since last version, so if you were using 0.0.2.x, please read the below carefully so you understand what has changed (A LOT!).
The major philosophical shift was to separate the duties of encryption and data structure handling into the CryptKeeper classes and SecureConfig* classes. This means that if you just want a consistent way to encrypt/decrypt strings and files that works across all of your data structures, the CryptKeeper classes will come in handy.
Config styles currently supported:
ConfigParser (see SecureConfigParser) Json (see SecureJson) -- whole-data encryption only. serialized dictionaries (see SecureConfig) -- whole-data encryption only.
Please let the maintainer (@nthmost) know if you want to see another type supported.
secureconfig is being developed in the context of an open-source-loving, “information wants to be free” kind of company that also does not wish to get totally pwned in a heartbeat when, say, bitbucket has a major security breach.
We have a lot of pre-existing code that makes use of ConfigParser ini-style files and also JSON config files. The best solution for protecting our services and sensitive information would be to create a drop-in replacement for ConfigParser that allows us to keep 99% of the way we interact with config files, and simply wraps the decryption step.
Of course, once you have decryption handled, you start to want simplified ways of encrypting as well.
That’s why secureconfig (as of 0.0.3) supports writing new config files. See “basic usage” sections below to see how you can easily turn a plaintext value or file into an encrypted value or file (depending on config style).
The CryptKeeper classes can even generate new keys for you. Just make sure you keep track of which keys match with your data; this library will not stop you from shooting yourself in the foot!
secureconfig also tries to be helpful in keeping stored keys secure. FileCryptKeeper has “paranoid” mode on by default, which means that it will check to see whether the key is in a directory protected by your operating system. If not, it will refuse to run. (Turn this off using paranoid=False, if you must.)
Finally, secureconfig contains a smattering of deployment utilities found in secureconfig.utils. Feel free to suggest new ones.
This library can be found at https://bitbucket.org/nthmost/python-secureconfig
Contributions and code/documentation critiques are warmly welcomed.
How secureconfig Works
At its core, secureconfig simply subclasses the configuration mechanisms we all know and love, and wraps certain operations (read-from-file and/or read-and-interpolate) in a decryption layer.
This library bases its operations on Fernet, a cryptography meta-protocol (see https://cryptography.io) developed to help programmers choose the best possible defaults for their encryption tasks.
The CryptKeeper classes handle key storage, en/decryption, and key generation. All SecureConfig* classes receive from_x class instantiation methods to set up an internal CryptKeeper.
Table of Methods of key storage - CryptKeeper class - SecureConfig* classmethod:
string -- CryptKeeper -- .from_key(key_string) file -- FileCryptKeeper -- .from_file(key_filename) environment variable -- EnvCryptKeeper -- .from_env(key_env_name)
All CryptKeeper classes have a default argument of proactive=True, which means that the CryptKeeper instance will try to store a key in that place whether it currently exists or not. If this place is not writeable, you’ll get your OS’s usual error for an attempted operation.
When proactive=False and locations do not exist, you’ll get a KeyError for environment variables or an OSError for files.
If CryptKeeper classes are instantiated without a key argument, they will generate a key automatically for you.
Another way to generate a new key is to use the CryptKeeper classmethod .generate_key().
NOTICE: You can’t assign a new key to a CryptKeeper object after it’s been created and have it work. (If that seems like misbehaviour, let me know; it’s changeable.)
All of the SecureConfig* classes can be used with or without encryption keys, although you’ll get a SecureConfigException(‘bad data or no encryption key’) if you try to parse a data structure (such as JSON) out of encrypted text.
Finally, in secureconfig is a class called SecureString, which is a subclass of the string object. Its special function is to zero out the memory location of the string payload. This class has its own section and explanation at the bottom.
SecureString must be considered HAZARDOUS MATERIALS and not implicitly trusted. See below for why.
Installation and Requirements
To install secureconfig, you’ll need to have the development libraries for libffi and libssl installed on your system. On ubuntu, therefore, you’d do this:
sudo apt-get install libffi-dev libssl-dev
Beyond this requirement, most users will find they can install secureconfig via pip:
pip install secureconfig
The following requirements form the backbone of secureconfig:
cryptography configparser cffi six pycparser
If you have any problems installing these requirements, please let the maintainer of this package know at https://bitbucket.com/nthmost/python-secureconfig
NEW SINCE 0.1.0:
SecureConfigParser is a subclass of the configparser module’s ConfigParser class.
The difference is that, when instantiated via one of the standardized cryptkeeper classmethods (see above) so that a private key is supplied, SecureConfigParser detects encrypted entries and decrypts them when demanded (i.e. when .get is used).
So, unlike SecureJson, this class encrypts and decrypts single values rather than entire files.
All of the usual ConfigParser methods are available.
In addition, you can set new values into the config to be encrypted by supplying encrypt=True as an argument to the .set method. See an example of this below.
from secureconfig import SecureConfigParser # starting with an ini file that has unencrypted entries: configpath = '/etc/app/config.ini' key_env = 'SCP_INI_KEY' scfg = SecureConfigParser.from_env('SCP_INI_KEY') scfg.read(configpath) username = scfg.get('credentials', 'username') password = scfg.get('credentials', 'password') connection = GetSomeConnection(username, password) # IMPORTANT: supply encrypt=True to encrypt values. scfg.set('credentials', 'password', 'better_password', encrypt=True) fh=open('/path/to/new_scfp.ini', 'w') scfg.write(fh) fh.close()
SecureJson is a very simple wrapper around JSON data. It decrypts whole files (or whole strings) and can encrypt new configurations as well.
Use one of the cryptkeeper classmethods above to instantiate with a key. SecureJson will happily process plaintext data as well if no key is supplied.
SecureJson is a subclass of SecureConfig (see below), and as such, as some ConfigParser-like operations included.
Basic usage (CHANGED SINCE 0.1.0):
from secureconfig import SecureJson configpath = '/etc/app/config.json.enc' config = SecureJson.from_file('.keys/aes_key', filepath=configpath) username = config.get('credentials', 'username') password = SecureString(config.get('credentials', 'password')) connection = GetSomeConnection(username, password) # SecureString overwrites its string data with zeroes upon garbage collection. del(password) # set a new password config.set('credentials', 'password', 'better_password') fh=open('/path/to/config.json.enc', 'w') config.write(fh) fh.close()
The way SecureConfig reads data back is via literal_eval. This approach may not be without its concerns, so please do not use this class to work with data you do not explicitly trust.
The lowly SecureConfig class’s lot in life is to be subclassed by other objects. But it can still be somewhat useful.
SecureConfig stores data in serialized dictionaries, which are then encrypted as a whole and stored as an undecipherable blob of information. The data can only be read and recovered by supplying the private key that it was encrypted with.
SecureConfig provides a .cfg dictionary for raw access. It also provides many ConfigParser style interactions (see class docstring), including .get and .set methods. This works as long as your data is at least 2-dimensional.
You can still use SecureConfig with 1-dimensional data (i.e. flat dictionary of key=value pairs); you just can’t use the ConfigParser style interactions.
Below is demonstrated the non-ConfigParser style of interacting with SecureConfig data.
Basic Usage (CHANGED SINCE 0.1.0):
from secureconfig import SecureConfig config = SecureConfig.from_file('.keys/aes_key', filepath='/path/to/serialized.enc') cfg = config.cfg username = cfg['username'] password = cfg['password'] connection = GetSomeConnection(username, password)
“RAM security is haaaard” –Noah Kantrowitz, https://twitter.com/kantrn/status/461654722558963712
SecureString is a subclass of the string object with one modification: when deleted and garbage-collected by python, or when its .burn() function is called, which explicitly zeroes out the data.
Now this documentation must spend due time convincing you why it is not “secure”.
Python generally tries to create references to ‘payload’ data in memory rather than copy payloads whenever possible, but in those and other scenarios, you may wind up having string data copied into other locations, and SecureString won’t have any idea.
In a “tight” scenario, e.g. where SecureString could be used to receive the password and then immediately be “burned after reading”, SecureString can be trusted to zero out the string data completely. Outside of these strict scenarios, a number of circumstances will create copies of your sensitive data in memory, such as concatenation of strings and use of the comparison operator on strings held in lists.
You must also keep in mind that, even if you del(secure_string) and explicitly run gc.collect(), your string will still be in memory if there are still references to that string lying around in other objects.
Also, if your python program does not complete gracefully, garbage collection may not run completely or at all, so SecureString memory will not be wiped. If you want to insert gc.collect() statements to proactively scrape these strings, that is an option, but there can be performance drawbacks to aggressively running garbage collection operations.
Finally, different python interpreters handle memory differently, and SecureConfig hasn’t yet been tested on more than just the standard python interpreter and the ipython interpreter.
Given the above, SecureString cannot at this time be implicity trusted as “secure”, since so much depends upon how it’s used.
Planned features include:
- more automated-deployment-oriented utils - asymmetric key deployments (e.g. RSA public key encryption)
Look for @nthmost on bitbucket if you’re interested and would like to contribute! Comments, critiques, and bug reports warmly welcomed. Pull requests encouraged.
–Naomi Most, spring 2014.