Skip to main content

A Key Management Infrastructure

Project description

This package provides a NIST SP 800-57 compliant Key Management Infrastructure (KMI).

To get started do:

$ python # Must be Python 2.7 or higher
$ ./bin/buildout     # Depends on successful compilation of M2Crypto
$ ./bin/runserver    # or ./bin/gunicorn --paste server.ini

The server will come up on port 8080. You can create a new key encrypting key using:

$ wget https://localhost:8080/new -O kek.dat --ca-certificate sample.crt \

or, if you want a more convenient tool:

$ ./bin/testclient https://localhost:8080 -n > kek.dat

The data encryption key can now be retrieved by posting the KEK to another URL:

$ wget https://localhost:8080/key --header 'Content-Type: text/plain' \
       --post-file kek.dat -O datakey.dat --ca-certificate sample.crt


$ ./bin/testclient https://localhost:8080 -g kek.dat > datakey.dat

Note: To be compliant, the server must use an encrypted communication channel of course. The --ca-certificate tells wget to trust the sample self-signed certificate included in the keas.kmi distribution; you’ll want to generate a new SSL certificate for production use.

Key Management Infrastructure

This package provides a NIST SP 800-57 compliant key management infrastructure. Part of this infrastructure is a key management facility that provides several services related to keys. All keys are stored in a specified storage directory.

>>> from __future__ import print_function
>>> import tempfile
>>> storage_dir = tempfile.mkdtemp()
>>> from keas.kmi import facility
>>> keys = facility.KeyManagementFacility(storage_dir)
>>> keys
<KeyManagementFacility (0)>
>>> from zope.interface import verify
>>> from keas.kmi import interfaces
>>> verify.verifyObject(interfaces.IKeyManagementFacility, keys)

One of the services the facility provides in the generation of new keys.

>>> verify.verifyObject(interfaces.IKeyGenerationService, keys)

The algorithm to generate a new pair of keys is somewhat involved. The following features are required:

  1. The key local to the data cannot be directly used as the encrypting key.
  2. The encrypting key must be stored using a cipher that is at least as strong as the key itself.
  3. The computer storing the data cannot also store the key.

This suggests the following algorithm to generate and store a new encrypting key:

  1. Create the key encrypting key (private and public).
  2. Create the encryption key.
  3. Use the public key encrypting key to encrypt both the encryption keys.
  4. Discard the public key encrypting key. It is important that this key is never stored anywhere.
  5. Store the encrypted encryption key in the key management facility.
  6. Return the private key encrypting key.

Let’s now use the key generation service’s API to generate a key.

>>> key = keys.generate()
>>> print(key.decode())

By default the system uses the AES 256 cipher, because public commentary suggests that the AES 196 or AES 256 cipher sufficiently fulfill the PCI, HIPAA and NIST key strength requirement.

You can now use this key encrypting key to extract the encryption keys:

>>> from hashlib import md5
>>> hash_key = md5(key).hexdigest()
>>> len(keys.get(hash_key))

Our key management facility also supports the encryption service, which allows you to encrypt and decrypt a string given the key encrypting key.

>>> verify.verifyObject(interfaces.IEncryptionService, keys)

Let’s now encrypt some data:

>>> encrypted = keys.encrypt(key, b'Stephan Richter')
>>> len(encrypted)

We can also decrypt the data.

>>> keys.decrypt(key, encrypted) == b'Stephan Richter'

We can also encrypt data given by a file descriptor

>>> import tempfile
>>> tmp_file = tempfile.TemporaryFile()
>>> data = b"encryptioniscool"*24*1024
>>> pos = tmp_file.write(data)
>>> pos =
>>> encrypted_file = tempfile.TemporaryFile()
>>> keys.encrypt_file(key, tmp_file, encrypted_file)
>>> tmp_file.close()

And decrypt the file

>>> decrypted_file = tempfile.TemporaryFile()
>>> pos =
>>> keys.decrypt_file(key, encrypted_file, decrypted_file)
>>> encrypted_file.close()
>>> pos =
>>> decrypted_data =
>>> decrypted_file.close()
>>> decrypted_data == data

And that’s pretty much all there is to it. Most of the complicated crypto-related work happens under the hood, transparent to the user.

One final note. Once the data encrypting key is looked up and decrypted, it is cached, since constantly decrypting the the DEK is expensive.

>>> hash_key in keys._KeyManagementFacility__dek_cache

A timeout (in seconds) controls when a key must be looked up:

>>> keys.timeout

Let’s now force a reload by setting the timeout to zero:

>>> keys.timeout = 0

The cache is a dictionary of key encrypting key to a 2-tuple that contains the date/time the key has been fetched and the unencrypted DEK.

>>> firstTime = keys._KeyManagementFacility__dek_cache[hash_key][0]
>>> keys.decrypt(key, encrypted) == b'Stephan Richter'
>>> secondTime = keys._KeyManagementFacility__dek_cache[hash_key][0]
>>> firstTime < secondTime

The Local Key Management Facility

However, using the master key management facility’s encryption service is expensive, since each encryption and decryption request would require a network request. Fortunately, we can

  1. communicate encryption keys across multiple devices, and
  2. keep encryption keys in memory.

It is only required that the data transfer is completed via an encrypted communication channel. In this implementation the communication protocol is HTTP and thus a sufficiently strong SSL connection is appropriate.

Let’s now instantiate the local key management facility:

>>> localKeys = facility.LocalKeyManagementFacility('http://localhost/keys')
>>> localKeys
<LocalKeyManagementFacility 'http://localhost/keys'>

The argument to the constructor is the URL to the master key management facility. The local facility will use a small REST API to communicate with the server.

For the purpose of this test, we are going to install a network component that only simulates the requests:

>>> from keas.kmi import testing
>>> testing.setupRestApi(localKeys, keys)

As with the master facility, the local facility provides the IEncryptionService interface:

>>> verify.verifyObject(interfaces.IEncryptionService, localKeys)

So en- and decryption is very easy to do:

>>> encrypted = localKeys.encrypt(key, b'Stephan Richter')
>>> len(encrypted)
>>> localKeys.decrypt(key, encrypted) == b'Stephan Richter'

Instead of forwarding the en- an decryption request to the master facility, the local facility merely fetches the encryption key pair and executes the operation locally. This approach has the following advantages:

  1. There is no general network latency for any en- and decryption call.
  2. The expensive task of en- and decrypting a message is delegated to multiple servers, allowing better scaling.
  3. Fetched keys can be cached locally decreasing the network calls to a once in a while process.

In this implementation, we do cache the keys in a private attribute:

>>> key in localKeys._LocalKeyManagementFacility__cache

A timeout (in seconds) controls when a key must be refetched:

>>> localKeys.timeout

Let’s now force a reload by setting the timeout to zero:

>>> localKeys.timeout = 0

The cache is a dictionary of key encrypting key to a 3-tuple that contains the date/time the key has been fetched, the encryption (public) key, and the decryption (private) key.

>>> firstTime = localKeys._LocalKeyManagementFacility__cache[key][0]
>>> localKeys.decrypt(key, encrypted) == b'Stephan Richter'
>>> secondTime = localKeys._LocalKeyManagementFacility__cache[key][0]
>>> firstTime < secondTime

The local facility also provides the IKeyGenerationService interface:

>>> verify.verifyObject(interfaces.IKeyGenerationService, keys)

The local method call is identical to the master one:

>>> key2 = localKeys.generate()
>>> print(key2.decode())

The operation is forwarded to the master server, so that the key is available there as well:

>>> hash = md5(key2)
>>> hash.hexdigest() in keys


The REST API of the master key management facility defines the communication with the local facility. When a new encryption key pair is created, we simply make a POST call to the following URL:


The request should have no body and the response is simply the key encrypting key.

So let’s have a look at the call:

>>> from keas.kmi import rest
>>> from webob import Request
>>> request = Request({})
>>> key3 = rest.create_key(keys, request).body
>>> print(key3.decode())

The key is available in the facility of course:

>>> hash = md5(key3)
>>> hash.hexdigest() in keys

We can now fetch the encryption key pair using a POST call to this URL:


The request sends the key encrypting key in its body. The response is the encryption key string:

>>> request = Request({})
>>> request.body = key3
>>> encKey = rest.get_key(keys, request)
>>> len(encKey.body)

If you try to request a nonexistent key, you get a 404 error: encryption key string:

>>> request.body = b'xxyz'
>>> print(rest.get_key(keys, request))
Key not found

A GET request to the root shows us a server status page

>>> print(rest.get_status(keys, Request({})))
200 OK
Content-Type: text/plain
Content-Length: 25
KMS server holding 3 keys

The Testing Key Management Facility

The testing facility only manages a single key that is always constant. This allows you to install a testing facility globally, not storing the keys in the database and still reuse a ZODB over multiple sessions.

>>> storage_dir = tempfile.mkdtemp()
>>> testingKeys = testing.TestingKeyManagementFacility(storage_dir)

Of course, the key generation service is supported:

>>> verify.verifyObject(interfaces.IKeyGenerationService, keys)

However, you will always receive the same key:

>>> def getKeySegment(key):
...     return str(key.decode().split('\n')[1])
>>> getKeySegment(testingKeys.generate())
>>> getKeySegment(testingKeys.generate())
>>> storage_dir = tempfile.mkdtemp()
>>> testingKeys = testing.TestingKeyManagementFacility(storage_dir)
>>> getKeySegment(testingKeys.generate())

All other methods remain the same:

>>> key = testingKeys.generate()
>>> testingKeys.getEncryptionKey(key) == b'_\xc4\x04\xbe5B\x7f\xaf\xd6\x92\xbd\xa0\xcf\x156\x1d\x88=p9{\xaal\xb4\x84M\x1d\xfd\xb2z\xae\x1a'

We can also safely en- and decrypt:

>>> encrypted = testingKeys.encrypt(key, b'Stephan Richter')
>>> testingKeys.decrypt(key, encrypted) == b'Stephan Richter'

Key Holder

The key holder is a simple class designed to store a key in RAM:

>>> from keas.kmi import keyholder
>>> holder = keyholder.KeyHolder(__file__)
>>> verify.verifyObject(interfaces.IKeyHolder, holder)

Encrypted Persistent Objects

This package provides an EncryptedPersistent class that takes care of data encryption in the storage. Usage is pretty simple: instead of subclassing persistent.Persistent, subclass keas.kmi.persistent.EncryptedPersistent:

>>> from keas.kmi.persistent import EncryptedPersistent
>>> class UserPrivateData(EncryptedPersistent):
...     def __init__(self, name, ssn):
... = name
...         self.ssn = ssn
...     def __repr__(self):
...         return '<UserPrivateData %s %s>' % (, self.ssn)
>>> userdata = UserPrivateData('Stephan Richter', '123456789')
>>> userdata
<UserPrivateData Stephan Richter 123456789>

The key used for encryption and decryption comes from an IKeyHolder utility that you’re supposed to provide in your application.

>>> from keas.kmi.testing import TestingKeyHolder
>>> from zope.component import provideUtility
>>> provideUtility(TestingKeyHolder())

None of the raw data appears in the pickle

>>> from zodbpickle import pickle
>>> pickled_data = pickle.dumps(userdata)
>>> b'Stephan' in pickled_data
>>> b'123456789' in pickled_data

We can successfully load it

>>> pickle.loads(pickled_data)
<UserPrivateData Stephan Richter 123456789>

Every persistent object is stored separately. Only the objects that inherit from EncryptedPersistent will be encrypted.

>>> import persistent.dict
>>> users = persistent.dict.PersistentDict()
>>> users['stephan'] = UserPrivateData('Stephan Richter', '123456789')
>>> users['mgedmin'] = UserPrivateData('Marius Gedminas', '987654321')
>>> pickled_data = pickle.dumps(users)
>>> b'stephan' in pickled_data
>>> b'123456789' in pickled_data

Persistent References

Enough pickling; we really should make sure our magic does not interfere with ZODB keeping track of persistent object references.

First let’s make our EncryptedPersistent objects have some references to other (encrypted and unencrypted) persistent objects

>>> users['stephan'].__parent__ = users
>>> users['mgedmin'].__parent__ = users
>>> users['stephan'].friend = users['mgedmin']
>>> users['mgedmin'].friend = users['stephan']

Now let’s create a database:

>>> import ZODB.DB
>>> import ZODB.MappingStorage
>>> db = ZODB.DB(ZODB.MappingStorage.MappingStorage())
>>> conn =
>>> conn.root()['users'] = users
>>> import transaction
>>> transaction.commit()

And we can open a second connection (while carefully keeping the first one open, to ensure it’s not reused and we actually load the pickles rather than receiving persistent objects from a cache) and load the whole object graph

>>> conn2 =
>>> users2 = conn2.root()['users']
>>> users2['stephan']
<UserPrivateData Stephan Richter 123456789>
>>> users2['mgedmin']
<UserPrivateData Marius Gedminas 987654321>

All the object references between persistent and encrypted persistent objects are preserved correctly:

>>> users2['stephan'].friend
<UserPrivateData Marius Gedminas 987654321>
>>> users2['mgedmin'].friend
<UserPrivateData Stephan Richter 123456789>
>>> users2['stephan'].__parent__ is users2
>>> users2['mgedmin'].__parent__ is users2
>>> users2['stephan'].friend is users2['mgedmin']
>>> users2['mgedmin'].friend is users2['stephan']

Data conversion

If you used to have simple persistent objects, and now want to convert them to EncryptedPersistent, think again. This is not secure. You already have unencrypted bits on your disk platters, and the only way to get rid of them is to physically destroy the disk.

But if you have a testing-only database with fake data, and would like to continue using it with a small conversion step, you can use the convert_object_to_encrypted() function.

>>> from keas.kmi.persistent import convert_object_to_encrypted

Here’s the old class definition that we’ll store:

>>> from persistent import Persistent
>>> class Password(Persistent):
...     def __init__(self, password):
...         self.password = password
>>> db = ZODB.DB(ZODB.MappingStorage.MappingStorage())
>>> conn =
>>> conn.root()['pwd'] = Password('xyzzy')
>>> transaction.commit()

And now we redefine the class:

>>> class Password(EncryptedPersistent):
...     def __init__(self, password):
...         self.password = password

Once again we have to use a different connection object (while keeping the first one alive) to avoid stepping on a ZODB cache:

>>> conn2 =
>>> pwd = conn2.root()['pwd']

If you try to use Password objects loaded from the database, you’ll get an error:

>>> pwd.password
Traceback (most recent call last):
ValueError: not enough values to unpack (expected 2, got 1)

But we can apply the conversion step:

>>> convert_object_to_encrypted(pwd)
>>> pwd.password

The converted state is stored in the DB

>>> transaction.commit()
>>> conn3 =
>>> pwd = conn3.root()['pwd']
>>> pwd.password


3.2.1 (2018-10-15)

  • The default initializeVector is now a byte string.

3.2.0 (2017-05-16)

  • Add support for Python 3.4, 3.5, 3.6.

3.1.1 (2017-04-19)

  • Removed all version constraints from buildout.cfg, since the app works fine with all the latest versions.

3.1.0 (2016-04-22)

  • Add an implementation of encrypt_file and decrypt_file. This allows chunked encoding and decoding of files. [pcdummy]

3.0.1 (2016-04-05)

  • Bumped the setuptools version in buildout.cfg.

3.0.0 (2014-01-06)

  • Switched from M2Crypto to PyCrypto, since M2Crypto is not maintained anymore.
  • Switched from deprecated repoze.bfg to pyramid.

NOTE: While I found code online to make the switch from PyCrypto to M2Crypto backwards compatible, I have not tested that functionality. Please try this on your data and let me know if you have issues.

NOTE 2: PyCrypto does not allow 512-bit RSA keys, so I increased the key size to 2048 bits. Old 512-bit keys should still work, but new ones will be always larger now.

2.1.0 (2010-10-07)

  • Added a cache for unencrypted DEKs in the key management facility, like it was already done for the local key management facility. This increases encryption and decryption performance by an order of magnitude from roughly 2ms to 0.2ms.

2.0.0 (2010-09-29)

  • Refactored REST server to be a simple repoze.bfg application.
  • The encrypted data encrypting keys (DEKs) are now stored in a directory instead of the ZODB. This increases transparency in the data store and makes backups easier.
  • Added caching to directory-based facility, so we do not need to read files all the time.

1.1.1 (2010-08-27)

  • Fixed deprecation warnings about md5 and zope.testing.doctest.

1.1.0 (2010-08-25)

  • Feature: Updated code to work with Bluebream 1.0b3.

1.0.0 (2009-07-24)

  • Feature: Update to the latest package versions.

0.3.1 (2008-09-11)

  • Relax M2Crypto version requirements to 0.18 or newer.

0.3.0 (2008-09-04)

  • A simple KeyHolder utility is available in keas.kmi.keyholder.

0.2.0 (2008-09-04)

  • Sample server shows how to enable SSL
  • Front page now shows the number of stored keys instead of a ComponentLookupError message.
  • Command-line client for testing a remote Key Management Server
  • Bugfix: LocalKeyManagementFacility was broken (AttributeError: ‘RESTClient’ object has no attribute ‘POST’)

0.1.0 (2008-09-03)

  • Initial Release
    • Key Generation Service
    • Encryption Service (Master and Local)
    • REST API for key communication between encryption services
    • Encrypted Persistent Storage

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for keas.kmi, version 3.2.1
Filename, size File type Python version Upload date Hashes
Filename, size keas.kmi-3.2.1.tar.gz (30.2 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page