Skip to content

Latest commit

 

History

History
297 lines (222 loc) · 15.4 KB

README.md

File metadata and controls

297 lines (222 loc) · 15.4 KB

Armadillo - Encrypted Shared Preference

Armadillo Logo

A shared preference implementation for secret data providing confidentiality, integrity and authenticity . Per default uses AES-GCM, BCrypt and HKDF as cryptographic primitives.

Download Build Status Javadocs Coverage Status Maintainability

Important Notice: If you migrate to v0.6.0 and use a user password and default key stretching function migration is needed due to a security issue. See migration guide in the changelog for v0.6.0

Features

  • No-Nonse State-of-the-Art Crypto: Authenticated Encryption with AES-GCM, key derivation functions Bcrypt and HKDF
  • Flexible: Tons of nobs and switches while having sane defaults
  • Modular: use your own implementation of symmetric cipher, key stretching, data obfuscation, etc.
  • Lightweight: No massive dependencies required like BouncyCastle or Facebook Conceal

Quick Start

Add the following to your dependencies (add jcenter to your repositories if you haven't)

compile 'at.favre.lib:armadillo:x.y.z'

A very minimal example

SharedPreferences preferences = Armadillo.create(context, "myPrefs")
    .encryptionFingerprint(context)
    .build();

preferences.edit().putString("key1", "stringValue").commit();
String s = preferences.getString("key1", null);

Advanced Example

The following example shows some of the configurations available to the developer:

String userId = ...
SharedPreferences preferences = Armadillo.create(context, "myCustomPreferences")
        .password("mySuperSecretPassword".toCharArray()) //use user provided password
        .securityProvider(Security.getProvider("BC")) //use bouncy-castle security provider
        .keyStretchingFunction(new PBKDF2KeyStretcher()) //use PBKDF2 as user password kdf
        .contentKeyDigest(Bytes.from(getAndroidId(context)).array()) //use custom content key digest salt
        .secureRandom(new SecureRandom()) //provide your own secure random for salt/iv generation
        .encryptionFingerprint(context, userId.getBytes(StandardCharsets.UTF_8)) //add the user id to fingerprint
        .supportVerifyPassword(true) //enables optional password validation support `.isValidPassword()`
        .enableKitKatSupport(true) //enable optional kitkat support
        .enableDerivedPasswordCache(true) //enable caching for derived password making consecutive getters faster
        .build();

A xml file named like f1a4e61ffb59c6e6a3d6ceae9a20cb5726aade06.xml will be created with the resulting data looking something like that after the first put operation:

<?xml version='1.0' encoding='utf-8' standalone='yes' ?>
<map>
    <!-- storage random salt -->
    <string name="585d6f0f415682ace841fb50d5980d60ed23a2ef">riIPjrL2WRfoh8QJXu7fWk4GGeAKlQoJl9ofABHZKlc=</string>
    <!-- 'key1':'stringValue' -->
    <string name="152b866fd2d63899678c21f247bb6df0d2e38072">AAAAABD/8an1zfovjJB/2MFOT9ncAAAAJaf+Z9xgzwXzp1BqTsVMnRZxR/HfRcO8lEhyKpL17QmZ5amwAYQ=</string>
</map>

KitKat Support

Unfortunately Android SDK 19 (KITKAT) does not fully support AES GCM mode. Therefore a backwards compatible implementation of AES using CBC with Encrypt-then-MAC can be used to support this library on older devices. This should provide the same security strength as the GCM version, however the support must be enabled manually:

SharedPreferences preferences = Armadillo.create(context, "myCustomPreferences")
        .enableKitKatSupport(true)
        ...
        .build();

In this mode, if on a KitKat device the backwards-compatible implementation is used, the default AES-GCM version otherwise. Upgrading to a newer OS version the content should still be decryptable, while newer content will then be encrypted with the AES-GCM version.

Description

Design Choices

  • AES + GCM block mode: To make sure that the data is not only kept confidential, but it's integrity also preserved, the authenticated encryption AES+GCM is used. GCM can be implemented efficiently and fast and is the usually alternative to encrypt then mac with AES+CBC and HMAC. The authentication tag is appended to the message and is 16 byte long in this implementation. A downside of GCM is the requirement to never reuse a IV with the same key, which is avoided in this lib.
  • Every put operation creates a different cipher text: Every put operation generates new salts, iv so the the resulting cipher text will be unrecognizably different even with the same underlying data. This makes it harder to check if the data actually has changed.
  • KDFs with Key Stretching features for user passwords: Add brute-force protection to possibly weak user provided passwords.
  • Minimum SDK 19 (Android 4.4): A way to increase security is to cap older implementation. SDK 19 seems to be a good compromise where most of the older security hack fixes are not necessary anymore, but still targeting most devices.
  • Use of JCA as Provider for cryptographic primitives: Various security frameworks exists in Java: BouncyCastle, Conscrypt, Facebook Conceal. The problem is that these libraries are usually huge and require manual updates to have all the latest security fixes. This library however depends on the default JCA provider (although the developer may choose a different one). This puts trust in the device and it's implementation, while expecting frequent security patches. Usually the default provider since KitKat is AndroidOpenSSL provider which is fast (probably hardware accelerated for e.g. AES) and heavily used by e.g. TLS implementation.
  • Android Keystore System is not used: In my humble opinion, the Android Keystore is the best possible way to secure data on an Android device. Unfortunately, due to the massive fragmentation in the ecosystem properly handling and using the Android Keystore System is not easy and has some major drawbacks. Due to working in a security relevant field I have a lot of experience with this technology, therefore the decision was made to not support it for this implementation i.e. to keep it simple.
  • Use of data obfuscation: To disguise the actual data format and appear as a pseudo random byte array, obfuscation is used. This deliberately uses non standard ways to make it a bit harder to reverse engineer.

User provided Passwords

A high entropy value not known to any system but the user is a good and strong base for a cryptographic key. Unfortunately user-based passwords are often weak (low-entropy). To mitigate that fact and help preventing easy brute-forcing key derivation functions with key stretching properties are used. These functions calculate pseudo-random data from it's source material which requires mandatory work.

The following implementations are available:

  • BCrypt: based on blowfish, has a variable cpu cost parameter and a fixed memory cost parameter (default)
  • PBKDF2: applies a pseudorandom function, such as hash-based message authentication code (HMAC), to the input password or passphrase along with a salt value and repeats the process many times to produce a derived key; no memory hardness

It is possible to provide any KDF implementation to the storage with providing a custom KeyStretchingFunction implementation.

Note, if you use key stretching put/get operations will get very slow (depeding on the work factor of course), so consider accessing the store in a background thread.

Encryption Fingerprint

This store bases part of it's security on so called fingerprinting. That basically means, during runtime entropy from e.g. the device, system or other parts are used to create a cryptographic key with which the data is encrypted. It basically is encryption with a semi-secret key.

This has the following benefits:

  • Binding the data to the executing runtime (ie. making it harder to lift the data and trying to read it in a different environment)
  • Strongly obfuscating the data bordering actual encryption when the used fingerprint is infeasible to guess
  • Be able to scope the data to a specific environment (e.g. when using the Android OS image build number, every update invalidates the data)

This store has a default implementation of EncryptionFingerprint which can only use generic data. In detail the following properties are incorporated:

  • Fingerprint of the APK signature
  • Android ID : a 64-bit number (expressed as a hexadecimal string) byte random value; on SDK 26 and higher, unique to each combination of app-signing key, user, and device - on SDK 25 lower only unique to user and device
  • Application package name, brand, model and name of the device
  • 32 byte hardcoded static random value

Enhancing the Strength of the Encryption Fingerprint

The security of this mechanism increases considerably if the user adds it's own data. Here are some suggestions:

  • Random values hardcoded, locally generated or provided by a remote service
  • Unique user-id (if the application has the concept of login)
  • Device Serial (requires dangerous permission SDK > 25)
  • Sim-ID/ICCID (if changing the sim should/can invalidate the data)
  • Android OS image build fingerprint (if you want to invalidate the data after OS update)

Key Derivation Process

The cryptographic key used to encrypt the data is composed of the following parts:

screenshot key derivation

  • User password (optional): provided by the caller and stretched with e.g. Bcrypt
  • Encryption Fingerprint (see section above)
  • Entry Key: the hashed version of the key passed by the caller; this will bind the data to that specific entry key
  • Entry Salt: a random 16 byte value unique to that specific entry that will be created on every put operation (will also be used for the key stretching function)
  • Storage Salt: a random 32 byte value unique to that specific storage, created on first creation of the storage

The concatenated key material will be derived and stretched to the desired length with HKDF derivation function.

Persistence Profile

Key

The key is hashed with HKDF (which uses Hmac with Sha512 internally) expanded to a 20 byte hash which will be encoded with base16 (hex). The key generation is salted by the encryption fingerprint, so different shared preferences will generate different hashes for the same keys.

Content

The diagram below illustrates the used data format. To disguise the format a little bit it will be obfuscated by a simple xor cipher.

screenshot gallery

The resulting data will be encoded with base64 and looks like this in the shared preferences xml:

<?xml version='1.0' encoding='utf-8' standalone='yes' ?>
<map>
    <string name="39e3e4f83dda81c44f8a9063196b28b3d5091fca">hwbchXlqDAQcig6q3UWxdbOb2wouDGGwjUGNIzREiy0=</string>
    <string name="62ef41ac992322bdd669e96799c12a66a2cce111">AAAAABAAajtOaVCq5yqu1TPxgLu2AAAAqUTxgPcAM6lyNTGgy7ZAoCjqcCdtxT6T</string>
</map>

Digital Signatures

Signed Commits

All tags and commits by me are signed with git with my private key:

GPG key ID: 4FDF85343912A3AB
Fingerprint: 2FB392FB05158589B767960C4FDF85343912A3AB

Build

Assemble the lib with the following command

./gradlew :armadillo:assemble

The .aar files can then be found in /armadillo/build/outputs/aar folder

Libraries & Credits

Similar Projects:

Further Reading

License

Copyright 2017 Patrick Favre-Bulle

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.