ejson-kms

ejson-kms is a utility for managing a collection of secrets in source control using AWS KMS.

The secrets are encrypted using secret-key cryptography (NaCl Secretbox: XSalsa20 + Poly1305), using key wrapping with a master key stored on HSM-backed storage at AWS. Encrypted secrets are stored in a JSON file that can easily be shared and versioned.

Quick start

1. Install ejson-kms (see Installation)
2. Create a KMS master key on AWS
3. Create a secrets file with ejson-kms init --kms-key-id="alias/MyKMSKey"
4. Add an encrypted secret with ejson-kms add secret
5. Use the decrypted credential in your bash scripts with eval "$(ejson-kms export)"

What is it

Software systems often need access to some shared credential. For example, your web application needs access to a database password, or an API key for some third party service.

ejson-kms is a simple tool that provides secure credential storage using a JSON file stored and versioned with your project, while delegating the security of your credentials to AWS KMS.

The main benefits provided by ejson-kms are:

• Secrets can be safely stored in a git repository.
• Changes to secrets are auditable on a line-by-line basis with git blame.
• Any number of access control policies can be implemented using IAM roles
• Usage audit is possible using encryption contexts and CloudTrail
• Secrets change synchronously with application source (as opposed to secrets provisioned by Configuration Management).
• Simple, well-tested, easily-auditable source.

How it works

File format

Secrets are stored in a JSON file with the following schema:

{
  "kms_key_id": "arn:aws:kms:eu-west-1:000123456789:alias/ejson-kms",
  "version": 1,
  "encryption_context": {
    "KEY": "VALUE"
  },
  "secrets": [
    {
      "name": "secret",
      "description": "Nuclear launch codes",
      "ciphertext": "EJK1;AQEDAHhZurRVk3ZWIqpympXccBmx1cOFJmQj8RBnIk01CJMnTAAAAH4wfAYJKoZIhvcNAQcGoG8wbQIBADBoBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDCgMdhWkV5uphiD5DQIBEIA7kkXS7izLJ9X4x5spWTqWjLmSY/dtcQBeaXSlzcQA5Hqd+dvdMShqcEvd3RfUzZGR89qZYzrTsfybRA4=;4vukkc/Z8K9rSkex+s7XpphIIldqNcPJjhvhOixiq6RM7pMfijLMsvTNEyQKLw=="
    }
  ]
}

Encryption context

AWS gives us the ability to store an arbitrary context with each secret, in the form of key-value pairs.

These key-value pairs are stored with the secret and are logged in CloudTrail for each encryption/decryption operation. You can use it for auditing purposes by adding, for example, the name of the project or the type of environment (production, staging, ...). Additionally, you can use it to further restrict access to your credentials with IAM policies.

Note: Since the context is stored with the secret and authenticated, it is read-only: once a secret has been encrypted, you cannot change the context.

Secret encryption

• For each secret, a data key is requested from AWS KMS (see GenerateDataKey).
• The encryption context is sent along and is stored alongside the encrypted data key. The name of the secret is added to the context automatically under the key Secret
• KMS returns the data key encrypted with the master key (stored on AWS servers) and the corresponding plaintext.
• That data key is used to encrypt one secret.
• A random nonce is generated and NaCL Secretbox is used for encryption.
• Under the hood, Secretbox uses XSalsa20 and Poly1305 to encrypt and authenticate messages. The length of messages is not hidden.
• Finally, the encrypted data key, the random nonce and the encrypted secret are each stored in the JSON file.

Secrets decryption

• For each secret, the encrypted data key, random nonce and encrypted secret are extracted from the JSON file.
• A request is made to AWS KMS to decrypt the encrypted data key (see Decrypt).
• At this stage the encryption context is authenticated and logged. The name of the secret is added to the context automatically under the key Secret
• Using the key plaintext and random nonce, the secret is decrypted using NaCL Secretbox.

Comparison with other tools

ejson

ejson is the main inspiration for this project. ejson-kms is thus very similar but differs in a few ways:

• While ejson uses public-key cryptography with keys stored on the filesystem, ejson-kms uses secret-key cryptography, key wrapping and a master key stored on HSM-backed storage in AWS servers.
• ejson has a free-form file format, any kind of schema can be implemented since it will encrypt all json keys that do not start with a _. ejson-kms has a fixed schema.
• ejson encourages secrets to be written in plaintext on the filesystem during encryption (You first add the plaintext to the file, then encrypt it). ejson-kms never writes your secrets in plaintext anywhere.

You can learn more about ejson in the write-up published on Shopify's blog here: https://engineering.shopify.com/79963908-secrets-at-shopify-introducing-ejson

credstash

credstash is the source of inspiration for ejson-kms's encryption workflow. It differs in a few ways:

• credstash depends on DynamoDB for credential storage, ejson-kms uses the filesystem. Consequently, auditing and versioning protocols are very different.
• credstash is a python tool and has a few dependencies. ejson-kms is distributed as a single, statically-linked binary.

You can learn more about credstash and KMS encryption here: https://blog.fugue.co/2015-04-21-aws-kms-secrets.html

Ansible Vault

Ansible Vault is a feature of Ansible designed to store secrets alongside playbooks. However, it uses whole-file encryption, meaning changes are not easily auditable and using git blame is impossible.

Hashicorp Vault

Hashicorp Vault is a tool designed to store secrets but also a number of features that are out of scope of ejson-kms such as certificate generation or AWS credentials management.

It is a complicated daemon you have to keep running in your infrastructure, as opposed to ejson-kms which you only have to install and use on the CLI.

Dependencies

ejson-kms as a binary has no dependencies.

It depends, however, on AWS KMS for cryptography.

Installation

Binaries for Linux, OSX and Windows are provided in the GitHub Releases page.

The binaries are signed using the GPG key C248DE6357445D6302F9A62E74BFD03C20CC21AF.

Linux

You can install ejson-kms on Linux with the following:

export EJSON_KMS_VERSION="4.2.0"
curl -Lo ejson-kms https://github.com/adrienkohlbecker/ejson-kms/releases/download/$EJSON_KMS_VERSION/ejson-kms-$EJSON_KMS_VERSION-linux-amd64
curl -Lo ejson-kms.asc https://github.com/adrienkohlbecker/ejson-kms/releases/download/$EJSON_KMS_VERSION/ejson-kms-$EJSON_KMS_VERSION-linux-amd64.asc
gpg --keyserver keyserver.ubuntu.com --recv-keys C248DE6357445D6302F9A62E74BFD03C20CC21AF
gpg --verify ejson-kms.asc
chmod +x ejson-kms

OSX

You can install ejson-kms using Homebrew:

brew install adrienkohlbecker/ejson-kms/ejson-kms

From source

ejson-kms requires a working Go 1.6+ installation.

You can install ejson-kms from source with:

go get -u github.com/adrienkohlbecker/ejson-kms

Usage

Detailed usage instructions are available in the doc folder.

init

Create an empty secrets file with ejson-kms init --kms-key-id="alias/MyKMSKey".

• Change the path of the file (by default ./.secrets.json) with --path=my_secrets.json
• Provide an encryption context with --encryption-context="key1=value1,key2=value2"

Note: the encryption context is read-only. If you want to change it you will have to start over with an empty file. A facility to edit the context might be added to ejson-kms in the future (contributions welcome!)

add

Add a secret with ejson-kms add SECRET_NAME

• ejson-kms will ask you to type the secret at runtime.
• Alternatively, you can use the form echo "password" | ejson-kms add secret, but be mindful of your bash history if you do so.
• To store the contents of a file (such as a TLS key), use cat tls.key | ejson-kms add tls_key
• Optionally, you can provide a description for this secret using --description="Nuclear launch codes". Use it to describe what the secret is used for, how to rotate it...
• The name of the credential can include lower-case letters, digits, and underscores. They cannot start with numbers (for compatibility with bash on export). Valid names: password, api_key, secret_123. Invalid names: Password, API KEY, 123-secret.

rotate

Rotate the value of a secret with ejson-kms rotate SECRET_NAME

• Secret entry is identical to the add command
• The secret will first be decrypted to check if the values are indeed different

rotate-kms-key

To rotate the KMS master key used in a secrets file, use ejson-kms rotate-kms-key NEW_KMS_KEY_ID.

Every secret will be decrypted with the old key, encrypted with the new key and the file will be overwritten.

export

To use your decrypted secrets, you can export them in a few formats with ejson-kms export --format=bash. The export will be output to standard out.

Currently there are four formats supported:

• bash: SECRET='password' (name is capitalized, value as-is except escaping of ' with '')
• dotenv: SECRET="password" (name is capitalized, value uses escape sequences (\t, \n, \xFF, \u0100) for non-ASCII characters and non-printable characters)
• json: { "secret": "password" }
• yaml: secret: password

To use in a bash script, do the following:

#!/bin/bash

eval "$(ejson-kms export)"

echo "$SECRET"

AWS authentication

ejson-kms will look for AWS credentials in the following locations and order:

• Environment variables:
  • Access Key ID: AWS_ACCESS_KEY_ID or AWS_ACCESS_KEY
  • Secret Access Key: AWS_SECRET_ACCESS_KEY or AWS_SECRET_KEY
  • Region: AWS_REGION
  • If using an IAM role: AWS_SESSION_TOKEN
• Shared credentials file:
  • If AWS_SHARED_CREDENTIALS_FILE is set, this path will be used
  • Otherwise $HOME/.aws/credentials on Linux/OSX and %USERPROFILE%\.aws\credentials on Windows
  • An AWS profile can be set with the AWS_PROFILE environment variable, otherwise it will use the default profile.
  • If you need to load the AWS region from ~/.aws/config, you need to set AWS_SDK_LOAD_CONFIG=true, otherwise you need to set AWS_REGION
• Instance profile: On EC2 instances with an assigned instance role

See also the AWS SDK session documentation.

IAM policies

Below are the basic IAM policies needed to give access to ejson-kms to a user. More complex policies can be devised, especially using encryption contexts. Refer to the documentation of AWS KMS and IAM for more information.

Secret writer

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Action": [
        "kms:GenerateDataKey"
      ],
      "Effect": "Allow",
      "Resource": "arn:aws:kms:us-east-1:AWSACCOUNTID:key/KEY-GUID"
    }
  ]
}

Secret reader

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Action": [
        "kms:Decrypt"
      ],
      "Effect": "Allow",
      "Resource": "arn:aws:kms:us-east-1:AWSACCOUNTID:key/KEY-GUID"
    }
  ]
}

Secret reader scoped by Encryption Context

{
  "Version": "2012-10-17",
  "Statement": {
    "Effect": "Allow",
    "Action": [
      "kms:Decrypt"
    ],
    "Resource": "arn:aws:kms:us-east-1:AWSACCOUNTID:key/KEY-GUID",
    "Condition": {
      "StringEquals": {
        "kms:EncryptionContext:MY_KEY": "MY_VALUE"
      }
    }
  }
}

Versioning

ejson-kms follows Semantic Versioning. For the versions available, see the releases on this repository.

Authors

Adrien Kohlbecker

See also the list of contributors who participated in this project.

License

ejson-kms is licensed under the MIT License - see the LICENSE file for details

Acknowledgments

This project was heavily inspired by ejson and credstash.