Using a combination of IAM roles, S3 buckets, and EC2 it is possible to use AWS as a trusted-third-party for distributing secret or otherwise sensitive data.
IAM roles allow specifying snippets of IAM policies in a way that can be used from an EC2 virtual machine. Combined with a private S3 bucket, this can be used to authorize specific hosts to specific files.
IAM Roles can be created in the AWS Console. While the policies applied to a role can be changed later, the name cannot so be careful when choosing them.
This cookbook requires Chef 11.8 or newer. It also requires the EC2 ohai plugin to be active. If you are using a VPC, this may require setting the hint file:
$ mkdir -p /etc/chef/ohai/hints
$ touch /etc/chef/ohai/hints/ec2.json
If you use knife-ec2 to start the instance, the hint file is already set for you.
By default, your role will not be able to access any files in your private S3 bucket. You can create IAM policies that whitelist specific keys for each role:
{
"Version": "2008-10-17",
"Id": "<policy name>",
"Statement": [
{
"Sid": "<statement name>",
"Effect": "Allow",
"Principal": {
"AWS": "arn:aws:iam::<AWS account number>:role/<role name>"
},
"Action": "s3:GetObject",
"Resource": "arn:aws:s3:::<bucket name>/<key pattern>"
}
]
}
The key pattern can include *
and ?
metacharacters, so for example
arn:aws:s3:::myapp.citadel/deploy_keys/*
to allow access to all files in the
deploy_keys
folder.
This policy can be attached to either the IAM role or the S3 bucket with equal effect.
Each EC2 VM can only be assigned a single IAM role. This can complicate situations where some secrets need to be shared by overlapping subsets of your servers. A possible improvement to this would be to make a script to create all needed composite IAM roles, possibly driven by Chef roles or other metadata.
node['citadel']['bucket']
– The default S3 bucket to use.
You can access secret data via the citadel
method.
file '/etc/secret' do
owner 'root'
group 'root'
mode '600'
content citadel['keys/secret.pem']
end
By default the node attribute node['citadel']['bucket']
is used to find the
S3 bucket to query, however you can override this:
template '/etc/secret' do
owner 'root'
group 'root'
mode '600'
variables secret: citadel('mybucket')['id_rsa']
end
While developing in a local VM, you can use the node attributes
node['citadel']['access_key_id']
and node['citadel']['secret_access_key']
to provide credentials. The recommended way to do this is via environment variables
so that the Vagrantfile itself can still be kept in source control without
leaking credentials:
config.vm.provision :chef_solo do |chef|
chef.json = {
citadel: {
access_key_id: ENV['ACCESS_KEY_ID'],
secret_access_key: ENV['SECRET_ACCESS_KEY'],
},
}
end
WARNING: Use of these attributes in production should be considered a likely security risk as they will end up visible in the node data, or in the role/environment/cookbook that sets them. This can be mitigated using Enterprise Chef ACLs, however such configurations are generally error-prone due to the defaults being wide open.
Within your S3 bucket I recommend you create one folder for each group of secrets, and in your IAM policies have one statement per group. Each group of secrets is a set of data with identical security requirements. Many groups will start out only containing a single file, however having the flexibility to change this in the future allows for things like key rotation without rewriting all of your IAM policies.
You can use any S3 client you prefer to manage your secrets, however make sure that new files are set to private (accessible only to the creating user) by default.
While citadel uses HTTPS, Chef does not verify certificates by default. You can
enable verification by adding ssl_verify_mode :verify_peer
to your client.rb.