Table of contents

This is for internal use by the PaaS team. Public-facing documentation is located at docs.cloud.service.gov.uk.

ADR003: AWS credentials

Context

Amazon Web Services (AWS) are our current Infrastructure as a Service (IaaS) provider. Our deployment tooling (Concourse, Terraform, BOSH, etc.) and Cloud Foundry components (Cloud Controller, RDS broker, blobstore clients, etc.) use the APIs to manage or access IaaS resources.

The most common mechanism for authenticating the API calls is to create an Identify and Access Management (IAM) user with the appropriate permissions, generate an Access Key ID and Secret Access Key for that user, and export those as environment variables. AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY are the standard environment variable names used by most utilities and libraries.

The problem with this approach is that it’s very easy to accidentally leak the plain text keys. They can appear in output from your shell, which you might copy+paste into a gist or email when debugging a problem. You might add them to your shell configuration or include them in a script, which can be pushed to a public code repository.

Our team have leaked keys like this on more than one occasion. It’s worth noting that even if you realise that you’ve done this, delete the commit and revoke the keys, they may have already been used maliciously because automated bots monitor sites like GitHub using the events firehose to detect any credentials.

As an alternative to using pre-generated keys, AWS recommends that you use IAM roles and instance profiles when accessing the API from EC2 instances. You delegate permissions to the EC2 instance and temporary credentials are made available from the instance metadata service. Most tools and libraries automatically support this. The credentials are regularly rotated and never need to be stored in configuration files.

Decision

To reduce the likelihood of us leaking AWS keys we will use IAM roles and instance profiles for all operations that run from EC2 instances. This includes everything that happens within Concourse and Cloud Foundry.

To reduce the impact of us leaking AWS keys we will use an IAM policy with an aws:SourceIp condition to enforce that IAM accounts for team members are only used from the office IP addresses.

The IAM roles, profiles, and policies will be managed by our aws-account-wide-terraform repo.

Status

Accepted

Consequences

We’ll still need to use AWS keys for operations that run outside of EC2. Care must be taken when storing and handling these credentials. These operations include:

  • Creation of Bootstrap Concourse instance
  • Running of aws-account-wide-terraform

Using IAM profiles has the drawback that any process running on the VM can get the same credentials. This model does not play well when it is required to assign the credentials to specific processes running in different containers (for example concourse, CF apps), as all the containers will have access to the AWS IAM profile.

We’ll need to maintain our own forks of some standard Concourse resources to add support for IAM roles and instance profiles because the maintainers don’t wish to support this feature (concourse/s3-resource/pull/22). These resources include:

We’ll need to use the office VPN to administer AWS when outside of the office. This matches what we have to do to administer Concourse or Cloud Foundry from outside the office. There are disaster recovery provisions for the VPN if the office has connectivity problems.