metadataproxy

The metadataproxy is used to allow containers to acquire IAM roles. By metadata we mean EC2 instance meta data which is normally available to EC2 instances. This proxy exposes the meta data to containers inside or outside of EC2 hosts, allowing you to provide scoped IAM roles to individual containers, rather than giving them the full IAM permissions of an IAM role or IAM user.

Installation

From inside of the repo run the following commands:

mkdir -p /srv/metadataproxy
cd /srv/metadataproxy
virtualenv venv
source venv/bin/activate
pip install metadataproxy
deactivate

Configuration

Modes of operation

See the settings file for specific configuration options.

The metadataproxy has two basic modes of operation:

1. Running in AWS where it simply proxies most routes to the real metadata service.
2. Running outside of AWS where it mocks out most routes.

To enable mocking, use the environment variable:

export MOCK_API=true

AWS credentials

metadataproxy relies on boto configuration for its AWS credentials. If metadata IAM credentials are available, it will use this. Otherwise, you'll need to use .aws/credentials, .boto, or environment variables to specify the IAM credentials before the service is started.

Role assumption

For IAM routes, the metadataproxy will use STS to assume roles for containers. To do so it takes the incoming IP address of metadata requests and finds the running docker container associated with the IP address. It uses the value of the container's IAM_ROLE environment variable as the role it will assume. It then assumes the role and gives back STS credentials in the metadata response.

STS-attained credentials are cached and automatically rotated as they expire.

Container-specific roles

To specify the role of a container, simply launch it with the IAM_ROLE environment variable set to the IAM role you wish the container to run with.

If the trust policy for the role requires an ExternalId, you can set this using the IAM_EXTERNAL_ID environment variable. This is most frequently used with cross-account role access scenarios. For more information on when you should use an External ID for your roles, see:

http://docs.aws.amazon.com/IAM/latest/UserGuide/id_roles_create_for-user_externalid.html

docker run -e IAM_ROLE=my-role ubuntu:14.04
docker run -e IAM_ROLE=their-role@another-account -e IAM_EXTERNAL_ID=random-unique-string ubuntu:14.04

Configurable Behavior

There are a number of environment variables that can be set to tune metadata proxy's behavior. They can either be exported by the start script, or set via docker environment variables.

Variable	Type	Default	Description
DEFAULT_ROLE	String		Role to use if IAM_ROLE is not set in a container's environment. If unset the container will get no IAM credentials.
DEFAULT_ACCOUNT_ID	String		The default account ID to assume roles in, if IAM_ROLE does not contain account information. If unset, metadataproxy will attempt to lookup role ARNs using iam:GetRole.
ROLE_SESSION_KEY	String		Optional key in container labels or environment variables to use for role session name. Prefix with Labels: or Env: respectively to indicate where key should be found. Useful to pass through metadata such as a CI job ID or launching user for audit purposes, as the role session name is included in the ARN that appears in access logs.
DEBUG	Boolean	False	Enable debug mode. You should not do this in production as it will leak IAM credentials into your logs
DOCKER_URL	String	unix://var/run/docker.sock	Url of the docker daemon. The default is to access docker via its socket.
METADATA_URL	String	http://169.254.169.254	URL of the metadata service. Default is the normal location of the metadata service in AWS.
MOCK_API	Boolean	False	Whether or not to mock all metadata endpoints. If True, mocked data will be returned to callers. If False, all endpoints except for IAM endpoints will be proxied through to the real metadata service.
MOCKED_INSTANCE_ID	String	mockedid	When mocking the API, use the following instance id in returned data.
AWS_ACCOUNT_MAP	JSON String	{}	A mapping of account names to account IDs. This allows you to use user-friendly names instead of account IDs in IAM_ROLE environment variable values.
ROLE_EXPIRATION_THRESHOLD	Integer	15	The threshold before credentials expire in minutes at which metadataproxy will attempt to load new credentials.
ROLE_MAPPING_FILE	Path String		A json file that has a dict mapping of IP addresses to role names. Can be used if docker networking has been disabled and you are managing IP addressing for containers through another process.
ROLE_REVERSE_LOOKUP	Boolean	False	Enable performing a reverse lookup of incoming IP addresses to match containers by hostname. Useful if you've disabled networking in docker, but set hostnames for containers in /etc/hosts or DNS.
HOSTNAME_MATCH_REGEX	Regex String	^.*$	Limit reverse lookup container matching to hostnames that match the specified pattern.

Default Roles

When no role is matched, metadataproxy will use the role specified in the DEFAULT_ROLE metadataproxy environment variable. If no DEFAULT_ROLE is specified as a fallback, then your docker container without an IAM_ROLE environment variable will fail to retrieve credentials.

Role Formats

The following are all supported formats for specifying roles:

• By Role:

  IAM_ROLE=my-role

• By Role@AccountId

  IAM_ROLE=my-role@012345678910

• By ARN:

  IAM_ROLE=arn:aws:iam::012345678910:role/my-role

Role structure

A useful way to deploy this metadataproxy is with a two-tier role structure:

1. The first tier is the EC2 service role for the instances running your containers. Call it DockerHostRole. Your instances must be launched with a policy that assigns this role.

2. The second tier is the role that each container will use. These roles must trust your own account ("Role for Cross-Account Access" in AWS terms). Call it ContainerRole1.

3. metadataproxy needs to query and assume the container role. So the DockerHostRole policy must permit this for each container role. For example:

   "Statement": [ {
       "Effect": "Allow",
       "Action": [
           "iam:GetRole",
           "sts:AssumeRole"
       ],
       "Resource": [
           "arn:aws:iam::012345678901:role/ContainerRole1",
           "arn:aws:iam::012345678901:role/ContainerRole2"
       ]
   } ]
   

4. Now customize ContainerRole1 & friends as you like

Note: The ContainerRole1 role should have a trust relationship that allows it to be assumed by the user which is associated to the host machine running the sts:AssumeRole command. An example trust relationship for ContainRole1 may look like:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "AWS": "arn:aws:iam::012345678901:root",
        "Service": "ec2.amazonaws.com"
      },
      "Action": "sts:AssumeRole"
    }
  ]
}

Routing container traffic to metadataproxy

Using iptables, we can forward traffic meant to 169.254.169.254 from docker0 to the metadataproxy. The following example assumes the metadataproxy is run on the host, and not in a container:

/sbin/iptables \
  --append PREROUTING \
  --destination 169.254.169.254 \
  --protocol tcp \
  --dport 80 \
  --in-interface docker0 \
  --jump DNAT \
  --table nat \
  --to-destination 127.0.0.1:8000 \
  --wait

If you'd like to start the metadataproxy in a container, it's recommended to use host-only networking. Also, it's necessary to volume mount in the docker socket, as metadataproxy must be able to interact with docker.

Be aware that non-host-mode containers will not be able to contact 127.0.0.1 in the host network stack. As an alternative, you can use the meta-data service to find the local address. In this case, you probably want to restrict proxy access to the docker0 interface!

LOCAL_IPV4=$(curl http://169.254.169.254/latest/meta-data/local-ipv4)

/sbin/iptables \
  --append PREROUTING \
  --destination 169.254.169.254 \
  --protocol tcp \
  --dport 80 \
  --in-interface docker0 \
  --jump DNAT \
  --table nat \
  --to-destination $LOCAL_IPV4:8000 \
  --wait

/sbin/iptables \
  --wait \
  --insert INPUT 1 \
  --protocol tcp \
  --dport 80 \
  \! \
  --in-interface docker0 \
  --jump DROP

Run metadataproxy without docker

In the following we assume _my_config_ is a bash file with exports for all of the necessary settings discussed in the configuration section.

source my_config
cd /srv/metadataproxy
source venv/bin/activate
gunicorn metadataproxy:app --workers=2 -k gevent

Run metadataproxy with docker

For production purposes, you'll want to kick up a container to run. You can build one with the included Dockerfile. To run, do something like:

docker run --net=host \
    -v /var/run/docker.sock:/var/run/docker.sock \
    lyft/metadataproxy

gunicorn settings

The following environment variables can be set to configure gunicorn (defaults are set in the examples):

# Change the IP address the gunicorn worker is listening on. You likely want to
# leave this as the default
HOST=0.0.0.0

# Change the port the gunicorn worker is listening on.
PORT=8000

# Change the number of worker processes gunicorn will run with. The default is
# 1, which is likely enough since metadataproxy is using gevent and its work is
# completely IO bound. Increasing the number of workers will likely make your
# in-memory cache less efficient
WORKERS=1

# Enable debug mode (you should not do this in production as it will leak IAM
# credentials into your logs)
DEBUG=False

Contributing

Code of conduct

This project is governed by Lyft's code of conduct. All contributors and participants agree to abide by its terms.

Sign the Contributor License Agreement (CLA)

We require a CLA for code contributions, so before we can accept a pull request we need to have a signed CLA. Please visit our CLA service follow the instructions to sign the CLA.

File issues in Github

In general all enhancements or bugs should be tracked via github issues before PRs are submitted. We don't require them, but it'll help us plan and track.

When submitting bugs through issues, please try to be as descriptive as possible. It'll make it easier and quicker for everyone if the developers can easily reproduce your bug.

Submit pull requests

Our only method of accepting code changes is through github pull requests.