AWS Identity and Access Management Construct Library

Security and Safety Dev Guide

For a detailed guide on CDK security and safety please see the CDK Security And Safety Dev Guide

The guide will cover topics like:

  • What permissions to extend to CDK deployments

  • How to control the permissions of CDK deployments via IAM identities and policies

  • How to use CDK to configure the IAM identities and policies of deployed applications

  • Using Permissions Boundaries with CDK

Overview

Define a role and add permissions to it. This will automatically create and attach an IAM policy to the role:

role = Role(self, "MyRole",
    assumed_by=ServicePrincipal("sns.amazonaws.com")
)

role.add_to_policy(PolicyStatement(
    resources=["*"],
    actions=["lambda:InvokeFunction"]
))

Define a policy and attach it to groups, users and roles. Note that it is possible to attach the policy either by calling xxx.attachInlinePolicy(policy) or policy.attachToXxx(xxx).

user = User(self, "MyUser", password=SecretValue.plain_text("1234"))
group = Group(self, "MyGroup")

policy = Policy(self, "MyPolicy")
policy.attach_to_user(user)
group.attach_inline_policy(policy)

Managed policies can be attached using xxx.addManagedPolicy(ManagedPolicy.fromAwsManagedPolicyName(policyName)):

group = Group(self, "MyGroup")
group.add_managed_policy(ManagedPolicy.from_aws_managed_policy_name("AdministratorAccess"))

Granting permissions to resources

Many of the AWS CDK resources have grant* methods that allow you to grant other resources access to that resource. As an example, the following code gives a Lambda function write permissions (Put, Update, Delete) to a DynamoDB table.

# fn: lambda.Function
# table: dynamodb.Table


table.grant_write_data(fn)

The more generic grant method allows you to give specific permissions to a resource:

# fn: lambda.Function
# table: dynamodb.Table


table.grant(fn, "dynamodb:PutItem")

The grant* methods accept an IGrantable object. This interface is implemented by IAM principal resources (groups, users and roles), policies, managed policies and resources that assume a role such as a Lambda function, EC2 instance or a Codebuild project.

You can find which grant* methods exist for a resource in the AWS CDK API Reference.

Roles

Many AWS resources require Roles to operate. These Roles define the AWS API calls an instance or other AWS service is allowed to make.

Creating Roles and populating them with the right permissions Statements is a necessary but tedious part of setting up AWS infrastructure. In order to help you focus on your business logic, CDK will take care of creating roles and populating them with least-privilege permissions automatically.

All constructs that require Roles will create one for you if don’t specify one at construction time. Permissions will be added to that role automatically if you associate the construct with other constructs from the AWS Construct Library (for example, if you tell an AWS CodePipeline to trigger an AWS Lambda Function, the Pipeline’s Role will automatically get lambda:InvokeFunction permissions on that particular Lambda Function), or if you explicitly grant permissions using grant functions (see the previous section).

Opting out of automatic permissions management

You may prefer to manage a Role’s permissions yourself instead of having the CDK automatically manage them for you. This may happen in one of the following cases:

  • You don’t like the permissions that CDK automatically generates and want to substitute your own set.

  • The least-permissions policy that the CDK generates is becoming too big for IAM to store, and you need to add some wildcards to keep the policy size down.

To prevent constructs from updating your Role’s policy, pass the object returned by myRole.withoutPolicyUpdates() instead of myRole itself.

For example, to have an AWS CodePipeline not automatically add the required permissions to trigger the expected targets, do the following:

role = iam.Role(self, "Role",
    assumed_by=iam.ServicePrincipal("codepipeline.amazonaws.com"),
    # custom description if desired
    description="This is a custom role..."
)

codepipeline.Pipeline(self, "Pipeline",
    # Give the Pipeline an immutable view of the Role
    role=role.without_policy_updates()
)

# You now have to manage the Role policies yourself
role.add_to_policy(iam.PolicyStatement(
    actions=[],
    resources=[]
))

Using existing roles

If there are Roles in your account that have already been created which you would like to use in your CDK application, you can use Role.fromRoleArn to import them, as follows:

role = iam.Role.from_role_arn(self, "Role", "arn:aws:iam::123456789012:role/MyExistingRole",
    # Set 'mutable' to 'false' to use the role as-is and prevent adding new
    # policies to it. The default is 'true', which means the role may be
    # modified as part of the deployment.
    mutable=False
)

Customizing role creation

It is best practice to allow CDK to manage IAM roles and permissions. You can prevent CDK from creating roles by using the customizeRoles method for special cases. One such case is using CDK in an environment where role creation is not allowed or needs to be managed through a process outside of the CDK application.

An example of how to opt in to this behavior is below:

# stack: Stack

iam.Role.customize_roles(stack)

CDK will not create any IAM roles or policies with the stack scope. cdk synth will fail and it will generate a policy report to the cloud assembly (i.e. cdk.out). The iam-policy-report.txt report will contain a list of IAM roles and associated permissions that would have been created. This report can be used to create the roles with the appropriate permissions outside of the CDK application.

Once the missing roles have been created, their names can be added to the usePrecreatedRoles property, like shown below:

# app: App

stack = Stack(app, "MyStack")
iam.Role.customize_roles(self,
    use_precreated_roles={
        "MyStack/MyRole": "my-precreated-role-name"
    }
)

iam.Role(self, "MyRole",
    assumed_by=iam.ServicePrincipal("sns.amazonaws.com")
)

If any IAM policies reference deploy time values (i.e. ARN of a resource that hasn’t been created yet) you will have to modify the generated report to be more generic. For example, given the following CDK code:

# app: App

stack = Stack(app, "MyStack")
iam.Role.customize_roles(stack)

fn = lambda_.Function(self, "MyLambda",
    code=lambda_.InlineCode("foo"),
    handler="index.handler",
    runtime=lambda_.Runtime.NODEJS_LATEST
)

bucket = s3.Bucket(self, "Bucket")
bucket.grant_read(fn)

The following report will be generated.

<missing role> (MyStack/MyLambda/ServiceRole)

AssumeRole Policy:
[
  {
    "Action": "sts:AssumeRole",
    "Effect": "Allow",
    "Principal": {
      "Service": "lambda.amazonaws.com"
    }
  }
]

Managed Policy ARNs:
[
  "arn:(PARTITION):iam::aws:policy/service-role/AWSLambdaBasicExecutionRole"
]

Managed Policies Statements:
NONE

Identity Policy Statements:
[
  {
    "Action": [
      "s3:GetObject*",
      "s3:GetBucket*",
      "s3:List*"
    ],
    "Effect": "Allow",
    "Resource": [
      "(MyStack/Bucket/Resource.Arn)",
      "(MyStack/Bucket/Resource.Arn)/*"
    ]
  }
]

You would then need to create the role with the inline & managed policies in the report and then come back and update the customizeRoles with the role name.

# app: App

stack = Stack(app, "MyStack")
iam.Role.customize_roles(self,
    use_precreated_roles={
        "MyStack/MyLambda/ServiceRole": "my-role-name"
    }
)

For more information on configuring permissions see the Security And Safety Dev Guide

Generating a permissions report

It is also possible to generate the report without preventing the role/policy creation.

# stack: Stack

iam.Role.customize_roles(self,
    prevent_synthesis=False
)

Configuring an ExternalId

If you need to create Roles that will be assumed by third parties, it is generally a good idea to require an ExternalId to assume them. Configuring an ExternalId works like this:

role = iam.Role(self, "MyRole",
    assumed_by=iam.AccountPrincipal("123456789012"),
    external_ids=["SUPPLY-ME"]
)

SourceArn and SourceAccount

If you need to create resource policies using aws:SourceArn and aws:SourceAccount for cross-service resource access, use addSourceArnCondition and addSourceAccountCondition to create the conditions.

See Cross-service confused deputy prevention for more details.

Principals vs Identities

When we say Principal, we mean an entity you grant permissions to. This entity can be an AWS Service, a Role, or something more abstract such as “all users in this account” or even “all users in this organization”. An Identity is an IAM representing a single IAM entity that can have a policy attached, one of Role, User, or Group.

IAM Principals

When defining policy statements as part of an AssumeRole policy or as part of a resource policy, statements would usually refer to a specific IAM principal under Principal.

IAM principals are modeled as classes that derive from the iam.PolicyPrincipal abstract class. Principal objects include principal type (string) and value (array of string), optional set of conditions and the action that this principal requires when it is used in an assume role policy document.

To add a principal to a policy statement you can either use the abstract statement.addPrincipal, one of the concrete addXxxPrincipal methods:

  • addAwsPrincipal, addArnPrincipal or new ArnPrincipal(arn) for { "AWS": arn }

  • addAwsAccountPrincipal or new AccountPrincipal(accountId) for { "AWS": account-arn }

  • addServicePrincipal or new ServicePrincipal(service) for { "Service": service }

  • addAccountRootPrincipal or new AccountRootPrincipal() for { "AWS": { "Ref: "AWS::AccountId" } }

  • addCanonicalUserPrincipal or new CanonicalUserPrincipal(id) for { "CanonicalUser": id }

  • addFederatedPrincipal or new FederatedPrincipal(federated, conditions, assumeAction) for { "Federated": arn } and a set of optional conditions and the assume role action to use.

  • addAnyPrincipal or new AnyPrincipal for { "AWS": "*" }

If multiple principals are added to the policy statement, they will be merged together:

statement = iam.PolicyStatement()
statement.add_service_principal("cloudwatch.amazonaws.com")
statement.add_service_principal("ec2.amazonaws.com")
statement.add_arn_principal("arn:aws:boom:boom")

Will result in:

{
  "Principal": {
    "Service": [ "cloudwatch.amazonaws.com", "ec2.amazonaws.com" ],
    "AWS": "arn:aws:boom:boom"
  }
}

The CompositePrincipal class can also be used to define complex principals, for example:

role = iam.Role(self, "MyRole",
    assumed_by=iam.CompositePrincipal(
        iam.ServicePrincipal("ec2.amazonaws.com"),
        iam.AccountPrincipal("1818188181818187272"))
)

The PrincipalWithConditions class can be used to add conditions to a principal, especially those that don’t take a conditions parameter in their constructor. The principal.withConditions() method can be used to create a PrincipalWithConditions from an existing principal, for example:

principal = iam.AccountPrincipal("123456789000").with_conditions({"StringEquals": {"foo": "baz"}})

NOTE: If you need to define an IAM condition that uses a token (such as a deploy-time attribute of another resource) in a JSON map key, use CfnJson to render this condition. See this test for an example.

The WebIdentityPrincipal class can be used as a principal for web identities like Cognito, Amazon, Google or Facebook, for example:

principal = iam.WebIdentityPrincipal("cognito-identity.amazonaws.com", {
    "StringEquals": {"cognito-identity.amazonaws.com:aud": "us-east-2:12345678-abcd-abcd-abcd-123456"},
    "ForAnyValue:StringLike": {"cognito-identity.amazonaws.com:amr": "unauthenticated"}
})

If your identity provider is configured to assume a Role with session tags, you need to call .withSessionTags() to add the required permissions to the Role’s policy document:

iam.Role(self, "Role",
    assumed_by=iam.WebIdentityPrincipal("cognito-identity.amazonaws.com", {
        "StringEquals": {
            "cognito-identity.amazonaws.com:aud": "us-east-2:12345678-abcd-abcd-abcd-123456"
        },
        "ForAnyValue:StringLike": {
            "cognito-identity.amazonaws.com:amr": "unauthenticated"
        }
    }).with_session_tags()
)

Parsing JSON Policy Documents

The PolicyDocument.fromJson and PolicyStatement.fromJson static methods can be used to parse JSON objects. For example:

policy_document = {
    "Version": "2012-10-17",
    "Statement": [{
        "Sid": "FirstStatement",
        "Effect": "Allow",
        "Action": ["iam:ChangePassword"],
        "Resource": ["*"]
    }, {
        "Sid": "SecondStatement",
        "Effect": "Allow",
        "Action": ["s3:ListAllMyBuckets"],
        "Resource": ["*"]
    }, {
        "Sid": "ThirdStatement",
        "Effect": "Allow",
        "Action": ["s3:List*", "s3:Get*"
        ],
        "Resource": ["arn:aws:s3:::confidential-data", "arn:aws:s3:::confidential-data/*"
        ],
        "Condition": {"Bool": {"aws:_multi_factor_auth_present": "true"}}
    }
    ]
}

custom_policy_document = iam.PolicyDocument.from_json(policy_document)

# You can pass this document as an initial document to a ManagedPolicy
# or inline Policy.
new_managed_policy = iam.ManagedPolicy(self, "MyNewManagedPolicy",
    document=custom_policy_document
)
new_policy = iam.Policy(self, "MyNewPolicy",
    document=custom_policy_document
)

Permissions Boundaries

Permissions Boundaries can be used as a mechanism to prevent privilege escalation by creating new Roles. Permissions Boundaries are a Managed Policy, attached to Roles or Users, that represent the maximum set of permissions they can have. The effective set of permissions of a Role (or User) will be the intersection of the Identity Policy and the Permissions Boundary attached to the Role (or User). Permissions Boundaries are typically created by account Administrators, and their use on newly created Roles will be enforced by IAM policies.

Bootstrap Permissions Boundary

If a permissions boundary has been enforced as part of CDK bootstrap, all IAM Roles and Users that are created as part of the CDK application must be created with the permissions boundary attached. The most common scenario will be to apply the enforced permissions boundary to the entire CDK app. This can be done either by adding the value to cdk.json or directly in the App constructor.

For example if your organization has created and is enforcing a permissions boundary with the name cdk-${Qualifier}-PermissionsBoundary

{
  "context": {
     "@aws-cdk/core:permissionsBoundary": {
	   "name": "cdk-${Qualifier}-PermissionsBoundary"
	 }
  }
}

OR

App(
    context={
        "PERMISSIONS_BOUNDARY_CONTEXT_KEY": {
            "name": "cdk-${Qualifier}-PermissionsBoundary"
        }
    }
)

Another scenario might be if your organization enforces different permissions boundaries for different environments. For example your CDK application may have

  • DevStage that deploys to a personal dev environment where you have elevated privileges

  • BetaStage that deploys to a beta environment which and has a relaxed permissions boundary

  • GammaStage that deploys to a gamma environment which has the prod permissions boundary

  • ProdStage that deploys to the prod environment and has the prod permissions boundary

# app: App


Stage(app, "DevStage")

Stage(app, "BetaStage",
    permissions_boundary=PermissionsBoundary.from_name("beta-permissions-boundary")
)

Stage(app, "GammaStage",
    permissions_boundary=PermissionsBoundary.from_name("prod-permissions-boundary")
)

Stage(app, "ProdStage",
    permissions_boundary=PermissionsBoundary.from_name("prod-permissions-boundary")
)

The provided name can include placeholders for the partition, region, qualifier, and account These placeholders will be replaced with the actual values if available. This requires that the Stack has the environment specified, it does not work with environment.

  • ‘${AWS::Partition}’

  • ‘${AWS::Region}’

  • ‘${AWS::AccountId}’

  • ‘${Qualifier}’

# app: App


prod_stage = Stage(app, "ProdStage",
    permissions_boundary=PermissionsBoundary.from_name("cdk-${Qualifier}-PermissionsBoundary-${AWS::AccountId}-${AWS::Region}")
)

Stack(prod_stage, "ProdStack",
    synthesizer=DefaultStackSynthesizer(
        qualifier="custom"
    )
)

For more information on configuring permissions see the Security And Safety Dev Guide

Custom Permissions Boundary

It is possible to attach Permissions Boundaries to all Roles created in a construct tree all at once:

# Directly apply the boundary to a Role you create
# role: iam.Role

# Apply the boundary to an Role that was implicitly created for you
# fn: lambda.Function

# Remove a Permissions Boundary that is inherited, for example from the Stack level
# custom_resource: CustomResource
# This imports an existing policy.
boundary = iam.ManagedPolicy.from_managed_policy_arn(self, "Boundary", "arn:aws:iam::123456789012:policy/boundary")

# This creates a new boundary
boundary2 = iam.ManagedPolicy(self, "Boundary2",
    statements=[
        iam.PolicyStatement(
            effect=iam.Effect.DENY,
            actions=["iam:*"],
            resources=["*"]
        )
    ]
)
iam.PermissionsBoundary.of(role).apply(boundary)
iam.PermissionsBoundary.of(fn).apply(boundary)

# Apply the boundary to all Roles in a stack
iam.PermissionsBoundary.of(self).apply(boundary)
iam.PermissionsBoundary.of(custom_resource).clear()

OpenID Connect Providers

OIDC identity providers are entities in IAM that describe an external identity provider (IdP) service that supports the OpenID Connect (OIDC) standard, such as Google or Salesforce. You use an IAM OIDC identity provider when you want to establish trust between an OIDC-compatible IdP and your AWS account. This is useful when creating a mobile app or web application that requires access to AWS resources, but you don’t want to create custom sign-in code or manage your own user identities. For more information about this scenario, see [About Web Identity Federation] and the relevant documentation in the [Amazon Cognito Identity Pools Developer Guide].

The following examples defines an OpenID Connect provider. Two client IDs (audiences) are will be able to send authentication requests to https://openid/connect.

provider = iam.OpenIdConnectProvider(self, "MyProvider",
    url="https://openid/connect",
    client_ids=["myclient1", "myclient2"]
)

You can specify an optional list of thumbprints. If not specified, the thumbprint of the root certificate authority (CA) will automatically be obtained from the host as described here.

Once you define an OpenID connect provider, you can use it with AWS services that expect an IAM OIDC provider. For example, when you define an Amazon Cognito identity pool you can reference the provider’s ARN as follows:

import aws_cdk.aws_cognito as cognito

# my_provider: iam.OpenIdConnectProvider

cognito.CfnIdentityPool(self, "IdentityPool",
    open_id_connect_provider_arns=[my_provider.open_id_connect_provider_arn],
    # And the other properties for your identity pool
    allow_unauthenticated_identities=False
)

The OpenIdConnectPrincipal class can be used as a principal used with a OpenIdConnectProvider, for example:

provider = iam.OpenIdConnectProvider(self, "MyProvider",
    url="https://openid/connect",
    client_ids=["myclient1", "myclient2"]
)
principal = iam.OpenIdConnectPrincipal(provider)

SAML provider

An IAM SAML 2.0 identity provider is an entity in IAM that describes an external identity provider (IdP) service that supports the SAML 2.0 (Security Assertion Markup Language 2.0) standard. You use an IAM identity provider when you want to establish trust between a SAML-compatible IdP such as Shibboleth or Active Directory Federation Services and AWS, so that users in your organization can access AWS resources. IAM SAML identity providers are used as principals in an IAM trust policy.

iam.SamlProvider(self, "Provider",
    metadata_document=iam.SamlMetadataDocument.from_file("/path/to/saml-metadata-document.xml")
)

The SamlPrincipal class can be used as a principal with a SamlProvider:

provider = iam.SamlProvider(self, "Provider",
    metadata_document=iam.SamlMetadataDocument.from_file("/path/to/saml-metadata-document.xml")
)
principal = iam.SamlPrincipal(provider, {
    "StringEquals": {
        "SAML:iss": "issuer"
    }
})

When creating a role for programmatic and AWS Management Console access, use the SamlConsolePrincipal class:

provider = iam.SamlProvider(self, "Provider",
    metadata_document=iam.SamlMetadataDocument.from_file("/path/to/saml-metadata-document.xml")
)
iam.Role(self, "Role",
    assumed_by=iam.SamlConsolePrincipal(provider)
)

Users

IAM manages users for your AWS account. To create a new user:

user = iam.User(self, "MyUser")

To import an existing user by name with path:

user = iam.User.from_user_name(self, "MyImportedUserByName", "johnsmith")

To import an existing user by ARN:

user = iam.User.from_user_arn(self, "MyImportedUserByArn", "arn:aws:iam::123456789012:user/johnsmith")

To import an existing user by attributes:

user = iam.User.from_user_attributes(self, "MyImportedUserByAttributes",
    user_arn="arn:aws:iam::123456789012:user/johnsmith"
)

Access Keys

The ability for a user to make API calls via the CLI or an SDK is enabled by the user having an access key pair. To create an access key:

user = iam.User(self, "MyUser")
access_key = iam.AccessKey(self, "MyAccessKey", user=user)

You can force CloudFormation to rotate the access key by providing a monotonically increasing serial property. Simply provide a higher serial value than any number used previously:

user = iam.User(self, "MyUser")
access_key = iam.AccessKey(self, "MyAccessKey", user=user, serial=1)

An access key may only be associated with a single user and cannot be “moved” between users. Changing the user associated with an access key replaces the access key (and its ID and secret value).

Groups

An IAM user group is a collection of IAM users. User groups let you specify permissions for multiple users.

group = iam.Group(self, "MyGroup")

To import an existing group by ARN:

group = iam.Group.from_group_arn(self, "MyImportedGroupByArn", "arn:aws:iam::account-id:group/group-name")

To import an existing group by name with path:

group = iam.Group.from_group_name(self, "MyImportedGroupByName", "group-name")

To add a user to a group (both for a new and imported user/group):

user = iam.User(self, "MyUser") # or User.fromUserName(this, 'User', 'johnsmith');
group = iam.Group(self, "MyGroup") # or Group.fromGroupArn(this, 'Group', 'arn:aws:iam::account-id:group/group-name');

user.add_to_group(group)
# or
group.add_user(user)

Instance Profiles

An IAM instance profile is a container for an IAM role that you can use to pass role information to an EC2 instance when the instance starts. By default, an instance profile must be created with a role:

role = iam.Role(self, "Role",
    assumed_by=iam.ServicePrincipal("ec2.amazonaws.com")
)

instance_profile = iam.InstanceProfile(self, "InstanceProfile",
    role=role
)

An instance profile can also optionally be created with an instance profile name and/or a path to the instance profile:

role = iam.Role(self, "Role",
    assumed_by=iam.ServicePrincipal("ec2.amazonaws.com")
)

instance_profile = iam.InstanceProfile(self, "InstanceProfile",
    role=role,
    instance_profile_name="MyInstanceProfile",
    path="/sample/path/"
)

To import an existing instance profile by name:

instance_profile = iam.InstanceProfile.from_instance_profile_name(self, "ImportedInstanceProfile", "MyInstanceProfile")

To import an existing instance profile by ARN:

instance_profile = iam.InstanceProfile.from_instance_profile_arn(self, "ImportedInstanceProfile", "arn:aws:iam::account-id:instance-profile/MyInstanceProfile")

To import an existing instance profile with an associated role:

role = iam.Role(self, "Role",
    assumed_by=iam.ServicePrincipal("ec2.amazonaws.com")
)

instance_profile = iam.InstanceProfile.from_instance_profile_attributes(self, "ImportedInstanceProfile",
    instance_profile_arn="arn:aws:iam::account-id:instance-profile/MyInstanceProfile",
    role=role
)

Features

  • Policy name uniqueness is enforced. If two policies by the same name are attached to the same principal, the attachment will fail.

  • Policy names are not required - the CDK logical ID will be used and ensured to be unique.

  • Policies are validated during synthesis to ensure that they have actions, and that policies attached to IAM principals specify relevant resources, while policies attached to resources specify which IAM principals they apply to.