Child pages
  • Topic - Security Controls
Skip to end of metadata
Go to start of metadata


Infrastructure security (whether in the cloud or not) is tightly bound to accounts, roles and identities.

There are multiple questions to consider when deciding on a cloud account strategy.  Some examples include:

  • Billing - How will resource utilization and consumption be measured and reported. Will there need to be internal chargebacks?

  • Business models -    Will central IT act as an institutional service provider for cloud infrastructure, or will there be multiple accounts across distributed IT and research groups? How will  financial responsibility be distributed across different uses of cloud infrastructure?

  • Governance -   How will resources be allocated, secured, controlled and audited?

  • Operational -   How will operational management, enterprise integration, network integration and AWS service limits be architected and managed?


The following should be considered as some best practices for account strategy and be adapted to meet the organizational needs:

  1. Develop and enforce common tagging practices which meet minimum requirements for billing and chargeback
  2. Implement a consolidated single master account which provides billing consolidation and reporting across all institutional units (sub-accounts)
  3. Define a sub-account creation policy based on specific requirements for isolation or delegation as per governance and security requirements; initially it is recommended to segregate based on operating environment (i.e., production vs. non-production)
  4. Consider additional VPCs as boundaries for workloads that require specialized controls
  5. Leverage a common services model using VPC peering to minimize duplication of resources across accounts

AWS Provisioning and IAM Roles at Penn State

In AWS an account owner provisions users who will have access to the AWS console. Using AWS Identity and Access Management (IAM) roles with specific levels of permissions can be assigned to users by the account owner.

When provisioning an AWS account, a root account is created. This account has access to do the highest level of privileged tasks in the AWS console. It can delete the account, it can create IAM resources and link other AWS accounts for billing purposes. If used improperly, it can delete its own data centers. It is not advisable to use this account unless absolutely necessary. However, AWS offers several ways to integrate Identity and Access Management capabilities that allow console users to operate with more granular privileges. In fact, many different SAML integrations are supported that allow for a multi-user console environment where users operate with specific roles (

An IAM role is similar to a user, it is associated with a permission policy that describes what is can and can not do in the AWS account. Roles can be a one to one or one to many association.  It is easy to make simple rules that allow a role to do simple tasks like launch and terminate EC2 instances or read from an S3 bucket. But, it is also possible to make complex roles that limit a role to operate with very specific privileges in a certain part of the environment. For example, a role could be created for a developer team that would only allow these users to operate in a specific dev-test "data center" (Virtual Private Cloud or VPC). Or, a role could be created that would only be able to spin up certain size instances to ensure that spending doesn't get out of control. An infrastructure role could be created such that only this group could be able to perform typical infrastructure tasks like modifying subnets, VPCs or load balancers configuration.

Integrating AWS IAM roles with campus IdM is possible in many different scenarios. A common scenario might be the integration of a campus Shibboleth IDP with AWS. Each AWS account needs to have the IDP metadata configured in it. This can be accomplished manually through the console or via the AWS API.  The AWS account will require the roles to be created in IAM with the level of privilege needed. This can be done manually through the console or via API. More information on using AWS IAM roles with Shibboleth can be found in this blog post:

In the simplest scenario, an IDP could be configured to pull group memberships for a user that is logging into AWS from an LDAP directory. One of the easiest ways to accomplish this is to develop a naming scheme for IAM roles that contain both Amazon Resource Names (the 12 digit number identifying the account)  and a role.

The IDP could be configured to look for LDAP groups with a certain regular expression string and parse that information into the attributes that AWS requires (awsRoles and awsRoleSessionName). For example, inserting a person into a group named could inform AWS that the user is logging into account 123456789012 and that the user should assume the IAM role with the name "read-only" in that account. If the user is in multiple groups and multiple attributes are asserted, the user has the choice of what account/role they wish to assume (see figure 1-1).

Grouper ( is a popular  enterprise access management system that is highly customizable and supports group management. It supports the use of naming stems, a string that forms the leading part of a Group's name. In this way, the possibility that different groups can be given the same name is substantially reduced, and the name of each group can be made to reflect something about the authority under which it was created. The permission on the stem can be configured such that only a particular application or process could modify it.

Example attribute resolver stanzas for group naming convention in the form of ""

<resolver:AttributeDefinition id="awsRoles" xsi:type="ad:Mapped" sourceAttributeID="MemberOf">

 <resolver:Dependency ref="ldap"/>



 name="" friendlyName="Role" />






<resolver:AttributeDefinition id="awsRoleSessionName" xsi:type="ad:Simple" sourceAttributeID="mail">

 <resolver:Dependency ref="ldap"/>




  friendlyName="RoleSessionName" />


A caveat with the use of roles is that roles do not support MFA. However, the institution could enforce MFA at the IDP or SSO layer. One of the obvious advantages of using IAM roles over IAM users is that there are no credentials on the AWS side to manage. When a user is taken out of the group or they leave the institution, they lose access to the AWS environment.

Here is a high level potential sequence of provisioning an account.


  1. Account is requested from user at some web page (m)

  2. Account is provisioned in DLT portal (in the case of institution having Net+ AWS contract) and root credentials are retrieved (m)

  3. Root credentials are escrowed in a secure location, user does not have access to these credentials (m)

  4. Credentials are used to turn logging (Cloudwatch) on in account (a)

  5. Credentials are used to provision standard IAM roles and IDP information into account (a)

  6. Account information (account number) is stored in a directory (grouper, LDAP, or AD for example) and groups created that have IAM roles to perform certain duties (a) e.g. owner, read-only, billing

  7. Campus IPAM service is called requesting private IP space and Virtual Private Cloud network space is provisioned in cloud service (a)

  8. MFA is turned on the root account (a)

  9. User is given login URL (in the case of AWS, it is an IDP initiated SSO handler in the form of (a)

Support request:

  1. Request is logged into ITSM tool (m)

  2. Staff member with appropriate skill level is placed into admin group for account (a)

  3. Staff member logs in and identifies/fixes account issue (m)

  4. Staff member closes incident (m)

  5. Staff member is removed from admin group for account (a)

(m) denotes manual process

(a) denotes intended to be automated process


Securing Workloads of Differing Sensitivities in AWS at UNC Chapel Hill -

We are just getting started but we are looking at Landing Zone and Control Tower ( as one way to implement auditing and compliance for workloads in AWS for our institution. Specifically, we are looking for a way to manage accounts, monitor the accounts, and set baseline policies for the accounts that are created for researchers or departments. We are also looking into Transit Gateway ( to manage network connections within AWS and back to campus. I expect many of you are already using Organizations ( along with service control and/or AWS Config.


Just as the practice of syslogging events to a separate host is standard in on-premise environment, logging in cloud environments is also a good practice.  In AWS, a technology called CloudTrail enables the ability to log any change to the infrastructure. Coupled with IAM policies, this functionality can ensure that a compromise of the environment can not be hidden. 


  • No labels