Get more out of service control policies in a multi-account environment

Many of our customers use AWS Organizations to manage multiple Amazon Web Services (AWS) accounts. There are many benefits to using multiple accounts in your organization, such as grouping workloads with a common business purpose, complying with regulatory frameworks, and establishing strong isolation barriers between applications based on ownership. Customers are even using distinct accounts for development, testing, and production. As these accounts proliferate, customers need a way to centrally set guardrails and controls.

In this blog post, we will walk you through different techniques that you can use to get more out of AWS Organizations service control policies (SCPs) in a multi-account environment. We focus on policy evaluation logic and how SCPs fit into it, show an overview of SCP inheritance, and describe methods for writing compact SCPs. We cover the following five techniques:

1. Consider the number of policies per entity
2. Use policy inheritance
3. Segment by workload type
4. Combine policies together
5. Compact your policies

AWS Organizations provides a mechanism to set distinct logical boundaries by using organizational units (OUs). This is useful when you have similar workloads across different AWS accounts that require common guardrails. SCPs are a type of organization policy that you can use to manage permissions in your organization. SCPs offer central control over the maximum available permissions for all accounts in your organization. SCPs help you make sure that your accounts stay within your organization’s access control guidelines. A key distinction of SCPs is that they are useful to set broad guardrails across your environment. You can think of guardrails as a way to enforce specific governance policies at varying levels of your environment, which we will discuss in this post.

Policy evaluation logic and how SCPs fit in

Before we dig into the details, let’s first look at how SCPs work from an overall policy perspective, along with the evaluation logic. An explicit Deny statement in any policy trumps an Allow statement. Organization SCPs that apply to any AWS account that is part of an organization in AWS Organizations require an Allow statement before proceeding in the policy evaluation flow.

For an in-depth look at how policies are evaluated, see Policy evaluation logic in the documentation.

Now, let’s walk through five recommended techniques that can help you get more out of SCPs.

1. Consider the number of policies per entity

An organization is a collection of AWS accounts that you manage together. You can use OUs to group accounts within an organization and administer them as a single unit. This greatly simplifies the management of your accounts. It’s possible to create multiple OUs within a single organization, and you can create OUs within other OUs, otherwise known as nested OUs. You have the flexibility to attach multiple policies to the root of the organization, to an OU, or to an account. For example, in an organization that has the root, one OU, and one account, attaching five SCPs to each of them would produce a total of 15 SCPs (five SCPs at the root, five SCPs at the OU, and five SCPs on the one account).

The number of SCPs that you can apply is limited, and being close to or at the quota could restrict your ability to add more policies in the future. The current published quotas are as follows:

• Maximum number of SCPs attached to the root: 5
• Maximum number of SCPs attached to each OU: 5
• OU maximum nesting in a root: 5 levels of OUs under a root
• Maximum number of SCPs attached to each account: 5

Note: For the latest information on quotas, see Quotas for AWS Organizations.

Consider the following sample organization structure to understand how you can apply multiple SCPs at different levels in an organization.

Figure 1: A sample organization showing the maximum number of SCPs applicable at each level (root, OU, account)

2. Use policy inheritance

Policy inheritance refers to the inheritance of policies that are attached to the organization’s root or to an OU. All accounts that are members of the organization root or OU where a policy is attached are affected by that policy, but inheritance works differently for Allow and Deny statements. For a permission to be allowed for a specified account, every SCP from the root through each OU in the direct path to the account, and even attached to the account itself, must allow that permission. In other words, a statement that allows access needs to exist at every level of a hierarchy; it’s not inherited. However, a Deny statement is inherited and evaluated at each level.

At this point, you should start thinking about the policies from a broader controls perspective: Controls that you want to implement on the whole organization should go into your organization’s root-level SCP. Controls should be more granular as you move down the hierarchy in AWS Organizations.

For example, when a Deny policy is attached to the organization’s root, all accounts in the organization are affected by that policy. When you attach a Deny policy to a specific OU, accounts that are directly under that OU or nested OUs under it are affected by that policy. Because you can attach policies to multiple levels in the organization, accounts might have multiple applicable policy documents, as shown in Figure 2.

Figure 2: Sample organization showing applicable policies

By default, AWS Organizations attaches an AWS managed SCP named FullAWSAccess to every root and OU when it’s created. This policy allows all services and actions.

Note: Adding an SCP with full AWS access doesn’t give all the principals in an account access to everything. SCPs don’t grant permissions; they are used to filter permissions. Principals still need a policy within the account that grants them access.

Additionally, the policies that are applied to an OU only affect the accounts or the child OUs under it and don’t affect other OUs created under the root. For example, a policy applied to the Sandbox OU doesn’t affect the Workloads OU.

The two tables that follow show examples of the policies that result from inheritance. As discussed previously, if an Allow isn’t present at all levels (root, OU, and account) the account won’t have access to any service. Consider the last example in the Sandbox OU table with a “Deny S3 access” SCP at the root, which limits access to Amazon Simple Storage Service (Amazon S3). Although there is “Allow S3 access” applied to the Sandbox OU and “Full AWS access” at the account level, the resultant policy on account A is “No service access” because there is no policy with an effect of “Allow” in the SCP at the root level.

The following table shows the inheritance of policies in the Sandbox OU.

The following table shows the inheritance of policies in the Workloads OU.

Some examples of common root-level policies are as follows:

• Prevent member accounts from leaving the organization (Leave, Delete, RemoveAccountFromOrganization).
• Prevent tampering with AWS CloudTrail (DeleteTrail, StopLogging, UpdateTrail, PutEventSelectors).
• Prevent users from disabling AWS Config (StopConfigurationRecorder, DeleteDeliveryChannel)
• Prevent sharing resources outside of your organization. This is useful to restrict sharing outside of your organization.
• Restrict which AWS Regions to use.

For sample SCPs, see Example service control policies. For insight into best practices for applying policies at different levels in an organization, see Best practices for SCPs in a multi-account environment.

3. Segment SCPs by workload type

A key feature of AWS Organizations is the ability to create distinct workload boundaries by using organizational units (OUs). You can think of OUs as a logical boundary where you can directly apply SCPs. You can also nest OUs up to five levels deep and apply different policies at each level. By using OUs, you can segment your workload types and create purpose-driven guardrails to match your security and compliance requirements.

To illustrate this, let’s take an example where there are three distinct workload types divided into three separate OUs: Infrastructure, Sandbox, and Workload, as shown in Figure 3. A best practice would be to tailor your SCPs to each specific OU type. Your security organization wouldn’t want to allow private workloads to be reachable from the internet. However, workloads that serve your external customers would require external network connectivity. To support innovation and experimentation, you can establish a Sandbox OU that has fewer policy restrictions but might limit connectivity back to your corporate data center.

For additional information on how to organize your OUs, see Recommended OUs.

Figure 3: Example organization showing different workloads

4. Combine policies together

Similar to AWS Identity and Access Management (IAM) policies, you can have multiple statements within a service control policy. You can combine statements in a single policy to avoid hitting the quota limit of five policies per account, OU, or root. An AWS full access policy is attached by default when you enable SCPs on an organization. You can combine the full access policy with additional controls and combine statements, as shown in the following example policy. Each SCP that you apply can have a policy size of 5,120 bytes. When combining statements, make sure that the resultant statement doesn’t alter your original intent. You can combine the Action elements in an SCP if the policy has the same values for Effect, Resource, and Condition.

AWS full access policy (143 bytes)

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": "*",
            "Resource": "*"
        }
    ]
}

You can combine this full access policy with the following deny policy:

Deny bucket deletion and Security Hub disablement (260 bytes)

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Deny",
            "Action": "s3:DeleteBucket",
            "Resource": "*"
        },
        {
            "Effect": "Deny",
            "Action": "securityhub:Disable*",
            "Resource": "*"
        }
    ]
}

The resulting combined policy is as follows:

Combined policy (274 bytes)

{
   "Version":"2012-10-17",
   "Statement":[
      {
         "Effect":"Allow",
         "Action":"*",
         "Resource":"*"
      },
      {
         "Effect":"Deny",
         "Action":[
            "s3:DeleteBucket",
            "securityhub:Disable*"
         ],
         "Resource":"*"
      }
   ]
}

5. Compact your policies

One difference between IAM policies and SCPs is that whitespace counts against the size quota in SCPs. Compacting related actions in a policy can help you shorten the policy. Following are four methods to compact your policy:

1. Remove whitespace. If you use the AWS Management Console, whitespace is automatically removed. However, if you don’t want to manually update policies by using the console every time, you can incorporate a script that removes the whitespace. (Method four later in this list provides an example of this type of script.)
2. Use wildcards and prefixes to combine multiple actions. For example, the following policy denies access to disable configuration in AWS Security Hub.

   {
        "Effect": "Deny",
        "Action":[
           "Securityhub:DisableSecurityHub", 
           "Securityhub:DisableOrganizationAdminAccount",
           "Securityhub:DisableImportFindingsForProduct"
        ],
        "Resource": "*"
       }

   By using wildcards and prefixes, you can rewrite this policy as follows:

     {
       "Effect": "Deny",
       "Action": "Securityhub:Disable*",
       "Resource": "*"
   }

   Important: When you combine actions together as in this example, be aware that there could be a potential impact if new actions are released in the future that start with the Disable keyword, because these actions will be covered by the wildcard and denied.

3. SCPs can be configured to work as either deny lists or allow lists. For additional details on allow lists and deny lists, see Strategies for using SCPs. We recommend that you use deny lists where possible, because they are more flexible and can help simplify your policies, which will result in less maintenance. To expand on this strategy, deny statements support conditions (as shown in the following example), and for specific resources to be specified. For example, when AWS adds a new service, you don’t have to go back and update your policy if you’ve used a deny statement. To support this, AWS Organizations attaches an AWS managed SCP named FullAWSAccess to every root and OU when it’s created. This policy allows all services and actions. Additionally, deny statements coupled with NotAction statements can help you write shorter policies.

   Consider the following scenario: Your security organization requires that application teams use specific AWS Regions. The recommended approach is to create a deny list that blocks everything except what is in the NotAction block. Following is an example where the SCP denies any operation outside of specified Regions that your organization has authorized for use.

   Note: The list includes AWS global services that cannot be allowlisted based on a Region.

   {
       "Version": "2012-10-17",
       "Statement": [
           {
               "Sid": "DenyAllOutsideEU",
               "Effect": "Deny",
               "NotAction": [
                   "a4b:*",
                   "acm:*",
                   "aws-marketplace-management:*",
                   "aws-marketplace:*",
                   "aws-portal:*",
                   "budgets:*",
                   "ce:*",
                   "chime:*",
                   "cloudfront:*",
                   "config:*",
                   "cur:*",
                   "directconnect:*",
                   "ec2:DescribeRegions",
                   "ec2:DescribeTransitGateways",
                   "ec2:DescribeVpnGateways",
                   "fms:*",
                   "globalaccelerator:*",
                   "health:*",
                   "iam:*",
                   "importexport:*",
                   "kms:*",
                   "mobileanalytics:*",
                   "networkmanager:*",
                   "organizations:*",
                   "pricing:*",
                   "route53:*",
                   "route53domains:*",
                   "s3:GetAccountPublic*",
                   "s3:ListAllMyBuckets",
                   "s3:PutAccountPublic*",
                   "shield:*",
                   "sts:*",
                   "support:*",
                   "trustedadvisor:*",
                   "waf-regional:*",
                   "waf:*",
                   "wafv2:*",
                   "wellarchitected:*"
               ],
               "Resource": "*",
               "Condition": {
                   "StringNotEquals": {
                       "aws:RequestedRegion": [
                           "eu-central-1",
                           "eu-west-1"
                       ]
                   }
               }
           }
       ]
   }

4. Shorten the Sid value in your policy: The Sid (statement ID) is an optional identifier that you provide for the policy statement. Remove it completely from your policy if it serves no purpose for you. We also have customers who find it effective to maintain a list of SID values and details on corresponding policies in an index file locally.

The following sample Python code can compress a provided policy by removing whitespace and Sid values.

You can export the compressed policy in the file named Compressed_Policy.json or show the output on the terminal by removing # from the following code.

import json
def compress_json(policy):
    statement = policy["Statement"]
    if not isinstance(statement, list):
        statement = [statement]
    for s in statement:
        s.pop("Sid", None)
   
    # json.dumps removes whitespace around separators in a JSON and converts it to a JSON formatted string.
    # To get the most compact representation, specify separators=(item_separator, key_separator)
    policy_without_whitespace = json.dumps(policy, separators=(',', ':'))
   
    return policy_without_whitespace

if __name__ == '__main__':
  path = input("Enter the path to policy file like: \n  /Users/swara/Desktop/policy.json or ./policy.json  \n >  ")
  with open(path) as f:
    policy = json.load(f)
   
original_len = len(str(policy))
mini_policy = compress_json(policy)
#To print the output on the screen
print(mini_policy)
compressed_len = len(str(mini_policy))
print("\n \t original length: {} -> compressed length: {} \n".format(original_len, compressed_len))
#To write output to a file named Compressed_Policy.json
with open("Compressed_Policy.json", "w") as Output_file:
     print(mini_policy, file=Output_file)

Example output on screen:

{"Version":"2012-10-17","Statement":[{"Action":["iam:AttachRolePolicy","iam:DeleteRole","iam:DeleteRolePermissionsBoundary","iam:DeleteRolePolicy","iam:DetachRolePolicy","iam:PutRolePermissionsBoundary","iam:PutRolePolicy","iam:UpdateAssumeRolePolicy","iam:UpdateRole","iam:UpdateRoleDescription"],"Resource":["arn:aws:iam::*:role/role-to-deny"],"Effect":"Deny"}]}

original length: 433 -> compressed length: 364

To download the sample python code and the example policy shown above, download the files compress-policy.py and policy.json.

Conclusion

In this post, we walked you through different techniques that you can use to get more out of service control policies in a multi-account environment. By using these techniques, you can establish a well-considered strategy for how your organization can adopt SCPs in a multi-account environment. You also learned about how SCPs fit into the overall policy landscape for AWS. SCPs are a powerful tool to help customers establish guardrails. As you evaluate your IAM strategy, consider what you’re trying to achieve. If you’re trying to establish broad guardrails for multiple accounts, then we suggest looking at SCPs first.

 
If you have feedback about this post, submit comments in the Comments section below. If you have questions about this post, contact AWS Support.

Want more AWS Security news? Follow us on Twitter.