A MITRE-based Analysis of a Cloud Attack

When you look at the details surrounding attacks in the cloud, how easy is it for you to quickly determine what happened and how to prevent a similar attack in your own environment? Applying the MITRE Att&ck Cloud Matrix is a great resource to classify and understand cloud-based attacks.

In this post, we will revisit a previous blog post to demonstrate how MITRE’s new Cloud Matrix can be applied to help us analyze the threat and prevent it from happening in your environment. We’ll recount the scenario, which exposes data in a storage bucket hosted by the Google Cloud Platform (GCP). Then we will examine some mitigations provided by MITRE, and look at how we could implement these in GCP.

Summary of the Scenario

The original scenario was an attack in which I escalated my privileges to remove a security perimeter established around a storage bucket that was storing sensitive data. The security perimeter only allowed people to access the bucket if they were coming from the correct IP address range. Below is a diagram, along with the steps taken to complete the attack.

The attack followed these basic steps:

1. A service account credential was accidentally exposed, which had OSLogin permissions (allows someone to SSH to virtual machines).
2. Using the exposed credential, I SSHd into a publicly exposed virtual machine.

The virtual machine was running with the default service account for Google Compute Engine (which has too many permissions), and the scope associated was, “Allow full access to all Cloud APIs.”

1. Using the default service account, I listed the other service accounts in that project.
2. I located a service account that had admin permissions and activated that service account from the virtual machine. At that point, I had administrative privileges at the Organization level.
3. Using the admin service account, I removed a VPC Security Perimeter.
4. After removing the VPC Security Control Perimeter, I accessed the data in the storage bucket that was previously protected.

The attack’s impact was the exposure of sensitive data. 

Breaking Down the Att&ck

Below are the tactics, techniques, and mitigations found in MITRE’s Cloud Matrix matched with parts of the attack. The numbers match with the attack steps in the section above. The text in italics under the MITRE mitigation column was taken directly from MITRE’s Cloud Matrix.

Along with MITRE’s Techniques and Mitigations, I’ve added some specific mitigative steps to take for GCP.

Step 1

Description: A valid credential was found in source control that was publicly exposed.

• Tactics: Initial Access
• Techniques: Valid Accounts
• Mitigations: Application Developer Guidance

Scan your source code for credentials, and then update any credentials found. This is not specific to GCP. However, the GCP service account credential is an RSA key, so you will need to be sure that your scans include private RSA keys.

Step 2

Description: A SSH connection was established from outside the organization to a virtual machine hosted by GCP using the exposed credential.

• Tactics: Initial Access
• Techniques: Exploit Public-Facing Application
• Mitigations: Network Segmentation

Only allow SSH authentication from allow listed IP addresses. This can be configured in GCP firewall rules.

Step 3 + 4

Description: The credentials of the publicly exposed virtual machine were used to assume another identity that has administrative privileges in GCP.

• Tactics: Privilege Escalation
• Techniques: Valid Accounts
• Mitigations: Privileged Account Management

The publicly exposed workload had a service account with no scope limitations, so its full capabilities would be allowed. In this case, the full capabilities meant that the virtual machine’s identity could be used to enumerate and impersonate other service accounts in the same GCP project.

To prevent this, you would need to scan the compute instances in your environment and make sure the “scopes” are not set to ‘Allow All’ while using the default service account. More information about service accounts and scopes for the compute engine is available here.

In addition, an administrative service account with an Organization-level binding had been created in the same GCP project as a publicly exposed workload. To prevent this, there are two main steps you can take:

1. Do not give service accounts administrative privileges. In addition, do not bind them at the Organization level.
2. If you must give a service account some administrative privilege or a binding at the Organization level, create that service account in a private project. Any publicly exposed workloads should be in projects separated from administrative service accounts.

Step 5

Description: A cloud perimeter control was removed, which protected sensitive data in part of the GCP environment.

• Tactics: Defense Evasion, Impact
• Techniques (proposed): Remove Boundary Security Control
• Mitigations: Filter Network Traffic

GCP does not currently allow using conditions in Identity and Access Management (IAM) policies, but it is available in beta, and may change. Much like AWS, the conditions may allow us to set permissions that will only work from certain IP addresses.  More information about it is available here.

Step 6

Description: Sensitive data was accessed from an IP address outside of the organization.

• Tactics: Collection
• Techniques: Data from Cloud Storage Object
• Mitigations: Filter Network Traffic

We did have a VPC service control perimeter, which disallowed access from outside IP addresses. However, the attack removed that control. The preventative steps in for the other techniques should prevent this from happening. However, we should also be alerted to the modification or removal of a VPC service control perimeter. In order to generate that type of alert, you must monitor the audit logs. More information on audit log monitoring is available here.

Conclusion

Without using the MITRE Att&ck Cloud Matrix, a common approach to this incident would be to make sure that no credentials are leaked like this in the future. However, as you can see from all of the content above, there are a number of other mitigations that should be considered as well.

Due to the abstract nature of the cloud, and the immense amount of services and configuration settings that could be deployed, it’s difficult to understand all the factors involved in an attack. The exercise of applying MITRE’s Att&ck Cloud Matrix, discovering possible mitigations in the matrix, and then mapping those mitigations to your specific environment should be something employed to maintain a strong security posture. This exercise will also help you understand and evaluate security tool features that are important for your own use cases as well.