Attack vectors for software supply chains are the various ways in which someone can intentionally or accidentally compromise your software.
Risks of vulnerable software include leaking credentials or confidential data, corruption of data, installation of malware, and application outages. These problems result in lost time, money, and customer trust.
The entry points for threats span the entire software lifecycle and can originate inside or outside of your organization.
The diagram legend includes two sets of threats:
- Letters A to H indicate attack vectors in the software supply chain that are described as threats threats in the Supply chain Levels for Software Artifacts (SLSA) framework.
- Numbers 1 to 4 indicate additional attack vectors that the SLSA framework does not directly describe.
Software Delivery Shield, a fully-managed software supply chain security solution on Google Cloud, incorporates best practices to help you mitigate both sets of threats.
The subsections in this document describe the threats in the context of source, builds, deployment, and dependencies.
These threats impact the integrity of your source code.
1: Write insecure code. A lack of secure coding practices can lead to writing code unintentionally includes vulnerabilities. Insecure developer workstations can also introduce malicious or insecure code. Mitigations include:
- Setting policies for developer workstations. Cloud Workstations provides fully-managed, preconfigured workstations that you can customize to align with your requirements.
- Local scanning of code. Cloud Code source protect (private preview) provides real-time security feedback, including vulnerability and license information for dependencies. Developers can also use the On-Demand Scanning API to scan container images for OS and language package vulnerabilities.
- Education about practices to make code more secure.
A: Submit bad code to the source repository. This not only includes malicious code, but also code that unintentionally introduces vulnerabilities to an attack such as cross-site scripting. Mitigations include:
- Requiring human review for changes to source code.
- Using code scanning and linting tools that integrate with IDEs and source control systems.
B: Compromise the source control system. Restricting access to the source control system and other systems in your build pipeline, and using multi-factor authentication helps to mitigate this risk.
When evaluating your source integrity, also examine supporting scripts and configuration that you use to build, and deploy your software. Include them in your source control system and code review processes, so that you can risk from vulnerabilities in these files.
See Safeguard source to learn more about protecting your source.
These threats compromise your software when you build or package it, or trick consumers of your software into using a bad version.
- C: Build with source that is not from the trusted source control system.
Mitigations that help to reduce this risk include:
- Using build services, such as Cloud Build, that generate provenance information so that you can validate that your builds use trusted source.
- Placing your CI/CD infrastructure in a network perimeter to prevent exfiltration of data from your builds. For Google Cloud services, use VPC Service Controls.
- Storing and using trusted copies of open source dependencies you need in a private artifact store such as Artifact Registry.
- D: Compromise the build system. Mitigations that help to reduce this risk
- Follow the principle of least privilege by restricting direct access to the build system to individuals who require it. In Google Cloud you can grant appropriate predefined roles or create custom roles.
- Use managed build services such Cloud Build. Cloud Build runs ephemeral builds builds by setting up a VM environment for each build and destroying it after the build.
- Place your CI/CD infrastructure in a network perimeter to prevent. exfiltration of data from your builds. For Google Cloud services, use VPC Service Controls.
- F: Package and publish software that was built outside of the official process. Build systems that generate and sign build provenance enable you to validate that your software was built by a trusted build system.
- G: Compromise the repository where you store your software for your
internal or external users. Mitigations that help to reduce this risk
- Storing and using trusted copies of open source dependencies you need in a private artifact stores such as Artifact Registry.
- Validating build and source provenance.
- Restricting upload permissions to dedicated non-human accounts and repository administrators. On Google Cloud, service accounts act on behalf of services and applications.
Deployment and runtime threats
H: Resolving dependencies by specifying a version range or a tag that isn't permanently attached to a specific build version can lead to several issues:
- Builds are not reproducible because dependencies that a build uses the first time can be different from dependencies that the build uses for future executions of the same build.
- A dependency might resolve to a compromised version, or a version with changes that break your software. Bad actors can take advantage of this uncertainty to cause your build to choose their version of a package instead of the version you intended to use. A number of best practices for dependencies can help to mitigate dependency confusion risks.
2: Compromise the deployment process. If you use a continuous deployment process, compromising that process can introduce unwanted changes to the software that you deliver to your users. You can mitigate risk by restricting access to your deployment service and testing changes in pre-production environments. Google Cloud Deploy can help you manage the continuous delivery process and promotion between between environments.
3: Deploy compromised or noncompliant software. Enforcing deployment policies can help mitigate this risk. You can use Binary Authorization to validate that container images are compliant with policy criteria and block deployment of container images from untrusted sources.
4: Vulnerabilities and misconfiguration in running software.
- New vulnerabilities are discovered regularly, which means that new findings can changes the security risk level for your applications in production.
- Some configurations increase risk of unauthorized access, such as running as the root user or allowing privilege escalation on execution of a a container.
The GKE security posture dashboard shows information about OS vulnerabilities and configuration issues in your running workloads.
In Cloud Run, you can also view security insights about your deployed revisions, including known vulnerabilities in container images you deployed.
See Safeguard builds to learn more about protecting your source, and Safeguard deployments to learn about protecting deployments.
Dependencies includes direct dependencies in your builds as well all transitive dependencies, the recursive tree of dependencies that are downstream from your direct dependencies.
In the diagram, E indicates using a bad dependency in your build. A bad dependency can include:
- Any software that your application depends on, including components you develop internally, commercial third-party software, open source software.
- Vulnerabilities originating from any of the other attack vectors. For example
- An attacker gains access to your source control system and modifies the version of a dependency that you project uses.
- Your build includes a component developed by another team in your organization. They publish build and publish the component directly from their local development environments and accidentally introduce a vulnerability in a library that they only use locally for testing and debugging.
- Intentional removal of an open source dependency from a public repository. The removal can cause consuming pipelines to break if they retrieve the dependency directly from the public repository.
See best practices for dependencies to learn about ways to to mitigate risks.
The overall integrity of your supply chain is only as strong as its most vulnerable part. Neglecting an attack vector increases risk of attack in that part of your supply chain.
At the same time, you don't need to change everything at once. The cumulative act effect, more popularly known as the swiss cheese model, applies to software supply chain security. Each mitigation that you implement reduces your risk, and when you combine mitigations across your supply chain, you increase protection against different types of attacks.
- Assess your security posture using frameworks and tools that help you evaluate your organization's ability to detect, respond to, and remediate threats.
- Learn about best practices to protect your software supply chain, and Google Cloud products designed to support those practices.
- Use Software Delivery Shield to protect your software across your software supply chain on Google Cloud. You can implement services gradually, based on your priorities and existing infrastructure.
- Assess your security posture.
- Learn about best practices to protect your software supply chain.
- Learn about Software Delivery Shield.