FastGPT GitHub Actions CI/CD Privileged Preview Workflow Artifact Escalation

Vulnerability Overview Title: Untrusted PR artifacts are pushed and deployed by privileged preview workflows Description: A potential GitHub Actions CI/CD security issue has been reported affecting the repository. The issue stems from untrusted PR build artifacts being promoted within trusted jobs, which possess registry permissions, PR/issue write access, signing capabilities, and deployment authority. Scope of Impact Affected Repository: Affected Branch and Commit: Affected Workflow Pairs: 1. Documentation Preview Image/Deployment Pipeline - - 2. FastGPT Preview Image Pipeline - - Remediation Steps 1. Do not promote untrusted PR build artifacts: Avoid promoting untrusted PR build artifacts within trusted jobs. 2. Enforce trusted sources: Before any push/deploy operation, require a trusted source. For example, mandate: - PRs from the same repository - Trusted actor associations - Labels approved by maintainers - Protected environment approvals - Or manual workflow dispatch by maintainers 3. Rebuild from trusted base branches: Rebuild deployable images from trusted base branches rather than loading artifacts produced by PRs. 4. Isolate preview images for external PRs: If preview images for external PRs are required, push them to an isolated, non-production registry namespace using low-privilege credentials. 5. Do not deploy untrusted PR build images: Do not deploy untrusted PR build images to Kubernetes clusters with significant permissions or network access. 6. Strictly bind artifacts: Strictly bind artifacts to the expected workflow run, repository, event type, commit SHA, and actor; reject artifacts from fork PR runs unless explicitly approved. 7. Verify artifact origin: Verify the origin of artifacts before executing , , , or . 8. Separate untrusted builds from trusted deployments: - Untrusted PR workflows: No secrets, no registry pushes, no deployment - Trusted deployment workflows: Only trusted inputs, protected environments, minimal credentials 9. Limit credential scope: Restrict and GHCR credentials to the minimum necessary permissions, and rotate them after untrusted preview artifacts have been promoted. POC Code Security Impact Untrusted PR code controls the Docker build context and image content. The generated image artifact is uploaded by the PR workflow. The trusted job downloads this artifact. The trusted workflow pushes the image to GHCR and deploys it using Kubernetes credentials in the docs preview. The pushed/deployed artifact is treated as a repository-controlled preview infrastructure output. Non-Destructive Maintainer Verification 1. Create a controlled fork PR that changes only a harmless tag within the Docker image build context. 2. Ensure the tag is visible only within the fixed string. 3. Confirm that the PR-requested build workflow constructs the image artifact containing the tag. 4. Confirm that the trusted workflow downloads the artifact from the PR run. 5. Confirm that the trusted workflow: - Loads the PR-built image - Tags it - Pushes it to GHCR - Deploys it via the Kubernetes preview deployment path (for docs preview) 6. Verify using temporary test credentials or by disabling the final push/deploy step. Expected Behavior Expected Vulnerable Behavior: - Image artifacts from fork PRs are promoted by the trusted . - The trusted workflow pushes or deploys the artifact. - No checks for trusted actors, maintainer approval, artifact origin verification, or source origin restrictions block the path. Expected Secure Behavior: - Untrusted fork PRs are not promoted by trusted workflows. - Push/deploy workflows run only for trusted actors or trusted branches. - Trusted workflows rebuild from trusted code or verify artifact origins before promotion. - Kubernetes and registry credentials are not used for artifacts produced by untrusted PR code. Suggested Remediation 1. Do not promote untrusted PR build artifacts. 2. Enforce trusted sources. 3. Rebuild from trusted base branches. 4. Isolate preview images for external PRs. 5. Do not deploy untrusted PR build images. 6. Strictly bind artifacts. 7. Verify artifact origin. 8. Separate untrusted builds from trusted deployments. 9. Limit credential scope. Additional Information Reporter: c121914yu Report Date: May 26, 2024 Vulnerability ID: GHSA-rvgc-2c29-g876 Severity: Medium CVE ID: CVE-2024-50562 Weaknesses: - CWE-266 - CWE-494 - CWE-829 Conclusion This vulnerability involves untrusted PR build artifacts being promoted to trusted jobs, which could lead to the deployment of malicious code. Remediation includes not promoting untrusted artifacts, enforcing trusted sources, rebuilding from trusted base branches, isolating preview images for external PRs, avoiding deployment of untrusted PR build images, strictly binding artifacts, verifying artifact origins, separating untrusted builds from trusted deployments, and limiting credential scope.

Goal Reached Thanks to every supporter — we hit 100%!

FastGPT GitHub Actions CI/CD Privileged Preview Workflow Artifact Escalation

More from this source