GHSA-52gf-6rpq-fgmx: High severity go/github.com/openchoreo/openchoreo vulnerability
Summary
The OpenChoreo API server (openchoreo-api) authorized requests to its exec and wirelogs endpoints against the project supplied by the caller in the request, rather than against the project that actually owns the target component. The target component was resolved by name only, and the authorization engine decided purely from the caller-supplied resource hierarchy — the component's real owning project was never checked. As a result, an authenticated user who holds a project-scoped exec or log-viewing grant on any single project could run commands in, and read the wirelogs of, components owned by other projects in the same namespace.
Impact
An authenticated tenant holding a project-scoped component:exec grant on just one project can execute arbitrary commands — including an interactive shell — inside the running pods of components owned by any other project in the same namespace. This exposes those workloads' environment variables and mounted Secrets (such as database credentials, API keys, and tokens), allows tampering with the workload, and enables further pivoting using the workload's own network position and credentials. The same flaw in the wirelogs endpoint lets a project-scoped wirelogs:view grant read any component's wirelogs in the namespace.
The blast radius is every component in the namespace, reachable from a single legitimate project grant. The boundary crossed is project-within-namespace: the request still pins the namespace, so the issue does not cross namespace boundaries. In effect this is a cross-project (cross-team) remote command-execution and data-access primitive available to any low-privileged tenant.
Patches
Fixed in 1.2.3 and 1.1.6. The fix aligns the exec and wirelogs endpoints with the pattern already used by other first-class component operations: the target component is fetched first and authorization is evaluated against its real owning project (comp.Spec.Owner.ProjectName) instead of the caller-supplied project query parameter.
Workarounds
If you cannot upgrade immediately:
- Restrict component:exec and wirelogs:view grants to trusted operators only. Any holder of such a grant on any project can reach every component in the same namespace. - Where stronger isolation is required, separate sensitive components into their own namespaces, since the flaw does not cross namespace boundaries.
Related advisories
- GHSA-rh53-xvx2-j327 — the cluster-gateway internal proxy applies no caller authorization, so this API-layer authorization bypass reaches the data-plane Kubernetes API with no compensating control: https://github.com/openchoreo/openchoreo/security/advisories/GHSA-rh53-xvx2-j327 - GHSA-c5f6-2rm9-2w8g — reported together as part of the same review.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
go/github.com/openchoreo/openchoreoto a version that resolves this vulnerability.Fixed in 1.1.6 - Upgrade
Upgrade
go/github.com/openchoreo/openchoreoto a version that resolves this vulnerability.Fixed in 1.2.3 - Upgrade
Upgrade
openchoreoto a version that resolves this vulnerability.Fixed in 1.2.3 - Upgrade
Upgrade
openchoreoto a version that resolves this vulnerability.Fixed in 1.1.6 - Upgrade
Upgrade to a fixed release to a version that resolves this vulnerability.
Patch GHSA-c5f6-2rm9-2w8g - Upgrade
Upgrade to a fixed release to a version that resolves this vulnerability.
Patch GHSA-rh53-xvx2-j327 - Compensating control
Restrict `component:exec` and `wirelogs:view` grants to trusted operators only.
- Compensating control
Where stronger isolation is required, separate sensitive components into their own Kubernetes namespaces, since the flaw does not cross namespace boundaries.
Event History
Frequently Asked Questions
What is the scope of potential cross-project access?
The affected authorization behavior allows access to components owned by other projects when they are in the same namespace. The caller only needs the relevant project-scoped permission on one project in that namespace, rather than permission on the target component's actual project.
What permissions could be abused?
A project-scoped component:exec grant can be used to run arbitrary commands, including an interactive shell, in pods belonging to components in other projects. A project-scoped log-viewing grant can similarly expose wirelogs for components outside the caller's project.
What could an attacker obtain or do after gaining exec access?
They could read workload environment variables and mounted Secrets, including database credentials, API keys, and tokens. They could also modify the workload and pivot using its network position and credentials.