Database/AI/ML frameworks & serving
Eclipse Che: authenticated SSRF in devfile URL fetching leaks the pod service-account token
Impact
Any authenticated Che user can make the Che server fetch a URL of their choosing and return the body, with no scheme or host allow-list. Via file:// that reads arbitrary files inside the Che server pod, including /var/run/secrets/kubernetes.io/serviceaccount/token - on a GPU cluster that token is a foothold into the same Kubernetes API that schedules every tenant's accelerator workloads. The same path reaches internal-only HTTP services and the cloud instance metadata endpoint at 169.254.169.254, so node-level cloud credentials are in range too. Che also attaches the caller's stored SCM personal access token as an Authorization header to the fetched host, and the same forwarding fires when a victim opens a workspace from a malicious devfile whose parent.uri points at an attacker server, so a developer's Git credentials can be exfiltrated without the attacker having API access at all. Affects 7.29.0 and later; the record states no fix is available.
Who can reach it
Any user with a Che account can call the two REST endpoints directly (authentication required, lowest privilege level is enough). The credential-forwarding variant needs only that a victim Che user opens a workspace built from an attacker-supplied devfile.
What to do
No fix is published. Mitigate: keep the Che/Dev Spaces route off any network an untrusted tenant can reach, restrict egress from the Che server pod so file-less targets like 169.254.169.254 and internal services are unreachable, and scope the Che server's service account down to the minimum it needs so a stolen token buys little. Rotate any SCM personal access tokens stored in Che if you suspect use, and watch the Red Hat tracker (CRW-11956) and the Eclipse vulnerability report for a patched release. No node drain or reboot is involved - this is a platform-level service, not a kernel or firmware change.
References
Related entries
- Gradio: CORS origin validation bypassCVE-2024-47084 · GradioHigh
- vLLM: malformed multimodal embedding input crashes the engine and drops every in-flight requestCVE-2025-62372 · vLLM (multimodal embedding input validation)High
- LangChain: template injection in prompt templates exposes Python object internalsCVE-2025-65106 · LangChain ChatPromptTemplate (prompt template parsing)High
- vLLM: --revision pin ignored for some FunAudioChat and Tarsier2 processor, tokenizer and config loadsCVE-2026-100653 · vLLM (revision pin not propagated to FunAudioChat and Tarsier2 Hugging Face artifact loads)High
- vLLM: prompt_logprobs over a KV-connector-cached prefix crashes EngineCore for all co-tenantsCVE-2026-116180 · vLLM (LogprobsTensors.empty_cpu / V1 model runner prompt_logprobs with KV connector)High
- Dagster: Vulnerability in Dagster Core prior to 1.13.1CVE-2026-41490 · DagsterHigh
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.