Description
vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with max_tokens=0 to exhaust decode-worker memory without bound until the worker restarts.
Metrics
CVSS 3.1 7.5/10
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:HWeakness enumeration
Affected software
| Vendor | Product | Versions |
|---|---|---|
| vllm-project | vllm | <= 0.29.0 |
References
- https://github.com/vllm-project/vllmProduct
- https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/distributed/kv_transfer/kv_connector/v1/nixl/push_worker.py#L162-L181Product
- https://github.com/vllm-project/vllm/pull/55677Issue Tracking, Patch
- https://www.vulncheck.com/advisories/vllm-through-0.29.0-memory-exhaustion-via-rejected-requestsThird Party Advisory
Timeline
- Published
- Sep 17, 2026
- Last modified
- Sep 28, 2026
- NVD status
- Analyzed