Resource exhaustion in vLLM - CVE-2025-29770

Description

The vulnerability allows a remote user to cause a denial of service.

The vulnerability exists due to uncontrolled resource consumption in outlines grammar cache in vllm/model_executor/guided_decoding/outlines_logits_processors.py when handling decoding requests with unique schemas. A remote user can send a stream of very short decoding requests with unique schemas to cause a denial of service.

The issue applies to the V0 engine only. The outlines backend can also be selected on a per-request basis using the guided_decoding_backend key in the extra_body field.

Remediation

External links

• https://github.com/vllm-project/vllm/security/advisories/GHSA-mgrm-fgjv-mhv8
• https://github.com/vllm-project/vllm/blob/53be4a863486d02bd96a59c674bbec23eec508f6/vllm/model_executor/guided_decoding/outlines_logits_processors.py