Incorrect Calculation of Buffer Size in vLLM - #VU128747

Detailed vulnerability description

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

The vulnerability exists due to incorrect calculation of buffer size in the extract_hidden_states speculative decoding proposer when processing requests with sampling penalty parameters. A remote user can send a specially crafted request with a penalty parameter to cause a denial of service.

Only deployments configured with extract_hidden_states speculative decoding are vulnerable.

Remediation

Sources

• https://github.com/vllm-project/vllm/security/advisories/GHSA-83vm-p52w-f9pw
• https://github.com/vllm-project/vllm/pull/38610