What happened
The vLLM Rust gRPC frontend does not enforce the logprob_token_ids limit, so a caller can send Generate or GenerateStream requests with output_logprobs enabled and more than the permitted number of output logprob_token_ids, exhausting resources or terminating EngineCore.
Why it matters
Allows a remote caller to terminate the engine and disrupt co-tenant serving (CVSS 5.9, CWE-667).
Attack vector
Send a Generate/GenerateStream request with output_logprobs enabled and more than 128 output logprob_token_ids.
Affected systems
vLLM >= 0.22.0, < 0.24.0 (Rust gRPC frontend).
Mitigation
Upgrade to vLLM 0.24.0.