vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).
Attack Parameters
Attack Vector
Network
Can be exploited remotely
Attack Complexity
Low
Easy to exploit
Attack Requirements
None
No additional conditions
Privileges Required
Low
Basic privileges needed
User Interaction
None
No user interaction needed
Impact Assessment
Confidentiality
None
No data leak
Integrity
None
No data modification
Availability
High
Complete denial of service
CVSS Vector v4.0
Weakness Type (CWE)
Vulnerable Products 1
| Configuration | From (including) | Up to (excluding) |
|---|---|---|
|
Vllm Vllm
cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*
|
— |
0.29.0
|