Description
LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint. Attackers can send a single request whose parameters cause pickle.loads to fail in InferSamplingParams, killing the inference thread while /health stays up and all later requests hang until restart.
Affected products
| Vendor | Product | Affected | Fixed |
|---|---|---|---|
| ModelTC | LightLLM | ≤ 1.2.0 | — |
References
- https://github.com/ModelTC/LightLLM/issues/1629 issue-tracking exploit
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/api_lightllm.py#L37-L43 technical-description
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/core/objs/sampling_params.py#L356-L357 technical-description
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/infer_batch.py#L484-L485 technical-description
- https://github.com/ModelTC/LightLLM product
- VulnCheck Advisory: LightLLM through 1.2.0 Unauthenticated DoS via pd_kv_trans_params Deserialization third-party-advisory