Vulnerability in ModelTC LightLLM

[ Why it matters ]Priority signals

• Public proof-of-concept exploit exists (CISA SSVC)

[ Description ]Official 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.

[ Fixes ]Fixes and patches

No fix or interim patch is linked from our sources yet. Check the vendor references below and limit exposure of affected systems until a fix is available.

[ References ]References

• github.com/ModelTC/LightLLM
• github.com/ModelTC/LightLLM/issues/1629
• github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/api_lightllm.py
• github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/core/objs/sampling_params.py
• github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/infer_batch.py
• www.vulncheck.com/advisories/lightllm-through-1.2.0-unauthenticated-dos-via-pd-kv-trans…