Description
LightLLM through 1.2.0 contains a memory exhaustion vulnerability in the NCCL control channel when started with --pd_trans_mode nccl, allowing unauthenticated attackers to exhaust KV-transfer worker memory. Attackers can call the exposed_set_value method to store unbounded key-value pairs without size limits, causing the worker process to crash and triggering node failure.
Affected products
| Vendor | Product | Affected | Fixed |
|---|---|---|---|
| ModelTC | LightLLM | ≤ 1.2.0 | — |
References
- GitHub Issue #1595 issue-tracking
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/mode_backend/pd/nccl_kv_transporter.py#L416-L420 technical-description
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/mode_backend/pd/nccl_kv_transporter.py#L474-L485 technical-description
- https://github.com/ModelTC/lightllm/blob/v1.2.0/lightllm/server/router/model_infer/mode_backend/pd/base_kv_move_manager.py#L121-L122 technical-description
- https://github.com/ModelTC/LightLLM product
- VulnCheck Advisory: LightLLM through 1.2.0 Unauthenticated Memory Exhaustion via NCCL Control Channel set_value third-party-advisory