vLLM Vulnerable to Remote Code Execution via Mooncake Integration
Critical severity
GitHub Reviewed
Published
Apr 29, 2025
in
vllm-project/vllm
•
Updated May 29, 2025
Description
Published to the GitHub Advisory Database
Apr 29, 2025
Reviewed
Apr 29, 2025
Published by the National Vulnerability Database
Apr 30, 2025
Last updated
May 29, 2025
Impacted Deployments
Note that vLLM instances that do NOT make use of the mooncake integration are NOT vulnerable.
Description
vLLM integration with mooncake is vaulnerable to remote code execution due to using
pickle
based serialization over unsecured ZeroMQ sockets. The vulnerable sockets were set to listen on all network interfaces, increasing the likelihood that an attacker is able to reach the vulnerable ZeroMQ sockets to carry out an attack.This is a similar to GHSA - x3m8 - f7g5 - qhm7, the problem is in
https://github.com/vllm-project/vllm/blob/32b14baf8a1f7195ca09484de3008063569b43c5/vllm/distributed/kv_transfer/kv_pipe/mooncake_pipe.py#L179
Here recv_pyobj() Contains implicit
pickle.loads()
, which leads to potential RCE.References