Overview
CVE-2025-30202 is a critical vulnerability affecting vLLM, a high-throughput and memory-efficient inference and serving engine. It exposes the system to potential denial of service (DoS) attacks and data leakage via ZeroMQ in multi-node vLLM deployment. This vulnerability poses a significant threat to all entities utilizing vLLM versions from 0.5.2 and prior to 0.8.5. It is noteworthy due to its potential to compromise system integrity and confidentiality.
Vulnerability Summary
CVE ID: CVE-2025-30202
Severity: High (7.5)
Attack Vector: Network
Privileges Required: None
User Interaction: None
Impact: Potential system compromise and data leakage
Affected Products
Escape the Surveillance Era
Most apps won’t tell you the truth.
They’re part of the problem.
Phone numbers. Emails. Profiles. Logs.
It’s all fuel for surveillance.
Ameeba Chat gives you a way out.
- • No phone number
- • No email
- • No personal info
- • Anonymous aliases
- • End-to-end encrypted
Chat without a trace.
Product | Affected Versions
vLLM | 0.5.2 to 0.8.4
How the Exploit Works
In a multi-node vLLM deployment, vLLM makes use of ZeroMQ for certain multi-node communication functions. The primary vLLM host opens an XPUB ZeroMQ socket and binds it to all interfaces. While the socket is typically opened for a multi-node deployment, it is only utilized when conducting tensor parallelism across multiple hosts.
Any client with network access to this host can connect to this XPUB socket unless its port is blocked by a firewall. Once connected, these arbitrary clients will receive all of the same data broadcasted to all of the secondary vLLM hosts. This data is internal vLLM state information that is not useful to an attacker. However, by potentially connecting to this socket many times and not reading the data published to them, an attacker can cause a DoS attack by slowing down or potentially blocking the publisher.
Conceptual Example Code
The following conceptual code represents how an attacker might continuously connect to the XPUB socket without reading the data, leading to potential DoS:
import zmq
context = zmq.Context()
socket = context.socket(zmq.SUB)
socket.connect("tcp://target_host:target_port")
while True:
# Continuously connect without reading the data
socket.recv_string(flags=zmq.NOBLOCK)