Summary
A Server-Side Request Forgery (SSRF) vulnerability exists in the MediaConnector class within the vLLM project's multimodal feature set. The load_from_url and load_from_url_async methods fetch and process media from user-provided URLs without adequate restrictions on the target hosts. This allows an attacker to coerce the vLLM server into making arbitrary requests to internal network resources.
This vulnerability is particularly critical in containerized environments like llm-d, where a compromised vLLM pod could be used to scan the internal network, interact with other pods, and potentially cause denial of service or access sensitive data. For example, an attacker could make the vLLM pod send malicious requests to an internal llm-d management endpoint, leading to system instability by falsely reporting metrics like the KV cache state.
Vulnerability Details
The core of the vulnerability lies in the MediaConnector.load_from_url method and its asynchronous counterpart. These methods accept a URL string to fetch media content (images, audio, video).
https://github.com/vllm-project/vllm/blob/119f683949dfed10df769fe63b2676d7f1eb644e/vllm/multimodal/utils.py#L97-L113
The function directly processes URLs with http, https, and file schemes. An attacker can supply a URL pointing to an internal IP address or a localhost endpoint. The vLLM server will then initiate a connection to this internal resource.
- HTTP/HTTPS Scheme: An attacker can craft a request like
{"image_url": "http://127.0.0.1:8080/internal_api"}. The vLLM server will send a GET request to this internal endpoint.
- File Scheme: The
_load_file_url method attempts to restrict file access to a subdirectory defined by --allowed-local-media-path. While this is a good security measure for local file access, it does not prevent network-based SSRF attacks.
Impact in llm-d Environments
The risk is significantly amplified in orchestrated environments such as llm-d, where multiple pods communicate over an internal network.
-
Denial of Service (DoS): An attacker could target internal management endpoints of other services within the llm-d cluster. For instance, if a monitoring or metrics service is exposed internally, an attacker could send malformed requests to it. A specific example is an attacker causing the vLLM pod to call an internal API that reports a false KV cache utilization, potentially triggering incorrect scaling decisions or even a system shutdown.
-
Internal Network Reconnaissance: Attackers can use the vulnerability to scan the internal network for open ports and services by providing URLs like http://10.0.0.X:PORT and observing the server's response time or error messages.
-
Interaction with Internal Services: Any unsecured internal service becomes a potential target. This could include databases, internal APIs, or other model pods that might not have robust authentication, as they are not expected to be directly exposed.
Delegating this security responsibility to an upper-level orchestrator like llm-d is problematic. The orchestrator cannot easily distinguish between legitimate requests initiated by the vLLM engine for its own purposes and malicious requests originating from user input, thus complicating traffic filtering rules and increasing management overhead.
Fix
See the --allowed-media-domains option discussed here: https://docs.vllm.ai/en/latest/usage/security.html#4-restrict-domains-access-for-media-urls
References
kexinoh Finder
d3do-23 Reporter
lonelyuan Reporter
huachenheli Remediation developer
DarkLight1337 Remediation reviewer
russellb Coordinator
sidhpurwala-huzaifa Coordinator
Summary
A Server-Side Request Forgery (SSRF) vulnerability exists in the
MediaConnectorclass within the vLLM project's multimodal feature set. Theload_from_urlandload_from_url_asyncmethods fetch and process media from user-provided URLs without adequate restrictions on the target hosts. This allows an attacker to coerce the vLLM server into making arbitrary requests to internal network resources.This vulnerability is particularly critical in containerized environments like
llm-d, where a compromised vLLM pod could be used to scan the internal network, interact with other pods, and potentially cause denial of service or access sensitive data. For example, an attacker could make the vLLM pod send malicious requests to an internalllm-dmanagement endpoint, leading to system instability by falsely reporting metrics like the KV cache state.Vulnerability Details
The core of the vulnerability lies in the
MediaConnector.load_from_urlmethod and its asynchronous counterpart. These methods accept a URL string to fetch media content (images, audio, video).https://github.com/vllm-project/vllm/blob/119f683949dfed10df769fe63b2676d7f1eb644e/vllm/multimodal/utils.py#L97-L113
The function directly processes URLs with
http,https, andfileschemes. An attacker can supply a URL pointing to an internal IP address or alocalhostendpoint. The vLLM server will then initiate a connection to this internal resource.{"image_url": "http://127.0.0.1:8080/internal_api"}. The vLLM server will send a GET request to this internal endpoint._load_file_urlmethod attempts to restrict file access to a subdirectory defined by--allowed-local-media-path. While this is a good security measure for local file access, it does not prevent network-based SSRF attacks.Impact in
llm-dEnvironmentsThe risk is significantly amplified in orchestrated environments such as
llm-d, where multiple pods communicate over an internal network.Denial of Service (DoS): An attacker could target internal management endpoints of other services within the
llm-dcluster. For instance, if a monitoring or metrics service is exposed internally, an attacker could send malformed requests to it. A specific example is an attacker causing the vLLM pod to call an internal API that reports a false KV cache utilization, potentially triggering incorrect scaling decisions or even a system shutdown.Internal Network Reconnaissance: Attackers can use the vulnerability to scan the internal network for open ports and services by providing URLs like
http://10.0.0.X:PORTand observing the server's response time or error messages.Interaction with Internal Services: Any unsecured internal service becomes a potential target. This could include databases, internal APIs, or other model pods that might not have robust authentication, as they are not expected to be directly exposed.
Delegating this security responsibility to an upper-level orchestrator like
llm-dis problematic. The orchestrator cannot easily distinguish between legitimate requests initiated by the vLLM engine for its own purposes and malicious requests originating from user input, thus complicating traffic filtering rules and increasing management overhead.Fix
See the
--allowed-media-domainsoption discussed here: https://docs.vllm.ai/en/latest/usage/security.html#4-restrict-domains-access-for-media-urlsReferences