Summary
Fluentd's out_http output plugin previously allowed placeholder values in the endpoint URL, such as ${tag}, to be populated from untrusted input.
When those dynamic values influenced the destination hostname, an attacker could redirect outbound HTTP requests to arbitrary internal services, enabling server-side request forgery (SSRF) and potential access to protected resources.
Affected context
This vulnerability arises in Fluentd deployments that use the out_http output plugin with endpoint configurations containing dynamic placeholders (for example, ${tag}) that can be influenced by external or semi-trusted data.
It is most relevant where Fluentd is allowed to make HTTP calls to internal networks, metadata services, or sensitive APIs, and where there is limited validation or restriction on the resolved destination hostnames.
Exploit preconditions
- Attack surface: HTTP output path in the Fluentd out_http plugin.
- Required access: Ability to inject or control data that resolves into endpoint placeholders (e.g., tags or fields derived from log/event input).
- Trigger point: Fluentd processing of events where the endpoint URL is constructed using placeholder values and a request is sent.
- Why exploitation is possible: The plugin trusts placeholder-derived data when building the endpoint URL, allowing untrusted values to control the destination hostname for outbound requests without sufficient validation or scoping.
Why this happens
The core issue is insufficient validation of dynamic endpoint configuration in the out_http plugin, where placeholders like ${tag} are directly substituted into the destination URL.
This aligns with CWE-918 (Server-Side Request Forgery): server-side code constructs outbound requests using input that can be influenced by an attacker, without enforcing strict boundaries on where those requests may go.
Abstractly, the insecure pattern is treating configuration-derived routing data as trustworthy, even when it may originate from or be shaped by untrusted event attributes, enabling attackers to re-route traffic to internal systems.
Potential impact
- Fluentd may send HTTP requests to internal network services that are not otherwise exposed externally.
- Attackers could probe internal APIs, databases, or admin interfaces reachable via HTTP.
- Sensitive data or configuration information could be accessed from internal endpoints (e.g., cloud metadata services).
- Internal services could be abused as part of lateral movement or chained attacks.
- Logging or monitoring infrastructure could be leveraged as a blind proxy to other protected resources.
General mitigation guidance
Mitigation generally centers on limiting how untrusted data can influence outbound request destinations and ensuring strict boundaries around network egress from logging components.
Common defensive patterns include:
- Avoiding or severely restricting the use of dynamic placeholders in endpoint URLs, especially those derived from user or event input.
- Applying strong validation and allowlists to any configuration values that can affect hostnames, ports, or schemes used for outbound HTTP calls.
- Confining Fluentd's network access through firewall rules, service meshes, or egress policies so that it can only reach explicitly approved endpoints.
- Treating log tags and event attributes as untrusted input and sanitizing or normalizing them before they are used to construct routing decisions.
- Designing output plugins and configurations so that operational flexibility does not rely on arbitrary user-controlled interpolation in critical network parameters.
How teams detect issues like this
Detection typically focuses on both configuration inspection and behavioral monitoring of outbound requests.
Approaches include:
- Static review of Fluentd configuration files to identify out_http endpoints that embed placeholders in hostnames or complete URLs, especially where placeholder sources are influenced by external events.
- Automated policy checks in CI/CD pipelines to flag dynamic URL construction, missing allowlists for destinations, or configurations that permit user-derived data in routing parameters.
- Network monitoring and logging to detect unusual or unexpected outbound connections from Fluentd instances, particularly to internal or high-value services.
- Code review of custom plugins or extensions for patterns where request destinations are determined by inputs that are not explicitly validated against approved targets.

