CVE-2025-4366 Overview
CVE-2025-4366 is an HTTP request smuggling vulnerability in Cloudflare's pingora-proxy, the proxying component of the open-source Pingora framework. The flaw allows attackers to inject malicious HTTP requests through manipulated request bodies when a cache HIT occurs on an HTTP/1.1 connection. Successful exploitation enables unauthorized request execution and cache poisoning against downstream origins. The vulnerability is tracked under CWE-444, Inconsistent Interpretation of HTTP Requests.
Critical Impact
An attacker can smuggle arbitrary headers and URLs into subsequent requests on the same HTTP/1.1 connection, poisoning cached responses served to other clients.
Affected Products
- Cloudflare Pingora (open-source proxy framework)
- Applications using the pingora-proxy module with caching enabled
- HTTP/1.1 connections proxied through vulnerable Pingora builds
Discovery Timeline
- 2025-05-22 - CVE-2025-4366 published to NVD
- 2026-06-17 - Last updated in NVD database
Technical Details for CVE-2025-4366
Vulnerability Analysis
The vulnerability lives in how pingora-proxy handles the request body on a cache HIT. When Pingora serves a cached response, it does not correctly drain or delimit the body of the incoming HTTP/1.1 request on the persistent connection. An attacker can append a second, attacker-controlled HTTP request inside the body of the first. Pingora treats the smuggled bytes as a new request on the same keep-alive connection, applying attacker-supplied headers and URLs to downstream processing.
The issue is classic HTTP request smuggling but triggered specifically by the cache-serving path. Because Pingora is deployed as a front-end proxy for high-traffic origins, one poisoned cache entry can be served to many clients before it expires.
Root Cause
The root cause is inconsistent parsing of HTTP/1.1 request framing between the cache layer and the connection handler. On a cache HIT, pingora-proxy did not fully consume the request body before reusing the connection for the next request. Trailing bytes controlled by the attacker were then parsed as a fresh HTTP request. The fix is present in commit fda3317ec822678564d641e7cf1c9b77ee3759ff in the Cloudflare Pingora repository.
Attack Vector
Exploitation requires a network-reachable Pingora deployment configured for HTTP caching and reachable over HTTP/1.1. The attacker sends a crafted request whose body contains a fully formed second HTTP request. When the outer request produces a cache HIT, Pingora emits the cached response and then interprets the leftover bytes as a new request. The smuggled request can target arbitrary URLs and inject arbitrary headers, enabling cache poisoning, credential theft through host header manipulation, and bypass of access controls enforced upstream. The published CVSS 4.0 score is 7.4, and exploitation requires user interaction and higher attack complexity, according to the vector recorded in NVD. No public exploit or CISA KEV entry exists at the time of publication, and the EPSS probability is approximately 0.42%.
No verified proof-of-concept code is public. Refer to the Cloudflare Pingora repository for the patched implementation and test cases.
Detection Methods for CVE-2025-4366
Indicators of Compromise
- Unexpected cache entries containing content from URLs that no legitimate client requested.
- HTTP/1.1 requests where the body length disagrees with Content-Length or Transfer-Encoding on connections terminated by Pingora.
- Origin logs showing requests with Host or path values that do not correspond to any front-end route.
- Sudden divergence between edge cache keys and origin access logs for the same URL.
Detection Strategies
- Inspect proxy access logs for consecutive requests on the same keep-alive connection where the second request arrived with sub-millisecond delay after a cache HIT.
- Deploy a Web Application Firewall rule that rejects HTTP/1.1 requests presenting both Content-Length and Transfer-Encoding: chunked headers.
- Correlate cache write events with upstream fetch logs; unexplained cache writes indicate poisoning.
- Run smuggling probes such as body-length desync tests against Pingora edges in a staging environment.
Monitoring Recommendations
- Forward Pingora access logs and origin logs into a unified analytics pipeline to correlate request framing anomalies.
- Alert on spikes in 4xx responses following cache HITs, which often accompany smuggling attempts.
- Track the version of Pingora deployed on each edge node and alert when an unpatched build is running.
How to Mitigate CVE-2025-4366
Immediate Actions Required
- Upgrade Pingora to a build that includes commit fda3317ec822678564d641e7cf1c9b77ee3759ff or later.
- Audit cache contents and purge entries created during the exposure window.
- Restrict client connections to HTTP/2 or HTTP/3 where feasible, since the flaw is specific to HTTP/1.1 framing.
- Review origin logs for anomalous Host, Authorization, or path values consistent with smuggled requests.
Patch Information
Cloudflare fixed the vulnerability in the Pingora source tree. Consumers must rebuild any downstream binary that vendors pingora-proxy against the patched revision. The fix commit fda3317ec822678564d641e7cf1c9b77ee3759ff is available in the Cloudflare Pingora GitHub repository.
Workarounds
- Disable HTTP caching in pingora-proxy until the patched build is deployed.
- Terminate HTTP/1.1 keep-alive after each request by forcing Connection: close on responses served from cache.
- Front Pingora with a proxy that strictly validates Content-Length and Transfer-Encoding headers and rejects conflicting framing.
- Restrict clients to HTTP/2 or HTTP/3, which use length-prefixed framing and are not affected by this class of smuggling.
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
