Apache HTTP/2 lettering beside a server with a yellow barrier
Security / The story

Apache fixes a critical HTTP/2 flaw. Check the module as well as the server

Version 2.4.71 closes a memory-corruption issue; configuration and distribution packages determine which installations need action.

GeekRadarPublished 4 min read

Apache HTTP Server 2.4.71 closes a critical flaw in HTTP/2 request handling, but identifying an exposed installation takes more than reading its version banner. Administrators need to establish whether the vulnerable module is loaded, whether HTTP/2 is enabled and whether their operating-system vendor has supplied a corrected package. Those three checks decide what needs attention.

Apache’s November 11 advisory identifies CVE-2026-119618 as a memory-corruption vulnerability that can permit remote code execution. The affected upstream HTTP Server range is 2.4.55 through 2.4.70. The project also identifies affected standalone mod_h2 releases, so environments that package the module separately should check that component’s advisory rather than assume the server number tells the whole story.

The feature that creates the exposure

HTTP/2 lets several requests share a connection. That is useful when a page needs images, scripts and styles without giving every transfer its own connection. Inside the server, that concurrency needs careful coordination: work happening for one request must not damage memory being used by another.

Here, the failure is in that coordination. Apache describes improper synchronization in mod_http2’s assign_headers() handling. Red Hat’s technical bulletin explains that the problem can corrupt memory through concurrent request processing. A reachable affected server does not require an attacker to log in first.

That makes the configuration boundary consequential. Red Hat says HTTP/1.1-only servers are not vulnerable to this particular flaw. An installed Apache package is therefore a reason to investigate, not proof that the service is exposed. Conversely, a login page in front of an application does not fix a problem in the web server processing the request before the application gets involved.

A distribution package has its own version history

The upstream fix is in 2.4.71. That does not mean every Linux administrator should replace a supported distribution package with a manually compiled upstream server. Vendors can apply security corrections to their maintained package branches while preserving older-looking version numbers. The relevant evidence is the vendor’s advisory and installed package release, not an internet scanner’s interpretation of a short banner.

Red Hat identifies RHEL 9, RHEL 10 and JBoss Core Services among affected products, and notes that mod_http2 is a separate package on the two listed RHEL branches. Its bulletin says fixes are available for most affected products, with remaining work being expedited. That wording is a reason to check the exact product entry; it does not establish that every supported configuration has received the same update.

The same discipline applies to appliances and containers. Finding one updated host does not establish that every image or bundled service uses that host’s package. Inventory the component that actually accepts traffic. This is the recurring practical lesson behind recent NetScaler remediation: the product name points you toward the investigation, while the deployed configuration determines the action.

Temporary mitigation needs verification too

Where an update cannot be applied immediately, Red Hat documents removing HTTP/2 from the active protocol configuration and restarting the service so it serves HTTP/1.1. That is a temporary reduction in exposure with a potential performance cost, not a substitute for installing the correction. Test it against the actual application and deployment arrangement rather than copying a generic configuration into production.

Afterward, verify what the running server negotiates. A saved file is not evidence that a service reloaded it, and a public test that reaches only a reverse proxy may say little about the origin behind it. Keep the checks aligned with the component being protected.

Apache credits separate reports from Anthropic and Google researchers. The useful outcome for operators is concrete: a corrected release and an identifiable exposure condition. Closing this incident means knowing which service was vulnerable, which build replaced it and what is now running—not merely seeing an update command finish.