Apache HTTP/2 RCE (CVE-2026-119618): A 9.8 You Can Rule Out in One Check
Bit4Bit DFIR, practical notes for people who actually run the systems.
On November 11, 2026, Apache shipped httpd 2.4.71 with a fix for a flaw that is easy to overstate and easy to miss. CVE-2026-119618 is a critical race condition in the HTTP/2 module, mod_http2. A remote attacker needs no login and no user interaction, and a successful attack can run code with the privileges of the Apache process. But the flaw is only reachable when the server actually negotiates HTTP/2, and plenty of Apache installs never do. One configuration check tells you which side you are on.
Apache rates the issue Critical. Red Hat assigns CVSS 3.1 9.8, with a network attack vector, low complexity, no privileges required, and no user interaction. No NVD score had been published at the time of writing.
What is actually affected
- Apache HTTP Server 2.4.55 through 2.4.70, with HTTP/2 enabled
- Separately distributed mod_h2 releases 2.0.8 through 2.0.42
- Fixed in httpd 2.4.71, released November 11, 2026
The precondition is the whole story. The vulnerable code path is only reachable when the server negotiates HTTP/2, through the Protocols directive with h2 or h2c. Apache's default is HTTP/1.1. A server that only serves HTTP/1.1 is not affected, even if mod_http2 is installed and loaded. If HTTP/2 is enabled on only one virtual host, that virtual host is the exposed surface.
If you cannot patch immediately, disabling HTTP/2 and serving HTTP/1.1 only removes the exposure. That is a real mitigation here, not a workaround that leaves the hole open.
How the flaw works
The bug is a race condition, CWE-362, in assign_headers() in mod_http2. The module is designed so worker threads do not use the connection thread's shared memory allocator. Each worker gets a private allocator instead. For one request shape, that rule breaks: a request with no body, a Content-Type header, and no Content-Length, arriving when the request header table is full. In that case the worker writes a synthesized Content-Length: 0 into the shared stream header table, and that write races with the connection thread on a lock-free free list. The result is heap memory corruption, which an attacker can turn into remote code execution.
The fix is a one-line change in the upstream commit: the write goes to the current request's own input header table instead of the shared request header table.
Apache credits Anthropic's OSS Scanner, which reported the flaw on October 26, and Anthony Weems and Simon Scannell of Google Cloud Vulnerability Research, who reported it on November 3.
How close is exploitation
This is where sources disagree, so here is the disagreement plainly.
Red Hat's CVE page says public research has demonstrated a reliable exploit against a default-configured server with mod_http2 enabled. A public crash-reproduction lab appeared on November 12, showing corruption on 2.4.70 under AddressSanitizer and ThreadSanitizer, and a clean run on 2.4.71 under the same traffic. That is a reproduction, not by itself a working remote code execution exploit.
Other reporting says no exploit was available and no exploitation in the wild had been reported. As of writing, the CVE is not in CISA's Known Exploited Vulnerabilities catalog. The safe reading: no confirmed attacks yet, but at least one major vendor says reliable exploitation has been demonstrated, and crash reproduction tooling is public. Patch on the vendor's word, not on the absence of attacks.
Vendor packages are moving: Red Hat shipped a fixed mod_http2 package for RHEL 9, and Debian unstable has 2.4.71-1. Check your own distribution's package rather than assuming a recent update covered this one. A server updated for an earlier Apache flaw is not automatically protected here.
What to do
If you run Apache with HTTP/2 enabled:
- Update to httpd 2.4.71 or install your distribution's fixed package, then restart Apache. The old process keeps serving the vulnerable code until it is restarted.
- Confirm the running version after the restart, not just the installed package.
- Rebuild container images that bundle Apache. Patching a running container is not enough if the image still ships a vulnerable build.
If you cannot patch immediately, remove h2 and h2c from the Protocols directive and serve HTTP/1.1 only, then restart Apache. Verify with a request that the server no longer negotiates HTTP/2.
If your Apache only serves HTTP/1.1, no action is needed for this CVE. Confirm the Protocols setting rather than assuming it, since HTTP/2 can be enabled on one virtual host while the rest of the server looks untouched.
The DFIR angle
If you have HTTP/2-enabled Apache hosts on affected versions, check for signs someone already probed or hit the flaw:
- Look for httpd child processes crashing or restarting unexpectedly around November 11 onward, especially on HTTP/2 virtual hosts.
- Review access and error logs for unusual HTTP/2 request patterns, including requests with a Content-Type header and no body or Content-Length.
- Check for processes running as the Apache user that should not be there, and for unexpected outbound connections from the web server.
- Treat any host that was both exposed and unpatched after November 11 as suspect until the logs say otherwise. Patching closes the hole, it does not undo code that already ran.
The lesson here is a familiar one: the severity score is not the risk. The risk is the score multiplied by whether the vulnerable code path is reachable on your server. For this CVE, that is one directive, and it is worth checking before you spend the weekend patching servers that were never exposed.
Sources: Apache httpd 2.4.71 release and security advisory, November 11, 2026; Red Hat CVE page for CVE-2026-119618 (CVSS 3.1 9.8); CIRCL TR-103 advisory, November 12, 2026; The Hacker News and Bitsight reporting on the disclosure and vendor fixes. Exploitation status differs by source, as noted above; the CVE was not in CISA KEV at the time of writing.
Top comments (0)