NGINX CVE-2026-123007: HTTP/2 heap overflow, CVSS 9.2
A critical NGINX flaw in HTTP/2 handling can crash workers and may allow code execution. See affected versions, how to check one server and how to fix it.
TL;DR
- What: a heap buffer overflow in NGINX when it handles HTTP/2 traffic, tracked as CVE-2026-123007.
- Impact: an unauthenticated attacker over the network can crash an NGINX worker process and corrupt its memory. Code execution is possible where ASLR is disabled or can be bypassed.
- Severity: a CVSS v4.0 score of 9.2, rated critical, and a CVSS v3.1 score of 8.1 from F5.
- Exploited: no source we checked says so yet. It was published on November 18, 2026.
- Fix: upgrade NGINX Open Source to 1.30.7 or 1.31.9, whichever matches your line, and follow F5's advisory for NGINX Plus and the Kubernetes products.
What the flaw is, in plain English
HTTP/2 lets a browser send many requests over one connection. To keep a fast sender from overwhelming a slow receiver, the protocol uses a flow control window: a running allowance of how much data may be in flight. CVE-2026-123007 is about that allowance. According to the NVD description, when NGINX processes certain undisclosed traffic, the window can grow past the allowed threshold, which can cause a heap buffer overflow.
In the best case for a defender, the overflow ends in a segmentation fault and the worker process restarts. NGINX starts a new worker, but every connection that worker held is dropped, and an attacker who repeats the trigger can keep doing that. In the worse case the overflow corrupts memory. The description says code execution is possible when Address Space Layout Randomization (ASLR) is switched off, or when an attacker finds a way around it. The weaknesses are catalogued as CWE-191 (integer underflow) and CWE-787 (out-of-bounds write).
The description also says this is a data plane issue only, with no control plane exposure. For NGINX that means the workers that serve your traffic are affected, not the management interface.
Why it matters now
NGINX sits in front of a very large number of sites and APIs: as a web server, a reverse proxy, a load balancer and, in Kubernetes, as an ingress controller. It is usually the first thing the internet touches, and it is reachable without logging in. A flaw in it that starts with nothing more than network access is the kind that automated scanners look for once details are public.
Two things make this one worth acting on quickly even though it is not reported as exploited. First, the severity: NVD records a CVSS v4.0 score of 9.2, and the vector shows no privileges and no user interaction are needed. Second, the conditions are common. HTTP/2 is enabled on a great many public NGINX servers, and many teams run several NGINX instances they do not track in one place: a proxy on a build host, an ingress controller in a cluster, a base image used by three services.
The score is not a prediction of exploitation. The CVSS v3.1 score from F5 is 8.1, lower than the v4.0 figure, because the vector marks the attack as complex to carry out. Use both numbers as a way to rank the work, not to decide whether to do it. For earlier NGINX and web server problems, see our notes on a Linux kernel flaw that can crash web and SSH servers and on the HTTP/2 flaws fixed in Node.js.
Am I affected?
NVD lists these releases as affected, all from F5 unless noted:
- NGINX Open Source from 1.9.5 up to 1.30.7 (excluding), and from 1.31.0 up to 1.31.9 (excluding).
- NGINX Plus from 37.0.0 up to 37.0.8.1 (excluding), and from 37.1.0 up to 37.3.0.1 (excluding), plus the R34, R35 and R36 releases.
- NGINX Ingress Controller from 5.0.0 up to 5.8.0 (excluding), plus other lines listed in the entry.
- NGINX Gateway Fabric, NGINX Instance Manager, NGINX App Protect WAF and NGINX App Protect DoS, each with ranges in the entry.
The entry also lists Red Hat among the affected vendors. The description adds that releases which have reached End of Technical Support were not evaluated, so an old release that is not on the list is not necessarily safe.
To check one machine by hand, start with the version, then see whether HTTP/2 is turned on anywhere in the configuration:
nginx -v
nginx -T 2>/dev/null | grep -n http2The second command prints the full effective configuration and shows every line that mentions HTTP/2. If it prints nothing, the described trigger is not enabled in that configuration. Run the commands inside a container for containerised NGINX, and on the host for a packaged one. If NGINX runs as an ingress controller, read the controller image tag from the deployment instead.
Distribution backports. Debian, Ubuntu, Red Hat and other distributions often fix a flaw without changing the upstream version number, so a package that reports an older version may already be patched. Do not decide from the version string alone: read your distribution's security tracker for this CVE and the changelog of the installed nginx package. Packages from nginx.org follow the upstream numbering.
How to fix CVE-2026-123007
- List every NGINX you run: servers, container images, base images your teams build from, and ingress controllers. Write down the version of each and which of the ranges above it falls into.
- Upgrade NGINX Open Source to 1.30.7 if your version is below that, or to 1.31.9 if you run the newer branch, or to a later release. For NGINX Plus, the Ingress Controller, Gateway Fabric and the other products, install the release F5's advisory names. For a distribution package, install its security update and confirm in its changelog that this CVE is listed.
- Restart or reload NGINX so the new binary is really running, then check the version again with
nginx -v. For containers, rebuild the image and redeploy it, because patching the host does nothing for an image. - If you cannot upgrade today and a server does not need HTTP/2, remove it from the listen directives of that server and reload. This removes the condition the description names. It is a stopgap, and it may slow clients down, so plan the upgrade as well.
- Read the F5 advisory for anything specific to your release, and check your logs for repeated worker process crashes, which would be a sign that someone has been trying the trigger.
The NVD entry does not give a fixed release for every product as a single line, so the numbers above come from the end of each affected range. Confirm them against the vendor advisory before you roll out.
How to check with Stackurity
Checking one server by hand is fine. Checking every server, container image and cluster owned by different teams is where this kind of issue slips through.
Stackurity keeps one live inventory of the software on your servers, deployed applications, repositories and container images, and matches it against new vulnerability advisories every hour. Its read-only agent finds server software such as NGINX on Linux servers, including nginx from nginx.org, and, when you opt in with the --docker flag, in the images of running containers. When an advisory matches, the finding names the version in use and the version to move to, and one fix groups every place that needs the same upgrade. The remediation documentation explains how fixes are ordered, most urgent first.
After you upgrade, Check again asks the agent for a fresh report and records whether the issue is really gone, so you do not have to take a ticket status on trust. There are limits. Stackurity does not cover appliance firmware or libraries copied into an application outside a package manager. It also does not read your NGINX configuration, so confirming whether HTTP/2 is enabled stays a manual check.
Start a free trial of Stackurity to see which of your servers and images run NGINX today.
FAQ
What is CVE-2026-123007?
It is a heap buffer overflow in NGINX configurations that use HTTP/2. When the flow control window exceeds the allowed threshold, a worker process can crash, and an attacker may be able to corrupt memory or run code, especially where ASLR is disabled or can be bypassed.
Which NGINX versions are affected?
NGINX Open Source from 1.9.5 up to 1.30.7 (excluding) and from 1.31.0 up to 1.31.9 (excluding) are listed, along with several NGINX Plus, Ingress Controller, Gateway Fabric and WAF releases. Check F5's advisory for the exact release that fixes your product.
Is CVE-2026-123007 exploited in the wild?
No source we checked says it is exploited, and we have not seen it in the CISA Known Exploited Vulnerabilities catalogue. It was published on November 18, 2026, so treat the upgrade as urgent for any server that accepts HTTP/2 from the internet.
Am I safe if I do not use HTTP/2?
The description says the issue affects configurations that use HTTP/2, so a server that does not enable it is outside the described trigger. Configurations drift, though, so upgrade anyway and confirm what each server really enables.
Does ASLR protect me?
It raises the bar for code execution, but it is not a fix. The description says code execution is possible when ASLR is off or can be bypassed, and a crash of the worker process is possible either way.
Does Stackurity find vulnerable NGINX versions?
Yes, for NGINX it finds on Linux servers and in images, including nginx from nginx.org. It does not cover appliance firmware or libraries copied into an application outside a package manager.
Sources: NIST National Vulnerability Database, CVE-2026-123007. See also our note on the Apache HTTP Server mod_http2 flaw.