On 18 November 2026, F5 and the nginx project fixed two security flaws in the HTTP/2 code of nginx: CVE-2026-123007 and CVE-2026-123633. F5 rates both at CVSS v4.0 9.2. Both can lead to code execution on the server without any login. A working exploit is public. No attacks have been confirmed so far.
According to Netcraft, nginx is the web server behind more than 225 million websites. It usually sits right at the front: as the web server, as a reverse proxy in front of applications, as the entry point into Kubernetes. A flaw there affects everything behind it.
Four weeks ago we wrote here about php2shell. Back then, attacks were running before a patch existed. This time the patch came first. This article explains what the flaws are, who is affected and what you should do now.
Security researchers call the set NGINX2Root. F5 and nginx list the flaws under their CVE numbers.
What is behind the nginx flaws?#
Both bugs sit in the way nginx manages HTTP/2 connections. HTTP/2 bundles many requests over a single connection. How much data may flow at any moment is governed by a credit that the other side keeps topping up. The flaws attack exactly this bookkeeping.
CVE-2026-123007 is an arithmetic error in this flow control. When a request finishes, nginx credits the connection without checking the upper limit. A later check then drops below zero, and with unsigned numbers that turns into a huge value. In the end nginx copies more data than the intended memory can hold. An attacker triggers this with a very large number of WINDOW_UPDATE messages over a long lived connection. The faulty code dates back to nginx 1.9.5 from 2015, the first version with HTTP/2. Wiz showed that the attack also works on a server that only serves static files, with no application behind it and no extra modules.
CVE-2026-123633 is a use after free bug. An HTTP/2 connection without active requests is marked as reusable, and the mark stays in place when a new request starts. When free connections run short, nginx cleans up exactly this connection while a request is still using it. The preconditions are HTTP/2, nginx as a reverse proxy in front of an application, and load up to the limit of worker_connections. A plain web server for static content does not meet them.
The documented impact is reading server memory, crashing worker processes and code execution with the rights of the nginx worker process.
Which nginx versions are affected?#
The short answer: every nginx version with HTTP/2 enabled up to and including 1.31.8.
| Flaw | Affected | Fixed in | Precondition |
|---|---|---|---|
| CVE-2026-123007 | 1.9.5 to 1.31.8 | 1.30.7 and 1.31.9 | HTTP/2 enabled |
| CVE-2026-123633 | 1.19.7 to 1.31.8 | 1.30.7 and 1.31.9 | HTTP/2, reverse proxy, scarce connections |
A server that only speaks HTTP/1.1 is not affected by these two flaws. Whether HTTP/2 is on is stated in the configuration: as the http2 parameter of a listen line or, from nginx 1.25.1, as a separate http2 on; line.
For NGINX Plus, the commercial edition from F5, builds 37.0.8.1 and 37.3.0.1 fix both flaws. Releases R36 and older get no fix and must move to a 37 release.
Are you affected?
Enter your nginx version as shown by nginx -v, for example 1.26.3. The check runs entirely in your browser, nothing is transmitted.
Patched versions: 1.30.7 and 1.31.9#
nginx published two versions on 18 November 2026:
- nginx 1.30.7, the stable branch for production use
- nginx 1.31.9, the mainline branch with new features
Older branches such as 1.28 and earlier no longer get updates from nginx.org. If you run nginx from Debian, Ubuntu, Red Hat or another distribution, the fix comes through the distribution. The version number then stays old, for example 1.22.1 or 1.24.0, and the fix is still included. What counts is the distribution's security notice, not the number that nginx -v shows.
The same release fixes eight further flaws, among them in HTTP/3, the DNS resolver and the mail proxy. F5 rates three of them at 9.2 as well. That is another reason why updating is the only complete answer.
Where nginx hides#
The harder question is: where does nginx run without being on any list? These places are easy to miss with an advisory like this one:
- Containers. Many Docker images ship nginx, as the web server or as a proxy in front of the actual application. Updating the host changes nothing there. The image has to be pulled in a new version or rebuilt, and the container recreated.
- Kubernetes. The widely used ingress-nginx controller was retired in March 2026. The Kubernetes project no longer ships security updates for it. If you still run it, you are waiting for a fix that will not come from there, and you need a migration plan now.
- Devices with a web interface. Firewalls, NAS systems, routers and other appliances often bring their own nginx. The fix only arrives through a vendor update. Wiz expects widespread attacks to start with exactly such products, because they run the same build everywhere.
- Derivatives. OpenResty, Tengine and Angie are based on nginx. Check whether their projects have taken over the fix.
- Hosting. With web hosting and managed hosting, the update is the provider's job.
Timeline: discovery, report, patch#
- 2015. nginx 1.9.5 introduces HTTP/2. The faulty flow control code dates from this version.
- 11 October 2026. Security firm Xint finds the flaws and reports them to F5 on 16 October.
- 21 October. Wiz independently reports CVE-2026-123633. F5 replies that another team has already reported it.
- 16 November. Wiz also reports CVE-2026-123007. Again, someone was faster.
- 18 November. F5 and nginx publish updates and advisories, CERT-EU warns the European institutions.
- Since then. An exploit by Rapid7 is public on GitHub. As of today, 19 November, no attacks have been confirmed.
In total, five teams found the same bugs independently of each other: Wiz, Xint, Rapid7, Ikotas Labs and Metnir. Wiz says it used an AI assisted analysis tool to do so.
How dangerous is it?#
The situation differs from php2shell. There, attacks were running before a patch existed. With nginx it is the other way round. CERT-EU reports no exploitation, F5 says it has seen no evidence of it, and the US agency CISA does not list the flaws in its catalogue of actively exploited vulnerabilities.
There are reasons for this. An attack takes effort: it first has to leak memory addresses, prepare the memory layout and be tuned to the exact nginx version and build. Hence the sober rating under the older CVSS 3.1 scale of 8.1 with high attack complexity. A CDN or web application firewall that terminates HTTP/2 itself can also block parts of the attack.
Still, waiting is not an option. The exploit is public, and attackers do not need to crack every configuration. A product that runs thousands of times with the same build is enough. Then there is a detail that Xint demonstrated under the name NGINX2Root: through a further, still unfixed flaw in the SCTP module of the Linux kernel, an attacker gets from the nginx process to root on the server. There is neither a CVE number nor a fix for this kernel flaw yet.
So this time there is a window. Nobody knows how long it will stay open.
Disabling HTTP/2 only buys time#
If you cannot update right away, disable HTTP/2 for the time being:
- Remove the
http2parameter from alllistenlines.listen 443 ssl http2;becomeslisten 443 ssl;. - From nginx 1.25.1, also set
http2 off;in theserverblock, replacing any existinghttp2 on;. - Check with
nginx -tand reload. Verify from the outside, for example withcurl -sI --http2 https://your-domain.com/: the response must start with HTTP/1.1.
This only closes the two HTTP/2 flaws, not the other eight from the same release. Pages load slightly slower without HTTP/2. Xint, which found the flaws, also advises against relying on it as a blanket workaround.
One more step is worth taking regardless: block the sctp kernel module on servers that do not need it. That is almost every web server. A single line install sctp /bin/false in a file under /etc/modprobe.d/ is enough. It removes one link from the chain to root.
How we protect our clients#
A flaw in the web server cannot be prevented. What can be organised is the time until the patch is in. That is what a managed service is about, and this is how we work:
We know where nginx runs. Including the places where it is not on the label: in the container, in the ingress, in the appliance. Without this list the first question after any advisory cannot be answered, and it is always the same: does this affect us at all?
We assess each system. An nginx that accepts HTTP/2 directly from the internet is more urgent than one behind a proxy or on the internal network. We make this call for each client before we touch anything.
We inform actively, in both directions. Our clients hear from us when action is needed. They also hear from us when their environment is not affected. Otherwise managing directors read a headline and are left alone with the question.
We patch by urgency. Whatever is exposed to the internet comes first, outside the maintenance window if necessary. Internal systems follow in the regular window. We say which case applies beforehand, not afterwards.
We keep watching after the patch. Crashing nginx worker processes are a warning sign with these flaws. That belongs in monitoring, not in a one off exercise.
Patching nginx in six steps#
If you run your systems yourself, this is the order:
- Take stock. List every system with nginx, including containers, ingress controllers and appliances.
nginx -vshows the version, inside a containerdocker exec <container> nginx -v. - Check HTTP/2.
nginx -T | grep http2shows where HTTP/2 is enabled. Systems that offer it directly to the internet come first. - Apply the patch. The target is 1.30.7, 1.31.9 or later, or for distribution packages the version named in the distribution's security notice. Pull or rebuild container images.
- Restart, do not just reload. A reload rereads the configuration but does not replace the running binary. Afterwards, check that the running process has the new version. Recreate containers, restarting the old container is not enough.
- Ask your providers. Ask hosting providers, agencies and appliance vendors for their status and get the answer in writing. If you still run ingress-nginx, plan the migration now.
- Check the logs. Search the
error.logforworker process … exited on signal, especially since 18 November. Very long HTTP/2 connections with many small control messages and processes started by the nginx user are also suspicious.
Frequently asked questions#
Am I affected if nginx only serves a static website?
Yes, if HTTP/2 is enabled. CVE-2026-123007 can be triggered on a server that only serves files. CVE-2026-123633, by contrast, requires nginx to act as a reverse proxy in front of an application.
Does a CDN such as Cloudflare protect me?
Possibly in part. A CDN or web application firewall that terminates HTTP/2 itself and opens a new connection to your server can block parts of the attack. That depends on the implementation and only helps if your server is not also reachable directly from the internet. It does not replace the patch.
Is updating my Linux system enough?
If nginx comes from your distribution and the distribution has shipped the fix, yes, once nginx has been restarted. nginx inside containers, from your own builds or inside appliances is not covered by a system update.
Are Apache, Caddy or Traefik affected too?
No. The flaws are in the nginx code. Other web servers and proxies have their own HTTP/2 implementations. You should, however, check nginx derivatives such as OpenResty, Tengine or Angie.
Are the flaws being exploited already?
As of 19 November 2026, no attacks have been confirmed. A public exploit exists, though, so this can change quickly.
Conclusion#
Two critical flaws in the most widely used web server, a public exploit, but no confirmed attacks yet. That is the rare situation in which you can be ahead of the attacker. The patch is the only complete answer, disabling HTTP/2 only buys time.
The lesson is the same as with php2shell, only clearer. Five teams independently found the same bugs, some of them eleven years old, at least one of them with AI tools. What five teams find in parallel, a sixth one can find too, and that one does not report it. Patch management has become a question of response time, and that time is measured in days.
If you are weighing whether to carry this yourself, you will find the arguments in our comparison MSP or in-house IT admin.
Not sure where nginx runs in your estate? Talk to us. We will check, and we will also tell you when no action is needed.
Sources#
- nginx.org, security advisories and releases 1.30.7 and 1.31.9 of 18 November 2026
- F5, overview of the November 2026 security advisories (K000164058)
- CERT-EU, Security Advisory 2026-022
- Wiz Research, technical analysis and timeline
- Xint, NGINX2Root and the chain through the Linux kernel
- Tenable, frequently asked questions, exploit status and NGINX Plus
- Safe Security, assessment of the two HTTP/2 flaws and workarounds
- Kubernetes, retirement of ingress-nginx

