News Security & Bugs

Critical nginx HTTP/2 Flaws (CVE-2026-123007, CVE-2026-123633): Check and Patch Your Server

nginx 1.30.7 and 1.31.9 fix two critical HTTP/2 flaws that can let an attacker run code without logging in. How to check your VPS, which package versions are fixed, and a stopgap.

nginx 1.30.7 (stable) and 1.31.9 (mainline), released on 18 November 2026, fix ten security issues. Two of them are in the HTTP/2 code and can let an attacker with no account run code on the server: CVE-2026-123007 and CVE-2026-123633. F5, the company behind nginx, scores both 9.2 (critical) on CVSS 4.0.

The flaws are in nginx itself, so a site with clean application code is still exposed if it serves HTTP/2 from an unpatched build. Hosting providers have started to email customers about it; DigitalOcean sent its notice on 21 November. If you run your own VPS, the update is yours to install. This article shows how to check a server, which package versions carry the fix, and what to do if you cannot update today.

We compiled this from the vendor and distribution advisories listed at the end. We did not test the flaws or run these commands on a test server for this article.

What you should do now

  1. Install the nginx security update from wherever your nginx comes from: your distribution, the nginx.org repository, a container image or a control panel.
  2. Restart nginx. A reload keeps the old program in memory.
  3. Confirm the installed package contains the fix. On Debian, Ubuntu and RHEL-family systems the number shown by nginx -v does not change.
  4. If you cannot update today, switch HTTP/2 off until you can.

The two HTTP/2 flaws

CVE-2026-123007CVE-2026-123633
What it isFlow-control accounting error that corrupts memoryUse-after-free on an HTTP/2 connection
Vulnerable nginx versions1.9.5 to 1.31.81.19.7 to 1.31.8
NeedsHTTP/2 enabledHTTP/2 enabled, requests proxied to a backend, and a worker running low on worker_connections
F5 score9.2 critical (CVSS 4.0), 8.1 high (CVSS 3.1)9.2 critical (CVSS 4.0), 8.1 high (CVSS 3.1)
nginx.org ratingMajorMedium
F5 advisoryK000163926K000163928

CVE-2026-123007 is the broader of the two. The faulty code arrived with the first HTTP/2 release in 2015, so every upstream version from 1.9.5 to 1.31.8 has it. Wiz Research reports that it achieved code execution on a server that only served static files, with no backend application and no third-party modules.

The ratings differ between sources, and both views are worth knowing. F5's advisories say an attacker can crash worker processes, and can execute code where address space layout randomization (ASLR) is disabled or can be bypassed. The research teams at Wiz and Xint both say they bypassed ASLR and ran commands with the privileges of the nginx worker. The nginx.org advisory page uses its own scale and lists the two issues as major and medium. Treat both as urgent.

Several teams found the bugs independently. F5 credits Xint, Wiz Research, Rapid7, Ikotas Labs, Metnir and Palo Alto Networks Unit 42 across the two CVE records.

Check your server

1. Is HTTP/2 switched on? Print the full configuration and look for the two ways of enabling it.

sudo nginx -T 2>/dev/null | grep -nE 'listen .*http2|http2 +on'

Output such as listen 443 ssl http2; or http2 on; means HTTP/2 is enabled. No output means it is not, and these two flaws cannot be reached on that server. Most HTTPS configurations written by control panels, Certbot guides and tutorials do enable it.

2. Does the outside world reach HTTP/2? Run this from another machine and replace the host name.

openssl s_client -connect example.com:443 -servername example.com -alpn h2 </dev/null 2>/dev/null | grep ALPN

ALPN protocol: h2 means the server accepts HTTP/2. If a CDN or proxy such as Cloudflare sits in front, this test reaches the proxy and not your nginx. Your server is still exposed if its own IP address accepts connections from anywhere.

3. Which package is installed?

# Debian and Ubuntu
apt policy nginx

# AlmaLinux, Rocky Linux, RHEL
rpm -q nginx

Compare the full package version with the table below. Do not rely on nginx -v: distributions patch their own build and keep the upstream number.

Fixed versions

Where nginx comes fromFixed in
nginx.org source or packages, stable1.30.7
nginx.org source or packages, mainline1.31.9
Ubuntu 26.101.30.4-7ubuntu1.2
Ubuntu 26.04 LTS1.28.3-2ubuntu1.13
Ubuntu 24.04 LTS1.24.0-2ubuntu7.20
Ubuntu 22.04 LTS1.18.0-6ubuntu14.23
Ubuntu 20.04, 18.04, 16.04Fixed in ESM packages, which need Ubuntu Pro
Debian 13 (trixie)1.26.3-3+deb13u11 (DSA-6642-1)
Debian 12 (bookworm)1.22.1-9+deb12u11 (DLA-4918-1)
RHEL 10nginx-1.26.3-16.el10_3.3 (RHSA-2026:100744, 20 November)
RHEL 9 and 8Updated nginx module streams (RHSA-2026:100745, 100746, 100747, 20 November)
NGINX Plus37.0.7.1 or 37.2.0.1 for CVE-2026-123633; 37.0.8.1 or 37.3.0.1 for CVE-2026-123007. Take the higher one to cover both.

Versions are as listed by nginx.org, Ubuntu, the Debian security tracker and Red Hat on 22 November 2026. We did not check the AlmaLinux and Rocky Linux errata; both rebuild Red Hat's packages, usually within days, so look up the CVE number in their errata pages before assuming the fix has arrived.

The nginx 1.28 branch and older are legacy branches upstream and did not receive a new release. If you compiled one of those from source, move to 1.30.7.

Update and restart

# Debian and Ubuntu
sudo apt update && sudo apt install --only-upgrade nginx
sudo systemctl restart nginx

# AlmaLinux, Rocky Linux, RHEL
sudo dnf upgrade nginx
sudo systemctl restart nginx

Use restart, not reload. A reload re-reads the configuration but the old binary stays in memory. A restart drops open connections for a moment.

Then confirm the fix is in the package you now run:

# Debian and Ubuntu
apt changelog nginx 2>/dev/null | grep -m1 CVE-2026-123007

# AlmaLinux, Rocky Linux, RHEL
rpm -q --changelog nginx | grep -m1 CVE-2026-123007

A matching line means the package changelog records the fix. Check that the service start time is after the update with systemctl status nginx.

Docker and Kubernetes. Pull or rebuild from an image that contains 1.30.7 or 1.31.9, or a distribution image with the patched package, and recreate the containers. A tag such as nginx:stable on a host that pulled it last month is still the old build. Ingress controllers and bundles built on nginx, such as OpenResty, ship their own builds: follow their advisories.

Control panels and managed hosting. The panel or the host ships nginx. Apply the panel's update and read its announcement. On shared and managed hosting, the provider patches the server.

If you cannot update today

Switching HTTP/2 off blocks both flaws. Sites keep working over HTTP/1.1, a little slower on pages with many files. Services that need HTTP/2, such as native gRPC, will stop working.

# nginx 1.25.1 and later: in each affected server block
listen 443 ssl;
http2 off;

# nginx 1.9.5 to 1.25.0: remove the http2 parameter
listen 443 ssl;

Remove http2 from every listen line that shares the same address and port, then apply and test:

sudo nginx -t && sudo systemctl reload nginx

Run the openssl check again. It should no longer print ALPN protocol: h2. Old worker processes can keep serving existing HTTP/2 connections after a reload, so check that they have exited, or restart nginx. Switch HTTP/2 back on after you have updated.

The other eight fixes

The same releases fix eight more issues. Three of them are also rated critical by F5 and are not blocked by switching HTTP/2 off:

  • CVE-2026-121810: use-after-free in HTTP/3, when nginx proxies to a backend.
  • CVE-2026-123634: use-after-free in the DNS resolver, in configurations that use the resolver directive.
  • CVE-2026-113425: use-after-free in the mail proxy module.

The rest cover a request-body loop over TLS, a crash with PROXY protocol version 2, a buffer overflow with alias and try_files, and two issues in the stream module. The update fixes all ten, which is why the HTTP/2 workaround is only a stopgap.

Xint, one of the teams that reported the bugs, also describes an unpatched Linux kernel flaw that it chained with the nginx bug to become root and to leave a Docker container. It is withholding the details until a kernel fix exists. Install kernel updates when your distribution ships them.

Has it been exploited?

When the CVE records were published on 18 November, the CISA assessment attached to them listed no known exploitation. The technical write-ups are detailed, so expect that to change. Check the current status before you decide the update can wait.

An attack on the flow-control flaw may leave nothing in the access log, according to Xint, because the HTTP/2 control frames it uses are not logged as requests. Two things are worth a look on a server that was exposed:

sudo grep -Ei 'worker process.*exited on signal|segfault' /var/log/nginx/error.log

Repeated worker crashes can be failed attempts, though crashes have other causes too. A shell or any other program started by an nginx worker is a much stronger sign. If you find one, follow our guide on what to do when a VPS is hacked.

Keeping up next time

This is the third critical web-stack fix in a week, after WordPress 7.1.5 and the PHP-FPM flaw. Automatic security updates install patched packages without waiting for you, but they do not always restart the service, so check after a release like this one. The VPS hardening checklist covers the rest.

Sources

The F5 advisory pages did not load as text when we prepared this article; F5's wording is taken from the CVE records it published.

Share story:

Continue reading other guides on How VPS.