nginx2root Check
Find out whether your site's nginx is open to CVE-2026-123007 and CVE-2026-123633 — unauthenticated code execution over HTTP/2.
The check is passive: one ordinary request to the home page, as a browser makes, and nothing an attack would send. Check your own sites or with the owner's permission. What is and is not allowed
Want what we publish, weekly?
Drop your email — one letter a week with the guides and tools we published. No re-checks for this tool yet.
One-click unsubscribe in every email. We never share email addresses. By subscribing you agree to our privacy policy.
What nginx2root is and where it is fixed
nginx2root is five nginx vulnerabilities fixed on November 18, 2026. The main one, CVE-2026-123007, has been in the HTTP/2 code since 1.9.5 (2015): a flow-control window error lets an unauthenticated client corrupt worker memory and, in researchers' tests, run code with the worker's privileges. The second, CVE-2026-123633, is also in HTTP/2: present since 1.19.7 (2021), it triggers when nginx proxies to a backend and runs short of worker_connections. F5 scores all five 9.2 on CVSS 4.0. As of November 19, 2026 no attacks are reported and none is in CISA KEV, but proof-of-concept code is public.
| Branch | Fixed in |
|---|---|
| 1.31 (mainline) | 1.31.9 |
| 1.30 (stable) | 1.30.7 |
| NGINX Plus | 37.0.8.1 / 37.3.0.1 |
| 1.29 and older | no fix in its own branch — upgrade to 1.30.7 or take a distribution package with the fix |
Turning HTTP/2 off blocks CVE-2026-123007 and CVE-2026-123633, not the release's other three (HTTP/3, resolver, mail proxy). Only the upgrade closes them all.
How to check on your server
From outside, the version is visible only when the server names it (it does by default: server_tokens on). A distribution package may carry the fix under the old version number, and only the configuration tells whether HTTP/2 is on. These commands only read and change nothing:
# version and modules
nginx -V 2>&1 | head -1
# is HTTP/2 on: listen ... http2 or http2 on lines
sudo nginx -T 2>/dev/null | grep -n -E 'listen .*http2|^\s*http2 on'
# will the server agree to HTTP/2 — every browser asks the same way
openssl s_client -connect example.com:443 -servername example.com -alpn h2 </dev/null 2>/dev/null | grep ALPN
# Debian / Ubuntu: does the package carry the fix
apt changelog nginx 2>/dev/null | grep -m3 -E 'CVE-2026-12(3007|3633)'
How to close the hole without breaking anything
In this order:
-
Upgrade nginx
An upgrade within the branch (1.30.6 → 1.30.7) is a fix release: install it right away, keeping a copy of the configuration. nginx from the nginx.org repository upgrades to the fixed version. From a distribution repository, wait for the fixed package and compare its version with the distribution's advisory: the number in nginx -v does not change.
# Debian / Ubuntu sudo apt update && sudo apt install --only-upgrade nginx # RHEL / AlmaLinux / Rocky Linux sudo dnf upgrade nginx # Docker: replace a pinned tag such as nginx:1.29 with a fixed one docker pull nginx:1.30.7 docker compose up -d -
Restart nginx, not just reload it
reload does not load the new binary: the old processes keep running the vulnerable code. Test the configuration first — the command below will not restart nginx if it has an error. The restart takes seconds but drops open connections, so do it at a quiet hour.
sudo nginx -t && sudo systemctl restart nginx -
Check the result
On the server — nginx -v and the package version. From outside — run the check on this page again.
-
If you cannot upgrade today, turn HTTP/2 off
A stopgap from CIRCL and F5 guidance: it blocks CVE-2026-123007 and CVE-2026-123633. The site keeps working over HTTP/1.1 and visitors will not notice. Not for nginx proxying gRPC (grpc_pass needs HTTP/2). A reload is enough to apply it.
# nginx 1.25.1 and later: in every server block http2 off; # nginx 1.9.5–1.25.0: remove http2 from listen listen 443 ssl; # was: listen 443 ssl http2; sudo nginx -t && sudo systemctl reload nginx -
If the server was vulnerable — look for signs of an attack
An attack on CVE-2026-123007 takes a very large volume of inbound traffic on one connection. CIRCL suggests looking for crashed workers (worker process exited on signal 11 in error.log), long HTTP/2 connections with heavy inbound traffic and almost no requests in access.log, processes started by nginx, and unusual outbound connections. A clean access.log does not rule an attack out: HTTP/2 control frames are not logged there.
# crashed nginx workers sudo grep -E 'exited on signal (11|6)' /var/log/nginx/error.log | tail -20
What you may check
The tool makes one ordinary GET request to the home page (following up to three redirects) and reads only the headers the site sends to every visitor. It learns about HTTP/2 the way any browser does: by offering it in the TLS handshake. No control frames of the kind the attack uses, no brute force, no load — the check sends only an ordinary request.
- Check your own sites or sites whose owner has allowed the check.
- Do not run the published attack demos to “make sure”: on a live server this can crash nginx, and on someone else's system without permission it is unauthorized access to computer information (Article 272 of the Russian Criminal Code). Creating, using and spreading malicious software is prohibited by Article 273. Most countries have similar laws.
- Seen a vulnerable version on someone else's site? Tell the owner through the site's contacts or /.well-known/security.txt and go no further.
This is an explanation, not legal advice.