Status, November 18, 2026. The nginx.org security advisories page now lists fixes in 1.30.7 (stable) and 1.31.9 (mainline) for the nginx2root set. Earlier in the day DugganUSA checked nginx.org and found no advisory and no fixed build, so if your distribution has not shipped a package yet, their advice still holds: turn HTTP/2 off. We are updating this post as packages and images land.
- The bug: CVE-2026-123007, memory corruption in nginx's HTTP/2 flow control, rated Major by nginx. It affects 1.9.5 through 1.31.8, which is every nginx that has had HTTP/2 since 2015. Xint, who named the set "nginx2root", reports code execution as the worker user with no backend and no extra modules needed. Wiz reports reproducing it against nginx serving plain static files.
- The rest of the set: an HTTP/2 use-after-free (CVE-2026-123633), an HTTP/3 use-after-free, a resolver use-after-free, a mail proxy use-after-free, and a TLS 1.3 request body infinite loop, all fixed in the same builds.
- Do now: find every nginx you run, including ingress controllers, appliances and container images. Upgrade to 1.30.7 or 1.31.9. Until you can, set
http2 off;and fully restart nginx, because a reload leaves the old workers serving open HTTP/2 connections. - What we did: we found seven internal nginx listeners with HTTP/2 on: six on 1.30.6 and our single sign-on proxy on 1.27.5. HTTP/2 was off on all seven within the hour, and we verified that each one now negotiates HTTP/1.1 only. The commands are below.
What Was Disclosed
On November 18 Xint published "NGINX2Root": five nginx remote code execution bugs, plus a separate Linux kernel bug they chained with one of them to get from the nginx worker to host root and out of a Docker container. The headline bug, CVE-2026-123007, is an integer overflow in HTTP/2 flow-control accounting that corrupts the heap. According to Xint, one researcher found it while preparing for a hacking competition, and an automated scan found the other four.
The nginx.org advisory list now carries the set with fixed versions:
| CVE | What | nginx severity | Vulnerable | Needs |
|---|---|---|---|---|
| CVE-2026-123007 | HTTP/2 flow control memory corruption | Major | 1.9.5 to 1.31.8 | HTTP/2 enabled |
| CVE-2026-123633 | HTTP/2 use-after-free | Medium | 1.19.7 to 1.31.8 | HTTP/2 over TLS with proxy_pass (per Xint) |
| CVE-2026-121810 | HTTP/3 use-after-free | Medium | 1.25.0 to 1.31.8 | HTTP/3 (QUIC) with an upstream |
| CVE-2026-123634 | Resolver use-after-free | Medium | 0.6.18 to 1.31.8 | A resolver and a proxy_pass built from the request |
| CVE-2026-113425 | Mail proxy use-after-free | Major | 0.5.33 to 1.31.8 | A mail block with auth_http |
| CVE-2026-118899 | Infinite loop reading a request body over TLS 1.3 | Major | 1.13.0 to 1.31.8 | TLS 1.3 |
Fixed in 1.30.7 and 1.31.9. The same advisory batch also fixes several minor and medium stream and PROXY protocol bugs.
How Bad Is It, Really
Reachability is broad: HTTP/2 is on by default in many distribution configs, cloud images and ingress controllers. Every TLS site that a modern browser reaches over HTTP/2 is a candidate. The bar for a working exploit is higher than the headline suggests. Rapid7's analysis, as reported by SecurityOnline, says the exploit needs a large amount of traffic (about 128 GB of HTTP/2 data) and depends on ASLR being off or bypassed. It also says the code runs as the nginx worker, not root. Root comes from the separate kernel bug in Xint's chain. A public proof of concept has been reported. DugganUSA had found no exploit code when they checked, and no exploitation in the wild has been reported so far.
Treat it as patch-now anyway. Exploits only get cheaper, and a worker shell on a reverse proxy sees every request and every upstream credential that passes through it.
Check Yourself in Five Minutes
Run these against every nginx you own, including the ones inside containers (docker exec <name> ...):
# version (vulnerable: 1.9.5 to 1.31.8, fixed: 1.30.7 / 1.31.9)
nginx -v
# is HTTP/2 or HTTP/3 on anywhere in the effective config?
nginx -T 2>/dev/null | grep -nE 'http2|http3|quic'
# from outside: what does the listener actually negotiate?
curl -sk --http2 -o /dev/null -w '%{http_version}\n' https://your.host:443/
# "2" means HTTP/2 is live; "1.1" means it is not
Check distribution changelogs as well as version strings: a backported fix may not change the version number.
The Interim Fix: Turn HTTP/2 Off
- nginx 1.25.1 and newer: replace
http2 on;withhttp2 off;in every server block. - Older nginx: remove
http2from everylistenline. - Do a full restart (
systemctl restart nginxor restart the container), not a reload. A reload leaves old workers running for connections that are already open. - If you do not use HTTP/3, remove
quicfromlistenlines and block UDP 443. - Use fixed upstream names instead of
proxy_pass http://$hoststyle variables, and keep any mail proxy listener off the internet. - gRPC through nginx needs HTTP/2, so those listeners have to be patched rather than downgraded.
- Behind a CDN, check whether the CDN talks HTTP/2 to your origin, and make sure the origin is not reachable directly. A directly reachable origin bypasses the CDN.
Browsers fall back to HTTP/1.1 without complaint. Most sites will not notice the change beyond slightly slower page loads with many assets.
What We Found on Our Own Network
We ran the checks above on our own infrastructure as soon as the advisory dropped. Seven internal nginx listeners had http2 on: six reverse proxies in front of internal dashboards and tools on nginx 1.30.6, and the proxy in front of our single sign-on on 1.27.5. The sign-on proxy matters most, because every login passes through it. None had HTTP/3 enabled, and none use a mail proxy or a request-built proxy_pass. Seven more nginx containers across our hosts ran vulnerable versions with HTTP/2 already off. They are not exposed to the HTTP/2 bugs, but they still get the upgrade.
We set http2 off; on all seven and restarted each container fully. We then checked from another machine with curl: every listener now negotiates HTTP/1.1. We made the same change in the configs those services deploy from, so a redeploy cannot bring HTTP/2 back. The images move to 1.30.7 as soon as the official image is published, and HTTP/2 stays off until then. Total time from reading the advisory to the last verified listener was under an hour, and nothing broke.
We are also adding this check to the external assessments we run for clients: the nginx version from the Server header plus whether the site negotiates HTTP/2.
Detection
These are frame-level bugs, so a WAF or access log does not see the payload. What you can watch:
- Worker crashes:
worker process ... exited on signal 11(or signal 6) in the nginx error log. A memory corruption attempt that misses usually kills a worker first. - Traffic shape: a single client pushing tens of gigabytes over one HTTP/2 connection, or an abnormal rate of WINDOW_UPDATE frames, fits the reported exploit and does not fit a normal browser.
- After the fact: unexpected child processes of the nginx worker user, outbound connections from the nginx host, and new files in writable paths the worker can reach.
Sources
- DugganUSA: nginx2root, HTTP/2 RCE, no patch yet, turn off HTTP/2 (Patrick Duggan). It has a clear mitigation checklist and a sober read of what was and was not confirmed when it was written.
- nginx.org security advisories
- Xint: NGINX2Root
- Wiz: Unauthenticated RCE in NGINX
- SecurityOnline: CVE-2026-123007 PoC report (Rapid7 analysis)
Questions about your exposure?
RedEye Security provides assessments for organizations that need to understand their real risk.
Talk to us