If you ran apt upgrade this morning because of the nginx HTTP/2 RCE (CVE-2026-123007, "nginx2root"), there's a good chance you patched nothing. On November 18 the Debian and Ubuntu packages were still listed as pending, the nginx:stable and nginx:mainline Docker tags still pulled vulnerable builds, and the community ingress-nginx controller that a lot of clusters still run will never get a fix, because the project was retired in March.
Xint calls it nginx2root, Wiz found it independently, and nginx.org lists it as "Memory corruption in HTTP/2 flow control", vulnerable from 1.9.5 through 1.31.8. That's every release since 2015. It needs no authentication and no backend app. A reachable TLS port with HTTP/2 on is enough.
So patch it, sure. But also count how many nginx binaries you're actually running, because I'd bet it's more than you think.
What the nginx HTTP/2 RCE actually is
Here's the mechanism from Xint's write-up. The client sets the initial stream window to 0 so nginx queues response DATA, grants that amount back with a connection-level WINDOW_UPDATE, then cancels the streams with RST_STREAM. Cleanup returns the same credit a second time without an upper-bound check, which pushes the connection send window past its 31-bit maximum. The unsigned window check then wraps, and later WINDOW_UPDATE frames keep getting accepted. That turns into an oversized memory copy, out-of-bounds writes and address disclosure. The final step overwrites a cleanup function pointer. Xint left the full procedure out of the post.
The address disclosure is why ASLR won't save you. Wiz says it confirmed an ASLR bypass on the official 1.31.7 and 1.31.8 builds. Xint says its exploit worked from inside the server, across a private network and over a public IP on AWS, GCP and Azure, and that it held up under concurrent normal traffic. Neither team has published the exploit. One Lobsters commenter said PoC scripts were already floating around anyway.
The nginx bug gets you code execution as the worker user. The "root" in nginx2root comes from a separate Linux kernel SCTP bug that Xint is mostly withholding until there's a kernel patch. In one of their setups that chain went from nginx inside a Docker container to a root shell on the host. So "it's only in a container" isn't the comfort it usually is.
A second HTTP/2 bug landed in the same release. CVE-2026-123633 is a use-after-free that shows up when nginx proxies HTTP/2 requests to an upstream while connection slots run low. Both are rated CVSS 4.0 9.2 by Wiz.
The same release also fixes CVE-2026-121810 (HTTP/3), CVE-2026-123634 (resolver) and CVE-2026-113425 (mail proxy auth_http). Turning off HTTP/2 does nothing for those three, so the mitigation further down only covers the two HTTP/2 bugs.
Fixed versions
| Product | Fixed in |
|---|---|
| nginx OSS stable | 1.30.7 |
| nginx OSS mainline | 1.31.9 |
| NGINX Plus (CVE-2026-123007) | 37.0.8.1, 37.3.0.1 |
| NGINX Plus (CVE-2026-123633) | 37.0.7.1, 37.2.0.1 |
| NGINX Plus R33 to R36 | No fix |
| F5 NGINX Ingress Controller | 5.8.0, HTTP/2 off by default (per Upwind) |
| Community ingress-nginx | No fix coming |
After the package upgrade, restart nginx (or do a live binary upgrade). Xint points out that a config reload doesn't replace the running executable, so systemctl reload nginx leaves the old, vulnerable binary serving traffic.
Check the version on your host (and don't trust it)
nginx -v nginx -V 2>&1 | grep -o -- "--with-http_v2_module" Distros backport fixes without changing the upstream version, so nginx -v is going to print 1.18.0 on Ubuntu 22.04 before and after the patch. Xint's example is exactly that package moving from 1.18.0-6ubuntu14 to 1.18.0-6ubuntu14.22 with the October 26 security updates while the version string stays put. Check the changelog instead (commands from Computing for Geeks):
# Debian/Ubuntu: 0 means the fix isn't in your package apt changelog nginx 2>/dev/null | grep -cE "CVE-2026-(123007|123633)" # RHEL, Rocky, Amazon Linux rpm -q --changelog nginx | grep -E "CVE-2026-(123007|123633)" Then check whether HTTP/2 is actually on. Grepping the config works, but asking the live server is more honest, since includes and defaults can lie to you:
sudo nginx -T 2>&1 | sed "s/#.*//" | grep -nE "listen[[:space:]].*http2|^[[:space:]]*http2[[:space:]]+on[[:space:]]*;" openssl s_client -connect 127.0.0.1:443 -alpn h2 </dev/null 2>&1 | grep -E "ALPN protocol:|No ALPN negotiated" If you see ALPN protocol: h2, you're exposed until you're patched.
There's an argument in the HN thread about whether Debian "disables HTTP/2 by default". Even if it does on a fresh install, I've never seen a production nginx box that kept the stock config. One Lobsters commenter checked a few of their servers and found HTTP/2 on all of them, several RHEL boxes and a Mastodon instance included.
Find the nginx copies apt doesn't know about
Containers are where the count goes up. Every one that bundles nginx ships its own binary, frozen at whatever version was in the base image on the day someone built it.
On a Docker host, start with what's running:
for c in $(docker ps -q); do printf '%s ' "$(docker inspect -f '{{.Name}} {{.Config.Image}}' "$c")" docker exec "$c" nginx -v 2>&1 || echo "no nginx binary" done Anything below 1.30.7 (or a 1.31.x below 1.31.9) needs a rebuild. Don't stop at images with "nginx" in the name either. Plenty of frontend images are FROM nginx:alpine with a different name on top, and plenty of "app" images run nginx in front of PHP-FPM.
Rebuilding doesn't help if the base tag hasn't moved yet. Computing for Geeks pulled nginx:stable on Nov 18 and got 1.30.6, and nginx:mainline gave them 1.31.8. Someone on HN said the official images lagged by a few hours. Pin nginx:1.31.9 or nginx:1.30.7 explicitly, rebuild, and run nginx -v inside the new image before you deploy it.
On Kubernetes, list every image that mentions nginx across the cluster:
kubectl get pods -A -o jsonpath='{range .items[*]}{.metadata.namespace}{"\t"}{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}' | grep -i nginx Then exec into anything suspicious and ask it directly with kubectl exec -n <ns> <pod> -- nginx -v.
The ingress-nginx problem
This is the one I'd fix before lunch. The community controller (kubernetes/ingress-nginx) was retired in March 2026. The Kubernetes announcement said there would be "no further releases, no bugfixes, and no updates to resolve any security vulnerabilities that may be discovered." Upwind says it ships with use-http2: true by default and bundles nginx 1.27.1, and that the Rancher and Chainguard forks were still unpatched when they wrote it. Chainguard has since merged a backport of the CVE-2026-123007 fix onto its bundled 1.27.1 (Nov 18), so if you're on that fork, update the image. Everyone on upstream kubernetes/ingress-nginx gets nothing.
So the cluster edge, the one component built to take internet traffic, is the copy with no patch on the way. Turn HTTP/2 off on it now:
kubectl -n ingress-nginx patch configmap ingress-nginx-controller \ --type merge -p '{"data":{"use-http2":"false"}}' Adjust the namespace and ConfigMap name if your Helm release is named differently. Then plan the migration you've been putting off since the retirement announcement in November 2025. Don't confuse it with F5's NGINX Ingress Controller (nginxinc/kubernetes-ingress), which is a different codebase and is fixed in 5.8.0. Upwind notes that F5's 2026-lts-r11 release still ships nginx 1.31.0 and is affected.
Can't upgrade today? Disable HTTP/2
Both Wiz and Xint give the same stopgap, and it blocks CVE-2026-123007 and CVE-2026-123633:
# nginx 1.25.1 and later server { listen 443 ssl; # remove "http2" from here http2 off; ... } # nginx 1.9.5 through 1.25.0: just drop http2 from the listen line listen 443 ssl; Two gotchas. First, a reload leaves old workers alive on their existing connections, and an attacker's long-lived HTTP/2 connection is exactly the kind that sticks around. Do a full restart, or at least confirm nothing is still shutting down:
ps -o pid,lstart,cmd -C nginx | grep "shutting down" Second, native gRPC needs HTTP/2, so this mitigation breaks it. One HN commenter couldn't use it because their service is mostly gRPC. If that's you, the mitigation isn't available and you have to upgrade.
Clients will just negotiate HTTP/1.1. For most sites behind a CDN, nobody will notice for a day or two.
Why apt upgrade isn't enough
Ubuntu 24.04 ships 1.24.0, Debian 12 ships 1.22.1, and both were pending as of Nov 18. Xint pulled Shodan on Nov 10 and found 11.94 million hosts with a visible version, 98.9% of them in the 1.9.5 through 1.31.8 range. Most of those owners will run apt upgrade, see nothing new, and assume they're fine.
Patch from nginx.org's repo if your distro is slow. Turn HTTP/2 off on anything you can't patch today. Then actually count your nginx binaries on the host, in containers, in sidecars and in the ingress. That count will take you longer than the patch.
How many copies of nginx did you find once you counted the ones inside containers and ingress controllers? And if you're still on community ingress-nginx after March, I'd like to hear what's kept you there.
Sources
- Wiz: Unauthenticated remote code execution in Nginx
- Xint: nginx2root
- nginx security advisories
- Upwind: where Kubernetes is exposed
- Computing for Geeks: check, disable HTTP/2, patch
- Kubernetes: Ingress NGINX retirement
- Discussion: Hacker News, Lobsters



