← Back to Blog
Hot TopicOn This Site

nginx HTTP/2 RCE, CVE-2026-123633 and CVE-2026-123007 (November 2026): Is Your Server Affected, and How to Patch It Step by Step

On 18 November 2026 nginx shipped 1.30.7 and 1.31.9 to fix ten vulnerabilities, led by two HTTP/2 memory bugs (CVE-2026-123633 and CVE-2026-123007) that researchers rate 9.2 and say can lead to unauthenticated remote code execution. A public proof of concept exists for one of them. Here is who is exposed, a two-minute check, the fixed versions for Ubuntu, Debian, Docker and Kubernetes, the http2 off stopgap, and what to do if you run the archived ingress-nginx controller.

·10 min read

On 18 November 2026 the nginx project released 1.30.7 (stable) and 1.31.9 (mainline), fixing ten vulnerabilities. The two that matter most are in HTTP/2: CVE-2026-123007, a memory corruption bug in HTTP/2 flow control, and CVE-2026-123633, a use-after-free. Several security firms rate both 9.2 on CVSS 4.0 and say they can lead to remote code execution without logging in, and Rapid7 has published a proof of concept for CVE-2026-123007. If your nginx serves HTTP/2 (most HTTPS sites do), update to 1.30.7, 1.31.9 or your distribution's patched package today, then restart nginx. If you can't patch yet, turn HTTP/2 off as a stopgap. If you run the community Kubernetes ingress-nginx controller, there is no fix coming, and that needs its own plan.

I'm Kunal Vohra, a security researcher and cybersecurity advisor with an M.Tech in Cybersecurity from NIT Kurukshetra, and a technical co-founder six times over. nginx sits in front of a huge share of startup stacks, including this site, which runs Next.js behind nginx. That's why I'd treat this one as a same-day job rather than a next-sprint ticket.

What was fixed, and how sure we are

The nginx.org security advisories page lists all ten CVEs fixed in this release, each marked "Not vulnerable: 1.31.9+, 1.30.7+". nginx labels its issues major, medium or minor rather than giving CVSS scores, and that's where the first disagreement comes in.

CVE-2026-123007, memory corruption in HTTP/2 flow control. Affects nginx 1.9.5 to 1.31.8. nginx calls it "major". Wiz scores it 9.2 and says it needs HTTP/2 over TLS on a 64-bit system, but nothing else: a plain static site with no backend is in scope. The catch for attackers is volume. Upwind estimates it takes around 60 GB sent on a single connection to crash a worker and around 128 GB to reach code execution. That's a lot of traffic, but it's traffic, not a precondition you control.

CVE-2026-123633, use-after-free in HTTP/2. Affects nginx 1.19.7 to 1.31.8. nginx calls it "medium"; Wiz, Orca, Sweet Security and others rate it 9.2 Critical. According to Upwind's analysis, it needs HTTP/2 on, nginx opening an upstream connection (any proxy_pass, fastcgi_pass, grpc_pass and so on, or a resolver lookup), and the attacker holding most of a worker's connection slots, roughly 960 of the default 1,024. In plain terms: nginx as a reverse proxy in front of an app, which is the most common startup setup there is.

Eight more in the same release. A use-after-free in HTTP/3 (CVE-2026-121810), a use-after-free in the resolver (CVE-2026-123634), a use-after-free in the mail proxy (CVE-2026-113425), an infinite loop reading request bodies over TLS 1.3 (CVE-2026-118899), an SNI bypass with client certificates in the stream module (CVE-2026-120967), a NULL pointer dereference with PROXY protocol v2 (CVE-2026-120993, mainline 1.31.4 to 1.31.8 only), a buffer overflow with alias plus try_files (CVE-2026-125283), and insufficient validation in stream ssl_preread (CVE-2026-123640). Most need specific configurations, but turning HTTP/2 off does nothing for them. Only the upgrade fixes the whole set.

Where sources disagree, I'll say so rather than pick a side:

  • Severity. nginx rates CVE-2026-123633 "medium"; the research firms rate it 9.2. My reading is that nginx weighs how hard the conditions are, while the firms weigh what happens if they're met.
  • ASLR. Wiz says address space layout randomization doesn't stop these bugs and that it demonstrated a bypass. Orca and nginx2shell.com say F5 treats code execution as possible only when ASLR is disabled or bypassed.
  • Exploitation. As of 18 to 19 November, nobody reported exploitation in the wild. Wiz built working exploits and is holding them back. Rapid7's proof of concept for CVE-2026-123007 is public (Upwind describes it as tuned for specific Debian 13 builds). With a public PoC and an internet full of nginx, I'd assume attempts start soon.

Five teams found these bugs independently: Wiz Research, Ikotas Labs, Metnir, Rapid7 and Xint. Wiz says it used its AI harness to find both. When five teams land on the same bug in the same weeks, the window between a fix and an exploit gets very short, the same pattern I wrote about in Google's exploit_agent report.

When five research teams find the same bug in the same month, assume attackers are the sixth.

Is this you? Who should care

You should act today if any of these are true:

  • You run nginx yourself on a VPS or cloud VM in front of Next.js, Node, Django, Rails, Laravel or anything else, with HTTPS. Certbot, Ansible roles and most tutorials produce exactly this setup.
  • You run nginx in Docker, including the official nginx image or images built on it.
  • You run Kubernetes with an nginx-based ingress: the community kubernetes/ingress-nginx, F5's NGINX Ingress Controller, or NGINX Gateway Fabric.
  • You serve gRPC through nginx. grpc_pass needs HTTP/2, so you can't use the stopgap; you must patch.
  • You use OpenResty, Tengine, Angie or freenginx. Upwind says their code still contains the vulnerable path; I found no fixed versions for them in what I read, so check with each project.
  • You pay for NGINX Plus or F5 products built on it.

You're probably not directly exposed if your app sits only behind a managed load balancer or CDN and no nginx is reachable from the internet. "Probably" is doing work there: if your origin server accepts connections from anywhere, an attacker skips the CDN and talks to nginx directly.

The two-minute check

Run this on every server or container that runs nginx:

# Version (distro builds keep an old upstream number, see below)
nginx -v

# Is HTTP/2 on, and is nginx proxying to a backend?
sudo nginx -T 2>/dev/null | grep -nE 'http2|grpc_pass|proxy_pass|fastcgi_pass|resolver|quic|http3'

# Package version on Ubuntu or Debian
apt-cache policy nginx

# From your laptop: what does the site actually negotiate?
curl -so /dev/null -w '%{http_version}\n' --http2 https://example.com/

If curl prints 2, HTTP/2 is on. If the nginx -T output shows http2 on a listen line or http2 on;, plus any *_pass directive, you're in the scope of both headline bugs.

One trap: Ubuntu and Debian backport fixes without changing the upstream version. A patched Ubuntu 24.04 server still says nginx 1.24.0. Judge distro packages by the full package version against the vendor notice, not by nginx -v. I covered the same trap in the PHP2Shell guide, and it catches people every time.

How to fix it, step by step

Steps 1 to 3 are the fix. Step 4 is for anything you can't patch today. Steps 5 and 6 are the ones teams skip.

Step 1: Install a fixed nginx

Ubuntu. Canonical's USN-9159-1 covers all supported releases. Fixed package versions reported on 19 November include 1.24.0-2ubuntu7.20 for 24.04, 1.18.0-6ubuntu14.23 for 22.04 and 1.28.3-2ubuntu1.13 for 26.04, with ESM releases also covered.

sudo apt update
sudo apt install --only-upgrade 'nginx*'
apt-cache policy nginx   # compare with the USN

Debian. DSA-6642-1 fixes Debian 13 (trixie) in 1.26.3-3+deb13u11. As of 19 November, Debian 12 (bookworm) had no fixed package. If you're on bookworm, use the stopgap in step 4 now and watch the Debian tracker, or move to the nginx.org packages.

nginx.org packages (the official nginx apt or yum repository): upgrade to 1.30.7 (stable) or 1.31.9 (mainline) with the same apt or dnf command.

RHEL, Rocky, Alma, Amazon Linux. As of 19 November, Red Hat had not shipped fixes for RHEL 8, 9 or 10. AlmaLinux 10 and the default AlmaLinux 9 stream had fixes; Amazon Linux 2023 ships 1.30.7-1.amzn2023.0.1 (ALAS2023-2026-3494). Where a fix exists:

sudo dnf upgrade --refresh 'nginx*'

Where it doesn't, apply step 4 and keep checking.

Alpine. 3.22, 3.23, 3.24 and edge are fixed; 3.21 was not as of 19 November.

Docker. A running container never patches itself. Pin a fixed tag and rebuild every image that uses nginx:

FROM nginx:1.30.7
docker compose build --pull
docker compose up -d

NGINX Plus. Wiz, Orca and Sweet Security report fixes in 37.0.7.1 and 37.2.0.1 for CVE-2026-123633 and 37.0.8.1 and 37.3.0.1 for CVE-2026-123007. Upwind says R33 to R36 have no fix. Check F5's advisories (K000163928 and K000163926) for your exact build.

Step 2: Restart nginx and confirm the new binary is running

Updating the package doesn't replace worker processes that are already running. Most write-ups say do a full restart, not a reload:

sudo nginx -t && sudo systemctl restart nginx
ps -o pid,lstart,cmd -C nginx   # every process should start after the upgrade

Step 3: Kubernetes ingress

This is where the news is worst, and it all rests on one source (Upwind), so verify against your own charts:

  • Community kubernetes/ingress-nginx: archived, ships nginx 1.27.1 with HTTP/2 on by default, and every release is affected with no fix planned. The Rancher and Chainguard forks were also unpatched. Turn HTTP/2 off now and plan a migration.
  • F5 NGINX Ingress Controller: fixed in 5.8.0, where HTTP/2 is off by default. The 2026-lts-r11 release still ships nginx 1.31.0 and is affected.
  • NGINX Gateway Fabric: fixed in 2.9.0. HTTP/2 is on by default there.

Find what you run, then switch HTTP/2 off on the community controller (adjust the namespace and ConfigMap name to your install):

kubectl get pods -A -o jsonpath='{range .items[*]}{.spec.containers[*].image}{"\n"}{end}' | grep -i nginx | sort -u

kubectl -n ingress-nginx patch configmap ingress-nginx-controller \
  --type merge -p '{"data":{"use-http2":"false"}}'

If you were already putting off the move from ingress-nginx to Gateway API or another controller, this is the push. An unmaintained edge proxy is a liability that only grows.

Step 4: If you can't patch today, turn HTTP/2 off

On nginx 1.25.1 and later, add http2 off; to each affected server block (or once in the http block) and remove http2 from every listen line. On older versions, only the listen change works:

server {
    listen 443 ssl;          # was: listen 443 ssl http2;
    listen [::]:443 ssl;
    http2 off;               # nginx 1.25.1+
    server_name example.com;
    # ...
}

Then sudo nginx -t && sudo systemctl restart nginx, and run the curl check again; it should print 1.1.

Two honest caveats. First, sources disagree on how complete this is: ma.ttias.be and Upwind say it blocks both HTTP/2 bugs, Sweet Security says it reduces exposure without fixing anything. Second, it does nothing for the other eight CVEs. It's a bridge to the upgrade, not a substitute. Your users won't notice much; HTTP/1.1 over TLS is slower for asset-heavy pages, not broken.

Step 5: Look at the kernel too

Xint reportedly chained CVE-2026-123007 with a separate, still unpatched Linux SCTP kernel bug to go from an nginx container to root on the host. The nginx patch does not fix the kernel bug. If you don't use SCTP (almost no web startup does), Upwind suggests blocking the module:

echo "install sctp /bin/false" | sudo tee /etc/modprobe.d/disable-sctp.conf
lsmod | grep sctp   # if it's already loaded, plan a reboot

This is the same lesson as the OpenAI Linux root exploits: a container is not a security boundary you should bet the company on.

Step 6: Check for signs of trouble

None of the sources I read publish specific indicators of compromise, so this is a general check rather than a signature:

  • Worker crashes in /var/log/nginx/error.log, such as lines with exited on signal 11 or signal 6, especially in bursts.
  • Single connections moving enormous amounts of data inbound, or many long-held idle connections filling your worker slots.
  • Unexpected processes, cron jobs, SSH keys or outbound connections from the nginx user.
  • If anything looks wrong, rotate secrets nginx could read (TLS private keys, basic auth files, any credentials in config) and rebuild from a known-good image.

What it means for startups, and what I'd do as a CTO

Most early teams treat nginx as plumbing: a config someone copied in year one, a Certbot run, never touched again. That's exactly why a bug like this hurts. Nobody owns it, nobody knows which servers run it, and the Kubernetes controller everyone installed from a tutorial is now archived.

If I were your CTO this morning, I'd do four things:

  1. Inventory every nginx in 30 minutes. VMs, containers, ingress controllers, the forgotten staging box. If you can't list them, that's finding number one.
  2. Patch what can be patched today; switch HTTP/2 off on the rest (Debian 12, RHEL, community ingress-nginx) and put a date on the real fix.
  3. Lock your origins. If a CDN or load balancer sits in front, make the origin accept traffic only from it. That shrinks who can reach nginx for every future bug, not just this one.
  4. Decide on ingress-nginx now, not when the next CVE lands. Moving off it is a week of careful work; doing it during an incident is much worse.

If you also run PHP-FPM behind nginx, do both jobs in one window: PHP shipped 8.5.13, 8.4.28, 8.3.37 and 8.2.36 on 19 November, and one of those fixes (CVE-2026-129289) involves nginx's fastcgi_keep_conn on. The routine is the same one I laid out for the Node.js November release and Java4Shell, and my startup security checklist covers making it a habit. If you'd rather have someone own this kind of morning for you, that's a big part of what I do as a technical advisor or fractional CTO.

Frequently asked questions

Is my nginx affected by CVE-2026-123633 or CVE-2026-123007? If it's older than 1.30.7 (stable) or 1.31.9 (mainline), or an unpatched distro package, and it serves HTTP/2, yes for CVE-2026-123007. If it also proxies to a backend, CVE-2026-123633 applies too. Check with nginx -T | grep http2 and compare your package version with your distro's advisory.

Does disabling HTTP/2 fix the nginx vulnerability? It's a stopgap. Some researchers say it blocks both HTTP/2 bugs; others say it only reduces exposure. It does not fix the other eight CVEs in the release. Upgrade as soon as you can.

Is the nginx HTTP/2 vulnerability being exploited? As of 18 to 19 November 2026, no exploitation in the wild was reported. Rapid7 has a public proof of concept for CVE-2026-123007 and Wiz has private working exploits, so treat it as urgent.

Is there a fix for Kubernetes ingress-nginx? According to Upwind, no: the community project is archived and every release is affected. Set use-http2: "false" in its ConfigMap now and plan a move. F5's NGINX Ingress Controller 5.8.0 and NGINX Gateway Fabric 2.9.0 are fixed.

Do I need to restart nginx or is a reload enough? Restart. A reload keeps the old binary as master, and most sources say to confirm every old process is gone after upgrading.

Am I safe behind Cloudflare or an AWS load balancer? Only if your origin is unreachable except through it. If anyone can connect to your server's IP directly, they can talk to nginx without going through the CDN.

Not sure how many nginx instances are hiding in your stack, or what to do about ingress-nginx? Book a 30-minute call and bring your server list and Helm charts.

Sources

Written By

Kunal Vohra

Kunal Vohra

Technical Co-Founder & Fractional CTO

I've co-founded 6+ startups across India, the UAE, and the US, spanning AI, Web3, fintech, and cybersecurity. I write about the technical and strategic decisions that determine whether a startup thrives or stalls.

Comments

Loading comments…

Enjoyed this article?

More writing on AI, Web3, and building startups.