Skip to Content

Critical NGINX vulnerability (nginx2root, HTTP/2) - troubleshooting and diagnostics

November 18, 2026 by
Critical NGINX vulnerability (nginx2root, HTTP/2) - troubleshooting and diagnostics
Administrator

What is this vulnerability

On November 18, 2026, critical vulnerabilities were disclosed in the HTTP/2 subsystem of the NGINX web server that allow for remote execution of arbitrary code without authentication (pre-auth RCE) — the attacker does not need a password or an account. Researchers from Xint combined their findings into a chain called nginx2root and demonstrated complete server takeover, including escaping from the Docker container; attacks were reproduced on AWS, Google Cloud, Microsoft Azure, and on physical machines.

Key records:

  • CVE-2026-123007 (nicknamed "nginx2root" / "NGINX2Shell") — a flaw in the HTTP/2 subsystem present in the open-source NGINX since version 1.9.5 (September 2015), meaning a "decade-old" hole. According to researchers, code execution is possible even on a server that serves only static files, without a backend application or third-party modules.
  • CVE-2026-123633 — a use-after-free type error (the worker process reuses already freed memory). It concerns scenarios where NGINX accepts HTTP/2 connections from clients and proxies requests to the backend, while the worker process experiences a shortage of free connections.

Both vulnerabilities have a critical score 9.2 out of 10 on the CVSS 4.0 scale. За оцінкою Wiz Research, NGINX використовується більш ніж на 225 млн сайтів у світі; станом на 10 листопада 2026 року вразливими були близько 98,9% хостів з розпізнаваними версіями NGINX. Підтверджених випадків експлуатації «в дикій природі» на момент розкриття не зафіксовано.

Which versions and configurations are at risk

The main condition for operation is enabled support HTTP/2 for incoming (client) connections. If HTTP/2 is disabled on the listeners, the server is not vulnerable, even on an old version. Important: this specifically refers to HTTP/2 between the client and NGINX, not the protocol to the backend.

Conditions for each vulnerability:

  • CVE-2026-123007 — it is sufficient to have HTTP/2 enabled; even a server serving only static content is vulnerable. For remote code execution, F5 states that ASLR must be disabled or bypassed.
  • CVE-2026-123633 — additionally, a reverse proxy scheme with a backend and a workload under load (shortage of free connections) is required.

One vulnerable instance of NGINX puts all applications behind it at risk.

Affected and fixed versions

CVE

Product

Affected versions

First fixed

CVE-2026-123007

NGINX Open Source

1.9.5–1.30.6 and 1.31.0–1.31.8

1.30.7 (stable), 1.31.9 (mainline)

CVE-2026-123633

NGINX Open Source

1.19.7–1.30.6 and 1.31.0–1.31.8

1.30.7 (stable), 1.31.9 (mainline)

CVE-2026-123007

NGINX Plus 37.x

37.0.0–37.0.7.1, 37.1.0–37.2.0.1

37.0.8.1, 37.3.0.1

CVE-2026-123633

NGINX Plus 37.x

37.0.0–37.0.6.1, 37.1.0–37.2.0

37.0.7.1, 37.2.0.1

Both

NGINX Plus R34–R36

R34–R36

switch to build 37.x with fixes for both

All open-source releases from 1.9.5 to 1.31.8 inclusive are affected.

How to fix

The main remedy is to update NGINX to the fixed version. A temporary alternative, if an immediate update is not possible, is to disable HTTP/2 for incoming connections.

Step 1. Check the version and whether HTTP/2 is enabled

nginx -v
nginx -T 2>/dev/null | grep -iE 'listen.*http2|http2 on'

If the version is within 1.9.5–1.31.8 and HTTP/2 is enabled — the server requires action.

Step 2. Update NGINX

Install a build that addresses both CVE: 1.30.7 (stable branch) or 1.31.9 (main branch). Note: a build newer than the October security release may still not contain the November fixes — check against the table above.

For Debian/Ubuntu from the official repository nginx.org:

apt update
apt install --only-upgrade nginx
nginx -v

If the fixed version is not yet in the distribution repositories, add the official nginx.org repo or wait for the distribution update, and until then — apply the temporary HTTP/2 disable (step 4).

Step 3. Restart the service, not just reload

Reloading the configuration (reload) does not replace the running binary that is already executing. After updating the package, you need to restart the process (or rebuild the container):

systemctl restart nginx
nginx -v
ss -tlnp | grep nginx

In Docker: rebuild the image with the updated nginx and recreate the containers (docker compose pull && docker compose up -d).

Step 4. Temporarily disable HTTP/2 (if updating is not possible yet)

In the configs, find the listen directives with http2 and remove them. In the new syntax — http2 on; replace with http2 off; (or comment it out). In the old one — remove the word http2 from the line listen 443 ssl http2;, leaving listen 443 ssl;.

nginx -t
systemctl reload nginx

Check from the client side that HTTP/2 is no longer offered:

curl -I --http2 https://your-domain/ 2>&1 | grep -i 'HTTP/'

The response should be HTTP/1.1, not HTTP/2. Reintroduce HTTP/2 only after updating.

Priority on many servers

First — servers that look to the Internet (website fronts, panels, APIs, reverse proxies), then — internal proxies and static sites.

How to diagnose if it has been exploited

This is a memory corruption vulnerability via HTTP/2, so it does not leave an obvious trace in access logs, as is the case with web attacks in URLs. At the time of disclosure, researchers had not released ready-made indicators of compromise. Therefore, the check is based on indirect signs: abnormal behavior of nginx processes and traces of the attacker’s persistence in the system.

1. Crashes and restarts of nginx worker processes

Exploitation of memory corruption often first crashes the worker processes. Look for sudden crashes or SIGSEGV signals:

journalctl -u nginx --since "7 days ago" | grep -iE 'signal|segfault|worker process|exited|core'
grep -iE 'signal|exited on signal|worker process' /var/log/nginx/error.log
dmesg -T | grep -iE 'nginx|segfault'

Frequent crashes/restarts of workers without an obvious reason — a reason for detailed analysis.

2. External processes on behalf of the user nginx

If the code was executed, the attacker will operate in the context of the worker (usually the user www-data or nginx). Check that no foreign processes are running from this user:

ps -u www-data -f
ps aux | grep -iE 'nginx' | grep -ivE 'nginx: (master|worker|cache)'
ls -la /proc/$(pgrep -f 'nginx: worker' | head -1)/exe

Concerning: shells (sh, bash), interpreters (python, perl), nc/ncat, curl/wget to unfamiliar addresses, execution from /tmp, /dev/shm, /var/tmp.

3. Outgoing (reverse/outbound) connections from the server

After RCE, a typical step is a reverse shell or payload download. Check for unusual outgoing connections, especially initiated by nginx processes:

ss -tnp | grep -iE 'nginx|www-data'
ss -tnp state established

Cross-check remote IPs against the list of expected ones (your backends, updates). Everything else is suspicious.

4. New/modified files and attachments

find /tmp /dev/shm /var/tmp /var/www -type f -mtime -14 -printf '%Tc %p\n' 2>/dev/null | sort
ls -la /var/www /usr/share/nginx/html
crontab -l -u www-data 2>/dev/null; ls -la /etc/cron.* /var/spool/cron/crontabs 2>/dev/null
grep -RiE 'wget|curl|base64|eval|bash -i|/dev/tcp' /var/www 2>/dev/null

Look for web shells (unknown .php/.sh in the root of the site), new cron jobs, added SSH keys in /root/.ssh/authorized_keys and home directories, modified systemd units.

5. For Docker environments

Due to the demonstrated exit beyond the container, also check the host: new processes from root, modified files on the host, attempts to access the Docker socket (/var/run/docker.sock), mounted volumes.

docker ps -a
docker diff <nginx-container>

docker diff will show all changes in the container's file system relative to the image — foreign files are immediately visible here.

6. Centralized analysis (Prometheus/Grafana)

If you are collecting metrics: sharp spikes in process restarts, memory usage by nginx workers, unusual outgoing connections or CPU on front servers in the window before the update — additional reasons for investigation.

If signs of compromise are found

Isolate the server from the network, take an image/snapshot for forensics, change all secrets and keys accessible from this host (SSH, API tokens, database passwords), and restore from a clean source, rather than 'fixing' on-site. For vulnerable but not compromised servers, the steps from the previous section are sufficient.


Internet in Ukraine: how businesses can avoid being left without access to servers