← Back to Blog

NGINX2Root (CVE-2026-123007): How to Check If Your nginx Is Vulnerable, Patch It and Turn HTTP/2 Off Safely

Published · by Leo Gashi, RS Computers

Nginx Security Debian

NGINX2Root is the name the research team Xint gave to five memory-safety bugs in nginx, led by CVE-2026-123007 in HTTP/2, that nginx fixed on 18 November 2026 in nginx 1.30.7 (stable) and 1.31.9 (mainline). Your nginx is vulnerable to CVE-2026-123007 if it is older than those versions (from 1.9.5 on) and has HTTP/2 switched on, unless your distribution has shipped a patched package, which on 19 November 2026 Debian 13 and Ubuntu had not.

This guide shows how to check your version and your HTTP/2 setting, how to patch on Debian, Ubuntu and Docker, and how to turn HTTP/2 off safely until a fixed package arrives. Every command below ran on a fresh Debian 13 test server with Debian's own nginx 1.26.3 on 19 November 2026; no exploit was run. If you are setting up a new server anyway, our KVM VPS and VDS plans in Amsterdam, Dublin and Prishtina come with Debian 13 and full root access, so you control when nginx is patched.

Key facts

Quick check and fix

  1. Version: nginx -v. Anything below 1.30.7 (or 1.31.x below 1.31.9) from nginx.org or Docker is vulnerable. Distribution packages keep the old number when patched, so also run zcat /usr/share/doc/nginx/changelog.Debian.gz | grep -c CVE-2026-123007; 0 means not patched.
  2. HTTP/2 on? sudo nginx -T 2>/dev/null | grep -nE '^\s*(http2|http3|mail|resolver|mirror|ssi)\b|listen[^;]*(http2|quic)'
  3. Patch: move to the nginx.org package (1.30.7) or a fixed Docker tag, then sudo systemctl restart nginx. A reload does not swap the binary.
  4. Cannot patch today: set http2 off; in every server block and remove http2 from every listen line, then sudo nginx -t && sudo systemctl reload nginx.
  5. Verify: curl -sk --http2 -o /dev/null -w '%{http_version}\n' https://your-domain/ must print 1.1, not 2.

What NGINX2Root is, in plain words

HTTP/2 tracks how much data each stream may still send, the "flow control window". nginx's changelog says an overflow of that window "might allow an attacker to cause a segmentation fault in a worker process, or might have potential other impact". The CVE record calls it a heap buffer overflow (CWE-787) that needs no login and no user action. So anyone who can open an HTTP/2 connection to your server could at least crash nginx workers, and in some conditions run code as the worker user, usually www-data.

Xint reported the bugs to F5 on 16 October 2026. The "2Root" part comes from a separate, still unpatched Linux kernel SCTP flaw that, in their research, turned worker access into root; Xint suggests blocking the sctp module on hosts that do not use SCTP until a kernel fix ships (Xint). Wiz also reported both HTTP/2 bugs to F5 and built exploits in its lab that it will not release (Wiz).

The five NGINX2Root CVEs

Versions and nginx ratings from nginx.org, scores and workarounds from F5's CVE records (123633, 123634, 121810, 113425), checked 19 November 2026. All five are fixed in 1.31.9 and 1.30.7.

CVEComponent, when exposednginx.org ratingCVSS 4.0 / 3.1Vulnerable versionsHTTP/2 off helps?
CVE-2026-123007HTTP/2 flow control (ngx_http_v2_module), any server with HTTP/2 onmajor9.2 / 8.11.9.5 to 1.31.8Yes
CVE-2026-123633HTTP/2 use-after-free, when proxying to a backend and worker_connections run shortmedium9.2 / 8.11.19.7 to 1.31.8Yes
CVE-2026-121810HTTP/3 (QUIC) use-after-free, when HTTP/3 is onmedium9.2 / 8.11.25.0 to 1.31.8No
CVE-2026-123634Resolver use-after-free, with a resolver, variables in the proxied server name and SSI or mirror subrequestsmedium9.2 / 8.10.6.18 to 1.31.8No
CVE-2026-113425Mail proxy (ngx_mail_proxy_module), only if you run nginx as a mail proxymajor9.2 / 8.10.5.33 to 1.31.8No

The same release fixes five more CVEs outside NGINX2Root, such as an endless loop reading request bodies over TLS 1.3 (CVE-2026-118899, "major"). Only the upgrade covers all ten.

Who is affected

Every stock nginx package on Debian 12 and 13 and on Ubuntu 16.04 to 26.10 is listed as vulnerable to CVE-2026-123007 by the two trackers. Exposure depends on configuration: Debian 13's default site serves port 80 without HTTP/2, so a plain install is not reachable through CVE-2026-123007 until you add an HTTPS site with http2 on;, which most HTTPS guides do. CVE-2026-121810 (HTTP/3) is "Not affected" on Ubuntu 24.04 and older, whose nginx has no QUIC code (Ubuntu); Debian 13's build has the HTTP/3 module. Software that bundles nginx (images, ingress controllers, panels) is fixed only when its maintainer ships 1.30.7, 1.31.9 or a backport.

Check your nginx version

On our Debian 13 server, as root:

nginx -v
dpkg -l nginx | tail -n 1
apt-cache policy nginx
nginx version: nginx/1.26.3
ii  nginx          1.26.3-3+deb13u10 amd64        small, powerful, scalable web/proxy server
nginx:
  Installed: 1.26.3-3+deb13u10
  Candidate: 1.26.3-3+deb13u10
  Version table:
 *** 1.26.3-3+deb13u10 500
        500 http://deb.debian.org/debian-security trixie-security/main amd64 Packages

1.26.3 is in every vulnerable range above. Debian and Ubuntu backport fixes without changing the upstream number, so the package changelog is the real test. Today it does not mention the new CVEs; the newest entry backports the October fixes from 1.30.6:

zcat /usr/share/doc/nginx/changelog.Debian.gz | grep -cE "CVE-2026-(123007|123633|123634|121810|113425)"
0

When a fixed Debian or Ubuntu package lands, this number goes above 0 after apt upgrade. To see which modules your build has:

nginx -V 2>&1 | tr ' ' '\n' | grep -E 'http_v2|http_v3'
--with-http_v2_module
--with-http_v3_module

For Docker, ask the running container, not the image you think you deployed. A container keeps the binary it started with until you recreate it:

docker ps --format '{{.Names}}  {{.Image}}  {{.Status}}'
docker exec web nginx -v
web  nginx:1.26.3  Up Less than a second
nginx version: nginx/1.26.3

Check whether HTTP/2 is on

nginx -T prints the full loaded configuration with all includes. This one grep finds HTTP/2, HTTP/3 and the features the other CVEs need. On our test server with two HTTPS sites it printed:

sudo nginx -T 2>/dev/null | grep -nE '^\s*(http2|http3|mail|resolver|mirror|ssi)\b|listen[^;]*(http2|quic)'
285:    http2 on;
292:    listen 8443 ssl http2;

Both forms enable HTTP/2. The http2 directive exists since nginx 1.25.1 (ngx_http_v2_module); the older listen ... http2 parameter still works and makes nginx -t print a "deprecated" warning. Then confirm from outside what clients really get. Run this against your own domain; here the lab server answered with HTTP/2 on both ports:

curl -sk --http2 -o /dev/null -w '%{http_version}\n' https://127.0.0.1/
echo | openssl s_client -connect 127.0.0.1:443 -alpn h2 2>/dev/null | grep -i alpn
2
ALPN protocol: h2

HTTP/2 is decided per site, by the host name the client sends (SNI). With two sites on port 443, one http2 off; and one http2 on;, our test gave 1.1 and 2, so test each domain with --resolve your-domain:443:SERVER-IP. For cleartext HTTP/2 on port 80, curl -s --http2-prior-knowledge -o /dev/null -w '%{http_version}\n' http://127.0.0.1/ printed 2 after we added http2 on; to the port 80 site and 0 with the stock config.

Patch nginx: Debian 13, Ubuntu, nginx.org and Docker

Where your nginx comes fromStatus on 19 November 2026What to do
Debian 13 (trixie) 1.26.3-3+deb13u10Vulnerable, no fixed package; sid fixed in 1.30.6-4Mitigate now, apt update && apt upgrade when a DSA appears, or switch to nginx.org
Debian 12 (bookworm) 1.22.1Vulnerable, no fixed packageSame as Debian 13
Ubuntu 16.04 to 26.10Vulnerable, no fixed version listedMitigate, watch the CVE page, or switch to nginx.org
Alpine edge and 3.24Fixed in 1.30.7-r0; 3.22 and 3.23 fixed in 1.28.3-r8Upgrade the package, restart (Alpine tracker)
nginx.org packages1.30.7 and 1.31.9 availableUpgrade, restart
Official Docker imagestable, stable-alpine, 1.30.7 run 1.30.7; latest, mainline, alpine run 1.31.9Pull and recreate the container

Tracker status changes daily; check the Debian page for CVE-2026-123007 before you decide.

Switch Debian 13 to the nginx.org stable package

These are nginx's own Debian instructions, as we ran them. Back up the config first, because this step replaces it:

cp -a /etc/nginx /root/etc-nginx-backup
apt install curl gnupg2 ca-certificates lsb-release debian-archive-keyring
curl -s https://nginx.org/keys/nginx_signing.key | gpg --dearmor | tee /usr/share/keyrings/nginx-archive-keyring.gpg >/dev/null
mkdir -m 700 -p /root/.gnupg
gpg --dry-run --quiet --no-keyring --import --import-options import-show /usr/share/keyrings/nginx-archive-keyring.gpg

On a fresh server the check failed with gpg: Fatal: /root/.gnupg: directory does not exist! until we created that folder. The output must contain the fingerprint nginx.org publishes, 573BFD6B3D8FBC641079A6ABABF5BD827BD9BF62; ours listed it together with two newer nginx signing keys from 2024. Then add the repository, pin it, and install:

echo "deb [signed-by=/usr/share/keyrings/nginx-archive-keyring.gpg] https://nginx.org/packages/debian `lsb_release -cs` nginx" | tee /etc/apt/sources.list.d/nginx.list
echo -e "Package: *\nPin: origin nginx.org\nPin: release o=nginx\nPin-Priority: 900\n" | tee /etc/apt/preferences.d/99nginx
apt update
apt-cache policy nginx | head -n 3
apt install nginx
nginx:
  Installed: 1.26.3-3+deb13u10
  Candidate: 1.30.7-1~trixie

Three things happened on our server that the instructions do not warn about:

  1. Debian's nginx-common package was removed and nginx.org's own nginx.conf was installed over Debian's. The new file includes only conf.d/*.conf, not sites-enabled/*, so all our sites silently disappeared: nginx -T no longer contained them.
  2. The new file runs workers as user nginx; instead of www-data. PHP-FPM's socket allows only www-data, which gives a 502 with (13: Permission denied), as shown in our nginx 502 Bad Gateway guide.
  3. nginx was left stopped: systemctl is-active nginx printed inactive.

The fix was to put the backed-up Debian file back, test, and start:

cp /root/etc-nginx-backup/nginx.conf /etc/nginx/nginx.conf
nginx -t && systemctl enable --now nginx
nginx -v
ps -o user=,pid=,cmd= -C nginx | head -n 2
nginx: configuration file /etc/nginx/nginx.conf test is successful
nginx version: nginx/1.30.7
root        1812 nginx: master process /usr/sbin/nginx -c /etc/nginx/nginx.conf
www-data    1813 nginx: worker process

Our sites answered again with the same config. Debian's extra module packages (libnginx-mod-*) are built for Debian's nginx, so list them with dpkg -l | grep nginx before you switch and plan for each one; our test server had none. nginx.org's page has the same steps for Ubuntu 22.04, 24.04 and 26.04.

Docker

Pull a fixed tag and recreate the container; restarting the old container is not enough. In our test the replacement started from nginx:stable reported:

docker rm -f web
docker run -d --name web -p 8080:80 nginx:stable
docker exec web nginx -v
nginx version: nginx/1.30.7

Pin an exact tag such as nginx:1.30.7 in Compose files if you want to know what runs. Images that bundle nginx inside something else need a rebuild by their maintainer.

Temporary mitigation: turn HTTP/2 off

F5's CVE record lists the workaround for CVE-2026-123007 as "None": the patch is the fix. Red Hat says the issue "can be mitigated by disabling HTTP/2 support", by removing the http2 parameter from listen lines and setting http2 off; on nginx 1.25 and later (Red Hat), and Xint and Wiz both advise turning HTTP/2 off on affected servers until the patch is installed. Browsers then use HTTP/1.1; services that need HTTP/2, such as native gRPC, will stop working, per Xint.

Before, as found by the grep above:

server {
    listen 443 ssl;
    http2 on;
    ...
}
server {
    listen 8443 ssl http2;
    ...
}

First we only added http2 off; to both blocks. Port 443 dropped to HTTP/1.1, but port 8443 kept HTTP/2, because the http2 parameter on listen wins over the directive:

1.1 200
2 200

After, with the parameter removed too:

server {
    listen 443 ssl;
    http2 off;
    ...
}
server {
    listen 8443 ssl;
    http2 off;
    ...
}
sudo nginx -t && sudo systemctl reload nginx
curl -sk --http2 -o /dev/null -w '%{http_version} %{http_code}\n' https://127.0.0.1:8443/
echo | openssl s_client -connect 127.0.0.1:443 -alpn h2 2>/dev/null | grep -i alpn
nginx: configuration file /etc/nginx/nginx.conf test is successful
1.1 200
No ALPN negotiated

On Ubuntu 24.04 and older (nginx 1.24 and below) the http2 directive does not exist; there you only remove http2 from listen. For the other three CVEs, F5's records give "Restrict listeners to trusted networks" for the mail proxy and "Avoid using SSI or mirror directives and avoid configuring a proxied server name with variables" for the resolver bug; for HTTP/3, Xint advises removing the quic listeners until you patch. After patching, put http2 on; back; on 1.30.7 our sites negotiated 2 again.

Restart after patching, not only reload

A reload re-reads the configuration and starts new workers, but the master process keeps running the old binary; replacing the executable needs a restart or nginx's binary upgrade procedure (Controlling nginx). You can see it in the process start times. After systemctl reload nginx our master kept PID 549 and its start time, and only the worker was new:

ps -o pid,lstart,cmd -C nginx | head -n 3
    PID                  STARTED CMD
    549 Thu Nov 19 02:26:29 2026 nginx: master process /usr/sbin/nginx -g daemon on; master_process on;
    724 Thu Nov 19 02:27:36 2026 nginx: worker process

After systemctl restart nginx, the same command showed a new master:

    PID                  STARTED CMD
    741 Thu Nov 19 02:27:37 2026 nginx: master process /usr/sbin/nginx -g daemon on; master_process on;
    742 Thu Nov 19 02:27:37 2026 nginx: worker process

If the master's start time is older than your upgrade, run sudo systemctl restart nginx. Open connections are cut, so do it at a quiet hour, but do it today.

Cloudflare or another proxy in front

A proxy helps less than people hope. Cloudflare's documentation says "HTTP/2 connection to the origin is enabled by default" and that Cloudflare uses it with any origin that announces HTTP/2 through ALPN (Cloudflare: HTTP/2 to Origin). So your nginx still speaks HTTP/2, only to Cloudflare. Cloudflare also warns that anyone who finds your origin IP can bypass it, and recommends blocking traffic that does not come from Cloudflare's IP ranges (Cloudflare IP addresses). We found no Cloudflare statement about these CVEs on 19 November 2026.

A proxy does not cover direct hits on your server's IP, ports it does not proxy (a mail proxy, for example) or hostnames that are not proxied. A rate limit like those in our nginx 429 guide is no fix either; the CVE-2026-123633 record speaks of a single "specially crafted request". Patch the origin, then lock ports 80 and 443 to the proxy's ranges.

Check your logs for anything unusual

We cannot tell you what an attack looks like on the wire, and we will not publish it. What you can check is the effect the CVE records describe: a crashed worker. When a worker dies, the master writes one line. We produced it by sending a worker the crash signal by hand on the test server:

grep -hE 'exited on signal|worker_connections are not enough' /var/log/nginx/error.log*
2026/11/19 02:27:39 [alert] 741#741: worker process 742 exited on signal 11

Signal 11 is a segmentation fault. A healthy server prints nothing here; a run of these lines on an unpatched HTTP/2 server is worth a closer look, and "worker_connections are not enough" is the condition two of the bugs need. Then check that the worker user runs nothing but nginx and owns no new files in temp folders:

ps -u www-data -o pid,lstart,cmd
find /tmp /var/tmp /dev/shm -user www-data 2>/dev/null; echo "find done"

On our server the first listed only nginx: worker process lines and the second only find done. Anything else, such as a shell or a downloaded binary, means you should treat the server as compromised: take it off the network, rebuild from a clean image and rotate keys. Our Debian 13 security checklist covers the hardening steps for the rebuild.

Frequently asked questions

What is NGINX2Root?

NGINX2Root is the name Xint gave to five nginx vulnerabilities fixed on 18 November 2026: CVE-2026-123007 and CVE-2026-123633 in HTTP/2, CVE-2026-121810 in HTTP/3, CVE-2026-123634 in the resolver and CVE-2026-113425 in the mail proxy. F5 scores each 9.2 in CVSS 4.0. The fixed versions are nginx 1.30.7 and 1.31.9.

Is my nginx vulnerable to CVE-2026-123007?

It is if your nginx is between 1.9.5 and 1.31.8 without a distribution backport and HTTP/2 is enabled with http2 on; or listen ... http2. Check with nginx -v, the package changelog for the CVE ID, and nginx -T | grep http2.

Which nginx version fixes NGINX2Root?

nginx 1.30.7 (stable) and 1.31.9 (mainline), both released 18 November 2026. Distribution packages may fix it by backport under an older version number, so check the changelog for the CVE IDs rather than the version alone.

Is Debian 13 nginx patched for CVE-2026-123007?

Not on 19 November 2026. Debian's tracker lists trixie's 1.26.3-3+deb13u10 as vulnerable and only sid as fixed. Use the HTTP/2 mitigation or the nginx.org 1.30.7 package until a Debian security update appears.

Does disabling HTTP/2 fully protect nginx?

No. It blocks CVE-2026-123007 and CVE-2026-123633 only. HTTP/3, resolver and mail proxy bugs are not affected by it, and the same release fixes five more CVEs. Remove http2 from listen lines as well, or HTTP/2 stays on.

Is there a public exploit for NGINX2Root?

Rapid7 says it has developed a proof of concept for CVE-2026-123007, and Xint published demonstrations without final steps. None of the sources we checked reported attacks in the wild on 19 November 2026; watch CISA's KEV catalog for changes.

Need a hand?

Finding every nginx in containers, panels and old VMs is usually the slow part. We do custom setups and systems built for you, including hardening an existing server or moving sites off an old one, quoted per job: write to us on Telegram or at info@rscomputers-ks.com. If the old box is too far gone to trust, start clean on a fresh Debian 13 VPS or VDS with NVMe in Amsterdam, Dublin or Prishtina, install nginx 1.30.7 from nginx.org on day one, and keep the free weekly backups for the next time something like this lands.

← All articles

Chat on Telegram