Direct answer: If you run NGINX, inventory every instance now and update NGINX Open Source to 1.30.7 or 1.31.9, NGINX Plus 37.0 to 37.0.8.1, or NGINX Plus 37.1/37.2 to 37.3.0.1. Then restart the service and confirm the running binary/package is fixed. The November 18 release fixes ten security issues, including the HTTP/2 memory-corruption flaw CVE-2026-123007 and four other bugs grouped under the name “nginx2root.”[1][2][3]
Do not rely on a version banner alone. Linux distributions often backport security fixes while retaining an older upstream version number. Compare the complete installed package build with your distribution’s security advisory.[1]
Five-minute response checklist
- Find NGINX on hosts, containers, ingress images, appliances and managed platforms.
- Record the installed package and running binary versions.
- Dump the active configuration and check whether HTTP/2, HTTP/3, dynamic resolver proxying or the mail proxy is enabled.
- Update through the package source or vendor that supplied NGINX.
- Restart NGINX; a configuration reload alone does not replace the running binary.[1]
- Verify the package build, process start time, configuration test and site health.
- Search logs for worker crashes and the indicators listed below.
- If Cloudflare proxies the site, confirm the origin cannot be reached directly; proxying reduces exposure to the HTTP/2 and HTTP/3 paths but does not replace patching.[1]
Read-only NGINX exposure check
Run these commands with the permissions normally used for server administration. They inspect the host and configuration; they do not send an exploit payload or change the server.
printf '%s\n' '=== binary ==='
command -v nginx || true
nginx -v 2>&1 || true
nginx -V 2>&1 || true
printf '%s\n' '=== package ==='
if command -v dpkg-query >/dev/null; then
dpkg-query -W -f='${Package} ${Version}\n' 'nginx*' 2>/dev/null || true
fi
if command -v rpm >/dev/null; then
rpm -qa | grep -Ei '^nginx' || true
fi
printf '%s\n' '=== running processes ==='
ps -eo pid,lstart,cmd | grep '[n]ginx' || true
printf '%s\n' '=== active configuration risk markers ==='
nginx -T 2>&1 | grep -En \
'listen .*http2|http2[[:space:]]+on|listen .*quic|http3[[:space:]]+on|resolver[[:space:]]|proxy_pass.*\$|mirror[[:space:]]|ssi[[:space:]]+on|mail[[:space:]]*\{|auth_http' \
|| true
Save the full nginx -T output privately for review because the filtered output only identifies likely risk markers. Do not paste the complete configuration into a public ticket: it can contain internal hostnames, paths and other sensitive deployment details.
Container and Kubernetes checks
A host can appear clean while a container still ships a vulnerable NGINX build. Inventory running images and inspect the binary inside the actual workload:
# Docker: inventory images used by running containers
docker ps --format '{{.ID}}\t{{.Image}}\t{{.Names}}'
# Run for each NGINX container ID or name
docker exec CONTAINER nginx -v 2>&1
# Kubernetes: list images across namespaces
kubectl get pods -A -o jsonpath='{range .items[*]}{.metadata.namespace}{"\t"}{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}'
Do not assume an ingress controller, hosting panel or appliance follows the same upstream version scheme. Check that product vendor’s advisory and fixed image digest.
Which nginx2root flaws matter to your configuration?
The five named nginx2root bugs require different features. Cloudflare’s analysis maps them as follows:[1]
- CVE-2026-123007: HTTP/2 enabled. This is the broadest exposure and the most urgent path.
- CVE-2026-123633: HTTP/2 proxying to a backend while the worker is low on available connections.
- CVE-2026-121810: HTTP/3/QUIC proxying to a backend.
- CVE-2026-123634: the
resolverdirective, a variable inproxy_pass, and a subrequest such asmirroror SSI. - CVE-2026-113425: the NGINX mail proxy with
proxy onandauth_http.
The official NGINX advisory page lists 1.30.7 and 1.31.9 as the first non-vulnerable Open Source releases for those five issues. It classifies CVE-2026-123007 and the mail-proxy flaw as “major,” while the other three are listed as “medium.”[2]
NGINX 1.31.9 also fixes CVE-2026-118899, an infinite loop while reading a request body over TLS 1.3, plus security issues involving PROXY protocol v2, alias with try_files, stream SNI validation and ssl_preread parsing.[2][4]
Why CVE-2026-123007 is urgent
CVE-2026-123007 is an HTTP/2 flow-control accounting error that can cause memory corruption. Rapid7 demonstrated unauthenticated code execution against a deliberately vulnerable NGINX 1.31.8 Docker target with HTTP/2 and TLS enabled, and assigned the flaw CVSS v4 9.2.[5]
The demonstrated route required roughly 128 GB of HTTP/2 traffic and depended on a specific tested Linux/AMD64 memory-copy implementation, so practical exploitation has constraints. Those constraints are not a reason to defer patching: they describe one known route, not a guarantee that every future technique will need the same conditions.[5]
As of Cloudflare’s November 19 analysis, the five nginx2root CVEs were not in CISA’s Known Exploited Vulnerabilities catalog and Cloudflare had not seen reports of exploitation in the wild. Rapid7 had published a local proof of concept, while fake exploit repositories were already circulating.[1]
Safe patch sequence
Ubuntu or Debian packages
First confirm which repository owns the package:
apt-cache policy nginx nginx-core nginx-full
Then use your normal approved update process:
sudo apt update
sudo apt install --only-upgrade nginx nginx-common
sudo nginx -t
sudo systemctl restart nginx
If the distribution package still displays an older upstream number, compare the complete package revision shown by dpkg-query or apt-cache policy with the distribution advisory. Do not replace a distribution package with an upstream binary during an incident unless you have tested the change and understand its module and configuration compatibility.
RHEL, Rocky, AlmaLinux, Fedora or compatible systems
sudo dnf --refresh upgrade nginx
sudo nginx -t
sudo systemctl restart nginx
For older systems using yum, follow that distribution’s supported command and advisory. Again, the package release suffix may carry the backport even when nginx -v still shows an older upstream number.
Official NGINX packages
If the host uses the official NGINX repository, move to 1.30.7 on the stable branch or 1.31.9 on the mainline branch. NGINX says 1.31.9 is available as source and binary packages for major Linux distributions.[2][4]
NGINX Plus
Rapid7’s remediation table lists these fixed builds:[5]
- NGINX Plus 37.0.0–37.0.7.1: update to 37.0.8.1.
- NGINX Plus 37.1.0–37.2.0.1: update to 37.3.0.1.
- NGINX Plus R34–R36: no fixed build in those branches; move to a fixed 37.x release.
Use the F5-supported upgrade path and test configuration compatibility before production rollout.
Temporary mitigations when you cannot patch immediately
These are short-term exposure reductions, not substitutes for the update:[1]
- Disable HTTP/2 if it is not required. On NGINX 1.25.1 and later, use
http2 off;; on older configurations, removehttp2from thelistendirective. - Remove QUIC listeners and disable HTTP/3 if it is not required.
- Replace variable-built
proxy_passdestinations with fixed upstream groups where possible. - Restrict NGINX mail listeners to trusted networks.
- Restart after configuration changes so existing connections close.
Test any mitigation in a staging environment first. Disabling HTTP/2 or HTTP/3 changes transport behavior and may affect performance, monitoring and client expectations.
Does Cloudflare protect an NGINX origin?
For orange-clouded HTTP hostnames, Cloudflare terminates client HTTP/2 and HTTP/3 connections at its edge and creates a separate connection to the origin. That prevents attacker-controlled protocol frames from reaching the origin through the normal proxied route for the HTTP/2 and HTTP/3 bugs.[1]
The protection has limits:
- It does not help if attackers can connect directly to the origin IP.
- It does not fully address the resolver/subrequest flaw.
- It does not protect NGINX mail proxy traffic.
- It does not cover DNS-only records, internal services or Spectrum traffic that passes the relevant protocol through.
- A WAF rule is not the fix because these are protocol-level memory-safety issues.[1]
Restrict origin access with Cloudflare Tunnel, a network firewall limited to Cloudflare IP ranges, or Authenticated Origin Pulls, then patch anyway.[1]
Post-update verification
Run this immediately after the maintenance window:
sudo nginx -t
nginx -v 2>&1
ps -eo pid,lstart,cmd | grep '[n]ginx'
systemctl --no-pager --full status nginx
curl -fsS -o /dev/null -w 'HTTPS %{http_code} %{time_total}s\n' https://YOUR-DOMAIN/
Confirm all of the following:
nginx -tsucceeds.- The full package build matches a fixed vendor advisory.
- Worker process start times are later than the package update.
- Public and internal health checks pass.
- Containers were recreated from fixed images rather than only restarting an old image.
- Load balancers, ingress controllers, appliances and standby nodes were included.
Detection and log review
Cloudflare lists several useful signals: unusually large inbound traffic on one long-lived HTTP/2 connection, worker process ... exited on signal 11, worker_connections are not enough, reusing connections, and could not cancel ... resolving on an unpatched resolver configuration. Unexpected programs launched by the NGINX worker account are a stronger sign of successful compromise.[1]
Example defensive searches:
# Adjust paths for your distribution and logging setup
sudo journalctl -u nginx --since '2026-11-18' | \
grep -E 'signal 11|worker_connections are not enough|could not cancel .* resolving|http2 flood detected|violated connection flow control'
sudo grep -RE \
'signal 11|worker_connections are not enough|could not cancel .* resolving|http2 flood detected|violated connection flow control' \
/var/log/nginx 2>/dev/null
Absence of these messages does not prove a server is safe or uncompromised. If you find unexplained worker crashes, child processes, configuration changes or direct-origin traffic, preserve logs and volatile evidence, isolate the affected system according to your incident-response plan, rotate secrets available to the NGINX worker, and rebuild from known-good artifacts.
FAQ
Is every NGINX server vulnerable to all five nginx2root bugs?
No. Each flaw requires a particular feature or configuration, but HTTP/2 is common and exposes the highest-priority CVE-2026-123007 path. Inventory configuration rather than assuming a reverse proxy, static server or container is unaffected.[1][2]
Is NGINX 1.30.7 fixed?
Yes. The official advisory lists 1.30.7 and 1.31.9 as non-vulnerable for the five nginx2root CVEs.[2]
Is a graceful reload enough after updating the package?
No. Cloudflare’s remediation guidance says to restart NGINX or complete a binary upgrade; a configuration reload does not replace the running binary.[1]
Does Cloudflare WAF block nginx2root?
Cloudflare says its protection for the HTTP/2 and HTTP/3 paths comes from terminating those protocols at the edge, not from a WAF signature. Directly reachable origins and other affected configurations still need remediation.[1]
Is CVE-2026-123007 being exploited in the wild?
Cloudflare reported no known in-the-wild exploitation as of November 19, and the CVE was not then in CISA’s Known Exploited Vulnerabilities catalog. A public technical proof of concept exists, so defenders should patch without waiting for confirmed attacks.[1][5]
Sources
- nginx2root: five new nginx vulnerabilities
- nginx security advisories
- nginx change log
- NGINX 1.31.9 protocol and security fixes
- Rapid7: CVE-2026-123007 nginx HTTP/2 vulnerability