What was released on November 18, 2026, and why it is serious
The disclosure was coordinated: on November 18, 2026, nginx.org released version 1.30.7 on the stable branch and 1.31.9 on the mainline. The 1.30.7 change log lists, along with the HTTP/2 flaws, a defect involving try_files and alias (CVE-2026-125283) and several other fixes.
On severity, two sources complement each other and should be quoted as they stand. The nginx.org security advisory rates CVE-2026-123007 as "major". Wiz, which documented the whole set, scores CVE-2026-123007 and CVE-2026-123633 at 9.2 on CVSS 4.0, which is "critical", and describes unauthenticated remote code execution.
Three nuances matter for the decision. The attack requires HTTP/2 to be enabled. CVE-2026-123007, an integer underflow on the HTTP/2 send window, is exploitable with the default configuration as soon as HTTP/2 is on: Wiz demonstrated it on a 64-bit system serving static files. Exploitation remains difficult, since it needs a memory leak and then heap grooming, but ASLR does not protect you. Finally, we have not established that a public exploit exists; that is no reason to wait.
The HTTP/2 flaws and their neighbors: affected and fixed versions
Scroll the table
| Flaw or product | Affected versions | Fixed in |
|---|---|---|
| CVE-2026-123007 — memory corruption, HTTP/2 flow control (major) | 1.9.5 to 1.31.8 | 1.30.7 and 1.31.9 |
| CVE-2026-123633 — HTTP/2 use-after-free (medium at nginx.org, 9.2 at Wiz) | 1.19.7 to 1.31.8 | 1.30.7 and 1.31.9 |
| CVE-2026-123634 — resolver (medium) | 0.6.18 to 1.31.8 | 1.30.7 and 1.31.9 |
| CVE-2026-121810 — HTTP/3 (medium) | 1.25.0 to 1.31.8 | 1.30.7 and 1.31.9 |
| CVE-2026-113425 — mail proxy module (major) | 0.5.33 to 1.31.8 | 1.30.7 and 1.31.9 |
| CVE-2026-125283 — try_files and alias | Range to read in the nginx.org advisory | 1.30.7 (CHANGES-1.30) |
| NGINX Plus — CVE-2026-123007 | See the vendor advisory | 37.0.8.1 and 37.3.0.1 (according to Wiz) |
Am I exposed? Four checks per server
A server is exposed when two conditions add up: HTTP/2 is enabled on at least one reachable server, and the running nginx process does not include the fix. The second condition is the one people miss: the binary installed on disk and the binary executing in memory are not necessarily the same.
A word on Cloudflare and other CDNs or WAFs. According to Wiz, a service that terminates HTTP/2 in front of your server may partially mitigate the attack. But if the origin IP is still directly reachable, the origin remains exposed: do not treat this mitigation as protection, and fix the server.
Check one server in ten minutes
Note the version and where it comes from
nginx -vgives the version of the binary on disk.nginx -V 2>&1 | head -3adds the build options. The origin decides the procedure:apt-cache policy nginxon Debian or Ubuntu,rpm -qi nginx | grep -E 'Version|Release|Vendor'on RHEL and derivatives. An old version number is not conclusive: see distribution packages below.Find out which version is really running
ps -o lstart= -p $(cat /run/nginx.pid)gives the start time of the master process. Compare it with the date of the last package update:grep ' upgrade nginx' /var/log/dpkg.log | tail -3on Debian or Ubuntu,rpm -q --last nginxon RHEL.sudo ls -l /proc/$(cat /run/nginx.pid)/exeshows(deleted)if the master runs a binary that has since been replaced. Adjust the PID file path if needed.Look for HTTP/2 in the effective configuration
sudo nginx -T 2>/dev/null | grep -n 'listen.*http2\|http2 on'nginx -Ttests then prints the full configuration, included files too. Before nginx 1.25.1, HTTP/2 is enabled only through thehttp2parameter oflisten. Since 1.25.1, thehttp2 on;directive exists too, athttporserverlevel. No line: HTTP/2 is not declared.Check from the outside, name by name
curl -sI --http2 https://your-domain.com | head -1HTTP/2 200means HTTP/2 is negotiated;HTTP/1.1 200means it is not. Repeat the test for each name served: sites sharing the same port can have different settings.Look at the neighboring modules
nginx -V 2>&1 | grep -o 'with-mail\|http_v3'
Without--with-mail, the mail proxy module (CVE-2026-113425) is not compiled. Without--with-http_v3_module, CVE-2026-121810 does not concern you. Write down the result next to the two previous checks.Classify the server
Exposed: HTTP/2 enabled and an unfixed version, or an old binary still in memory. To plan within the week: HTTP/2 off but a version in the affected range. Not concerned: fixed package and process restarted after the update. Keep this table per client: it is the answer to give when a client asks whether they were affected.
Fixed versions by distribution: nginx -v is not enough
This is the most common objection: nginx -v prints 1.22.1 or 1.26.3, far below 1.30.7, and yet the machine may be fixed. Distributions backport security fixes into their own branch without changing the upstream number: only the package suffix moves. The reliable criterion is therefore the full package number, read with dpkg -s nginx | grep '^Version' on Debian or Ubuntu, or rpm -q nginx on the RHEL family, compared with the table below.
For distributions whose status was not checked, read the distribution's errata or security tracker: zgrep CVE-2026-123007 /usr/share/doc/nginx*/changelog.Debian.gz on Debian or Ubuntu, rpm -q --changelog nginx | grep CVE-2026-123007 on the RHEL family. A matching line proves the fix is backported; no line proves it is not yet.
Fixed packages by distribution
Scroll the table
| Distribution | Fixed package | Reference |
|---|---|---|
| Debian 12 (bookworm) | `1.22.1-9+deb12u11` | DLA-4918-1 |
| Debian 13 (trixie) | `1.26.3-3+deb13u11` | DSA-6642-1 |
| Debian forky | Not fixed; fix in sid, `1.30.6-4` | Debian security tracker |
| Ubuntu 26.04 | `1.28.3-2ubuntu1.13` | Ubuntu security tracker |
| Ubuntu 24.04 | `1.24.0-2ubuntu7.20` | Ubuntu security tracker |
| Ubuntu 22.04 | `1.18.0-6ubuntu14.23` | Ubuntu security tracker |
| Ubuntu 20.04 (ESM) | `1.18.0-0ubuntu1.7+esm5` | Ubuntu security tracker |
| AlmaLinux, Rocky, RHEL | Status not verified: read the distribution errata | Distribution errata |
| nginx.org repository | 1.30.7 (stable) or 1.31.9 (mainline) | nginx.org security advisories |
"I ran nginx -s reload": no, that is not enough
nginx -s reload and systemctl reload nginx send a signal to the master process, which rereads the configuration and replaces the workers. The master, however, remains the same process, started from the old executable. After a package update, a reload therefore leaves the old code in memory: that is what the (deleted) mention in front of /proc/<pid>/exe shows.
To load the new binary, there are two routes. Restarting the service is the simplest and cuts connections for a few seconds. Hot replacement of the binary with the USR2 signal, described in the nginx documentation, avoids that cut but requires a good knowledge of your systemd setup. In both cases, you then check that the old processes are gone.
Fix, restart, check
Test the configuration first
sudo nginx -t
The expected answer containssyntax is okandtest is successful. An invalid configuration leaves the service stopped after a restart.Update the package
sudo apt update && sudo apt install --only-upgrade nginxon Debian or Ubuntu,sudo dnf upgrade nginxon the RHEL family. Then read the package version and compare it with the table:dpkg -s nginx | grep '^Version'.Restart, do not reload
sudo systemctl restart nginx
The interruption lasts a few seconds. Pick a quiet slot for production servers.Variant without interruption: hot replacement
sudo kill -USR2 $(cat /run/nginx.pid)starts a new master with the new binary. Thensudo kill -WINCH $(cat /run/nginx.pid.oldbin)stops the old master's workers, andsudo kill -QUIT $(cat /run/nginx.pid.oldbin)stops the old master. After each signal, check that the site still answers.Check that the old processes are gone
pgrep -a nginxmust list only recent processes.ps -o lstart= -p $(cat /run/nginx.pid)must give a date later than the package update.sudo ls -l /proc/$(cat /run/nginx.pid)/exemust no longer show(deleted).Check the service and HTTP/2 negotiation
curl -sI --http2 https://your-domain.com | head -1confirms the site answers. Note the date and the version in your fleet table, then move to the next server.
Docker and Nginx Proxy Manager: the image carries its own nginx
In a container, the nginx concerned is the one in the image, never the host's. Check it with docker exec <container> nginx -v. To fix it, pull or rebuild the image, then recreate the container: docker compose pull && docker compose up -d --force-recreate, or docker compose build --pull && docker compose up -d for an image built with FROM nginx:…. A bare docker restart keeps the old image.
Nginx Proxy Manager ships its own nginx. At the time of writing, the image we looked at contained nginx 1.31.8, an affected version. We have not established that a fixed image is already published: check the version inside the container with docker exec <container> nginx -v, and do not wait to act on HTTP/2. In the meantime, turn off the HTTP/2 setting of each proxy host, in its SSL tab. Our guide Nginx Proxy Manager on a VPS reminds you how it is deployed.
Workaround until you update: turn HTTP/2 off
On nginx 1.25.1 or newer, add http2 off; in each server block. On all versions, remove the http2 parameter from the listen directive: listen 443 ssl http2; becomes listen 443 ssl;. Then sudo nginx -t and sudo systemctl reload nginx: here a reload is enough, since only the configuration changes. Check with curl -sI --http2 https://your-domain.com | head -1, which must answer HTTP/1.1. The cost is a slight performance drop on pages with many resources. This is a stopgap: it does not replace the update, and it does not cover the HTTP/3 or mail proxy flaws.
Detection: look for traces without concluding too fast
Wiz points to three indicators. The first is worker crashes: grep -c 'exited on signal' /var/log/nginx/error.log counts lines such as worker process 12345 exited on signal 11. Compare with previous weeks, and also read journalctl -u nginx --since '7 days ago'.
The second is unexpected child processes: pstree -ap $(cat /run/nginx.pid) should show under the master only nginx: worker process entries and, if you use a cache, its manager. A shell or any tool launched by nginx is a strong signal. The third is long HTTP/2 connections sending many WINDOW_UPDATE frames: nginx does not log them by default, so look for them in the logs of a proxy or a network tool.
These traces are indicators, not proof, and their absence does not prove the server is clean. On a server you doubt, keep the logs before rotating them, then decide with the client whether TLS keys and certificates must be changed.
Troubleshooting: the errors that follow the update
nginx: [emerg] unknown directive "http2". You wrote http2 on; on a version older than 1.25.1, where the directive does not exist. Fix: update nginx, or drive HTTP/2 through the http2 parameter of listen (present to enable it, removed to disable it).
nginx -t succeeds, but the old binary is running. nginx -t tests the new binary and the configuration on disk, not the master process already started. Fix: restart, then check that (deleted) no longer appears in ls -l /proc/<pid>/exe.
apt keeps the old version. An apt-mark hold nginx or a file in /etc/apt/preferences.d/ freezes the package. Fix: apt-mark showhold, then sudo apt-mark unhold nginx. If nothing blocks it, the distribution may not have published its fixed package: see the table, and apply the workaround.
The container still answers with the old version. The image is cached or the tag is pinned. Fix: docker compose pull, docker compose up -d --force-recreate, then docker exec <container> nginx -v.
nginx: [emerg] bind() to 0.0.0.0:443 failed (98: Address already in use). An old master stayed alive after an interrupted hot replacement. Fix: pgrep -a nginx, stop the old master with sudo kill -QUIT $(cat /run/nginx.pid.oldbin), then restart.
Dozens of servers, one fix: the agency workload
For an agency, this fix is not one command, it is a tour of the fleet: one server per client, different origins (a Debian package here, the nginx.org repository there, a Docker image elsewhere), configurations written by several people. Each machine needs the inventory, the update, the restart, the check, then the answer to the worried client. Our guide on the October nginx configuration audit covers another wave, on the configuration side. If PHP is also in your fleet, see the PHP guide: the same tour is to be planned.
The most durable way to reduce this workload is to reduce the number of web servers you administer yourself. With ServOrbit.com reseller hosting, your clients' sites run in cPanel accounts that you manage from WHM, white-label: you keep the client relationship and your prices, we run the infrastructure. For machines that must stay under root access, the administration option is offered on our VPS Cloud.
Official documentation
servorbit.com/ma-en/blog/nginx2root-cve-2026-123007-update-nginx-http2