Nginx2root : http2 is on by default in the standard container. What's the plan?

סיכום מבוסס בינה מלאכותית

The discussion addresses the nginx2root vulnerability (CVE-2026-123007), which affects HTTP/2 enabled in the default Discourse Nginx container. spinitom raised concerns about remote unauthenticated execution risks and asked for a patched base image or a way to disable HTTP/2 via app.yml. tgxworld confirmed a new base image was being released to address the issue.

For immediate mitigation without a rebuild, pfaffman shared a script to disable HTTP/2 in running containers. sam later announced the patch was available and strongly urged all self-hosters to rebuild their images immediately using ./launcher rebuild app. Jagster clarified the equivalent commands for multi-container setups.

Some users, like fuzzyk, were confused because the Nginx version remained 1.26.3. david explained that security patches were applied on top of the existing Debian-packaged Nginx version rather than upgrading the binary, as Debian had not yet released a fixed package. Users can verify they have the patched image by checking /VERSION inside the container (should be 2.0.20261119 or later) or checking the package version with apt list nginx.

To ensure admins notice the critical update, david noted that all Discourse releases were bumped to trigger the standard “update available” notification on the dashboard and via email, prompting a rebuild. agemo suggested improvements for how security topics are surfaced to admins who may not check meta regularly. The response culminated in a linked topic detailing the official response to the Ruby2Shell and nginx2root vulnerabilities.

“nginx2root” ( NGINX2Root: New Remote Code Execution Vulnerabilities in nginx Require Urgent Patching - Xint Blog | Insights from Offensive Security Researchers ) lists five nginx CVEs, remote and unauthenticated. It says 1.31.9 fixes them, due Nov 19.

Two of the five need HTTP/2, which the Discourse standard container enables.

After a fresh rebuild, templates/web.ssl.template.yml:46 writes http2 on; into /etc/nginx/conf.d/outlets/server/20-https.conf, and nginx -v in the container says 1.26.3.

  1. Is a patched nginx in the base image planned, and when?
  2. Is there a supported way to turn HTTP/2 off from app.yml? (I’m using a run: hook that seds http2 on; to http2 off;.)

We’re currently working on releasing a new base image to address the vulnerability. We’ll share an announcement soon once that is done.

Here’s a script to disable http/2 so you can do the upgrade when it’s more convenient. I used it fo a bunch of servers, but you should check for yourself.

#!/usr/bin/env bash
# Temporary: turn off nginx http2 in running Discourse containers, no rebuild.
# Usage: bin/disable-http2-live user@host [user@host ...]
#        xargs bin/disable-http2-live < hosts.txt
# Lost on rebuild; the after_ssl hook in the container yml covers that.
# Delete this script once a patched release is out.
for host in "$@"; do
  echo "== $host"
  ssh -o BatchMode=yes -o ConnectTimeout=10 "$host" sudo bash -s <<'EOF' || echo "FAILED: $host"
set -e
files="/etc/runit/1.d/install-ssl /etc/nginx/conf.d/outlets/server/20-https.conf"
# Containers publishing 443 without these files (traefik etc.) are left alone.
for c in $(docker ps --filter publish=443 --format '{{.Names}}'); do
  docker exec "$c" sh -c "grep -qs 'http2 on;' $files || { echo '$c: nothing to do'; exit 0; }
    sed -i 's/http2 on;/http2 off;/' $files && nginx -t && sv reload nginx && echo '$c: patched'"
done
EOF
done

All NGINX instances on the company’s production servers have been checked and upgraded; Discourse needs to follow suit quickly.

Discourse is following suit, we have a patched image building it should be available in a few hours.

this is now patched … we strongly recommend all self hosters rebuild image ASAP:

./launcher rebuild app

And for us self-hosting amateurs with 2-container setup this does same thing, right?

git pull && ./launcher bootstrap web_only && ./launcher destroy web_only && ./launcher start web_only

nginx-1.30.7 stable and nginx-1.31.9 mainline versions have been released, with vulnerability fixes for use-after-free in resolver (CVE-2026-123634), memory corruption in HTTP/2 flow control (CVE-2026-123007), use-after-free in HTTP/3 (CVE-2026-121810), use-after-free in HTTP/2 (CVE-2026-123633), infinite loop while reading client request body over TLS 1.3 (CVE-2026-118899), use-after-free in mail proxy module (CVE-2026-113425), NULL pointer dereference when using PROXY protocol version 2 (CVE-2026-120993), buffer overflow when using alias with try_files (CVE-2026-125283), SNI bypass in client certificate validation in stream module (CVE-2026-120967), and insufficient validation in stream ssl_preread module (CVE-2026-123640).

How can i verify the patch?

See Determine Ruby Version in Images (regular expression cve) - #6 by RGJ

Hate to be “that guy” but… how is this patched? I rebuilt, it pulled discourse/base:2.0.20261119-0203 but I have nginx 1.26.3. Which is… old, to say the least.

Am I just missing something?

/edit: So after digging some more, if I have package version nginx=1.26.3-3+deb13u10+discourse3, the patches are in??

Command to check is apt list nginx.

We have applied the security patches on top of nginx 1.26.3, because Debian has not yet shipped a new nginx package with the fixed version. Implementation of our patch can be found here

To verify you are running our latest docker image, you can run this command:

./launcher enter app
# cat /VERSION
2.0.20261119

If you see anything earlier than 20261119, then you need to upgrade. The image upgrade is being forced for anyone using the standard install.

It would be nice if Discourse admins received a clear warning message like “Urgent: you need to update now!” instead of this “boring” message:

We have just bumped the version for all Discourse releases, so admins will see an ‘update available’ message in their main dashboard, and will receive the update-available email notification:

When they click through, they’ll see the prompt to rebuild the image.

I meant to suggest something like this weeks ago around another earlier security issue. I felt as a topic these events can get lost in the mix, better late than never:

Point today is, I hadn’t logged in in weeks to meta, so by chance I came in on the tail end of this issue being sorted, great timing on my part lucky me, but we can’t rely on that for everyone.