News: nginx2root: what it means for 3CX phone systems, and how to mitigate it
Current 3CX V20 for Linux ships its own copy of nginx 1.30.6 with HTTP/2 turned on, which puts it in scope for two of the nginx2root vulnerabilities disclosed on November 18, 2026 (CVE-2026-123007 and CVE-2026-123633). If you run your own 3CX system, below is what we found, how to check it, and the interim fix we applied while we wait for 3CX to ship a patched build.
What nginx2root is
nginx2root is a set of five nginx vulnerabilities published by Xint, with write-ups from Wiz and CIRCL. Two of them need HTTP/2:
- CVE-2026-123007, an HTTP/2 flow control bug that allows unauthenticated remote code execution. It affects nginx 1.9.5 through 1.31.8, and Rapid7 has published a proof of concept.
- CVE-2026-123633, a use-after-free when an HTTP/2 connection is torn down while a request is connecting to an upstream server. It affects nginx 1.19.7 and later.
The other three need a DNS resolver with a request-derived proxy_pass and mirror, HTTP/3 (QUIC), or the nginx mail proxy. nginx fixed all five in 1.30.7 (stable) and 1.31.9 (mainline).
The researchers also chained an unpatched Linux kernel SCTP bug to go from the nginx worker to root on the host. That kernel bug has no CVE or patch yet.
As of November 18 there are no reports of these being exploited in the wild, and they're not in CISA's Known Exploited Vulnerabilities catalog.
Why 3CX is affected
None of the published write-ups mention 3CX, and we haven't seen a 3CX advisory yet. On 3CX V20 for Linux, we found:
/usr/sbin/nginxbelongs to the3cxpbxpackage, not Debian's nginx. 3CX started shipping its own nginx build with its June 2026 security hotfix. Upgrading Debian's nginx won't touch it, so the fix has to come from 3CX.3cxpbx20.0.10.1785, the current build as of November 18, ships nginx 1.30.6. An older 20.0.9 build we checked has nginx 1.30.2. Both are below the fixed 1.30.7.- The stock config has
http2 on;and proxies to an upstream (proxy_pass http://gateway). That meets the published prerequisites for both HTTP/2 CVEs. - The stock config doesn't use a resolver with
mirror, QUIC, or the mail proxy, so the other three CVEs don't apply as published. - 3CX V20 for Linux runs on Debian 12 with kernel 6.1. The
sctpmodule wasn't loaded on any system we checked, but Debian can load it on demand. - One summary of the Rapid7 proof of concept says it needs a CPU with AVX-512. Xint says their exploit worked across different hardware, so an older CPU isn't a mitigation.
We haven't checked 3CX on Windows or 3CX's own hosted instances.
How to check your 3CX system
Run these as root over SSH. For the curl check, use your PBX's FQDN so the certificate name matches, and add :5001 if that's your web console port.
dpkg -l 3cxpbx | tail -1
/usr/sbin/nginx -v
nginx -T | grep -n http2
curl -sk --http2 -o /dev/null -w '%{http_version}\n' https://pbx.example.com/
If curl prints 2, that endpoint still offers HTTP/2 and is in scope for the HTTP/2 bugs. Run the curl from a machine whose curl supports --http2.
The interim fix
-
Turn HTTP/2 off in the real config file.
/etc/nginx/sites-enabled/3cxpbxis a symlink to/var/lib/3cxpbx/Bin/nginx/conf/nginx.conf, and a plainsed -ion the link replaces it with a regular file. Edit the target instead:cp -p /var/lib/3cxpbx/Bin/nginx/conf/nginx.conf /root/3cx-nginx.conf.bak sed -i 's/http2 on;/http2 off;/' /var/lib/3cxpbx/Bin/nginx/conf/nginx.conf nginx -tIf your build has
http2on alistenline instead, remove it there. -
Reload nginx:
systemctl reload nginx. That's what we did. New connections get HTTP/1.1, but old workers keep serving any existing HTTP/2 connections until those close.systemctl restart nginxcuts them off right away, at the cost of a short drop for web client and app sessions. -
Check again from outside. The curl check above should now print
1.1. -
Block SCTP autoload and confirm it took:
echo 'install sctp /bin/false' > /etc/modprobe.d/disable-sctp.conf python3 -c 'import socket; socket.socket(socket.AF_INET, socket.SOCK_STREAM, 132)'The second command should fail with "Protocol not supported". The block doesn't unload a module that's already loaded, and it does nothing if SCTP is built into the kernel. If
lsmod | grep sctpshows it loaded, find out what's using it before you plan a reboot.
After the change, the 3CX admin console and API answered normally, and trunk and extension status didn't change. So far nothing in 3CX has complained about HTTP/2 being off.
What to watch for
- A 3CX update may rewrite that nginx.conf and turn
http2 onback on. After every update, checknginx -v,nginx -T | grep http2and the outside curl. If nginx is still below 1.30.7 and HTTP/2 is back, repeat step 1. - Once 3CX ships nginx 1.30.7 or later, install it and restart nginx, because a reload keeps the old binary running. Confirm the version, then turn HTTP/2 back on if you want it.
- Leave the SCTP block in place until your kernel has a fix.
- If your remote extensions and apps all connect from known addresses, limiting access to the web port at your firewall helps no matter what. Many deployments can't do that.
Sources
- Xint: nginx2root
- CIRCL TR-105
- Wiz: CVE-2026-123633 and CVE-2026-123007
- nginx security advisories
- 3CX: June 2026 HTTP/2 hotfix