DEV Community

Danila Vershinin for GetPageSpeed

Posted on Originally published at getpagespeed.com AI-assisted

nginx2root: why nginx -v can't tell you if you're patched

On 2026-11-18 five NGINX CVEs went public under the name nginx2root. Two hit HTTP/2 (CVE-2026-123007, CVE-2026-123633). One hits HTTP/3 with an upstream (CVE-2026-121810). One needs a resolver, a variable in proxy_pass and subrequests such as mirror (CVE-2026-123634). The last one is the mail proxy with auth_http (CVE-2026-113425). Upstream fixed them only in 1.30.7 (stable) and 1.31.9 (mainline). Older branches get no upstream release.

That last sentence is where most people go wrong.

Your distro fixes it without changing the version

Distributions don't jump to 1.30.7. They backport the patches into the version they already ship. After the fix, nginx -v still prints 1.18.0, 1.22.1 or 1.24.0, exactly as before. The only thing that changes is the package release:

dpkg -l nginx      # Debian / Ubuntu
rpm -q nginx       # RHEL family, Amazon Linux
Enter fullscreen mode Exit fullscreen mode

On Ubuntu 24.04 the fixed package is 1.24.0-2ubuntu7.20. On AlmaLinux 9 it's 1.20.1-32.el9_9.1.alma.4, on AlmaLinux 10 1.26.3-16.el10_3.1.alma.3. nginx -v on those boxes still says 1.24.0, 1.20.1 and 1.26.3. A version-string scanner flags them as vulnerable, and a human comparing against "1.30.7" concludes the same.

It also works the other way. AlmaLinux's first builds on the day fixed only the two HTTP/2 CVEs. So "the package was updated today" doesn't mean "all five are fixed". You need to know which CVEs a specific package release covers.

Where things stood on 2026-11-19:

  • Ubuntu 22.04, 24.04, 26.04: fixed that morning in USN-9159-1 (16.04-20.04 for Ubuntu Pro only)
  • Debian 12 and 13: vulnerable, no DSA
  • RHEL 8/9/10: affected, no erratum; Rocky follows RHEL
  • AlmaLinux 9/10 and the nginx:1.24 / nginx:1.26 streams: backports shipped
  • Amazon Linux 2 (nginx1 extra) and 2023: fixed at 1.30.7
  • Alpine 3.22-3.24: fixed (1.28.3-r8 backport, 1.30.7-r0)
  • CentOS 7: nothing. EPEL is frozen at nginx 1.20.1-10.el7, and the nginx.org CentOS 7 repo stopped at 1.26.1

A reload doesn't load the new binary

nginx -s reload and systemctl reload nginx re-read the config. The running master keeps executing the old binary. After any package update, do a full restart. Then confirm:

readlink /proc/$(cat /run/nginx.pid)/exe
Enter fullscreen mode Exit fullscreen mode

If the output ends in (deleted), the process is still running the binary the package manager just replaced. You're not patched yet.

Turning off HTTP/2 is a partial measure

Removing http2 from your listen lines (or setting http2 off) blocks the two HTTP/2 CVEs. It does nothing for the other three. Check your nginx -T output for HTTP/3, for a resolver combined with a variable in proxy_pass, and for the mail module with auth_http. If none of those appear, disabling HTTP/2 is a reasonable stopgap until a package lands. If any do, it isn't enough.

Close the root step too

The "root" in nginx2root comes from a Linux kernel SCTP bug that turns worker-level code execution into root. That bug has no CVE and no fix yet. If the host doesn't use SCTP, block the module:

echo 'install sctp /bin/false' | sudo tee /etc/modprobe.d/disable-sctp.conf
Enter fullscreen mode Exit fullscreen mode

This doesn't fix the NGINX CVEs, but it removes the escalation path.

Checking it without reading changelogs

I maintain the GetPageSpeed package repository, and we built a free browser-based checker for this: https://www.getpagespeed.com/nginx2root. Paste the output of nginx -v, rpm -q nginx or dpkg -l nginx and it accounts for the distro backports. Paste nginx -T and it maps your config to the CVEs that are actually reachable.

Top comments (1)

Collapse
 
alexcodebytes profile image
Oleksandr •

The distinction between the upstream version and the distro package release is really important here. Too many vulnerability checks still rely on version strings and end up producing false positives for backported fixes. The point about systemctl reload is just as valuable — updating the package doesn't mean the running master process is using the new binary. Checking both package status and the actual running executable should be part of the remediation workflow, not an afterthought.