Dev.to Security 🔐 Cybersecurity 👁 0 📖 5 min read

NGINX2Root: Five RCEs in the Server That Runs the Web, and the Kernel Bug Nobody Patched

On November 18, F5 published an out-of-band security release for NGINX, the web server and reverse proxy that fronts a large share of the internet. Ten CVEs landed in a single day, outside the project's normal release rh

On November 18, F5 published an out-of-band security release for NGINX, the web server and reverse proxy that fronts a large share of the internet. Ten CVEs landed in a single day, outside the project's normal release rhythm. Researchers had already given the event a name: NGINX2Root. Five of the flaws are memory-safety bugs that can hand a remote attacker code execution inside the NGINX worker process. The "2Root" half of the name comes from a separate bug entirely - a use-after-free in the Linux kernel's SCTP implementation, present since 2018, still unpatched and still without a CVE - that can turn that first foothold into full root on the host, escaping the container along the way.

As of November 18, none of the flaws is known to be exploited in the wild, and none appears in CISA's Known Exploited Vulnerabilities catalog. Public proof-of-concept code, however, was out within hours.

The five RCE-class bugs

All five are memory-safety flaws, but they live in different parts of the server:

  • CVE-2026-123007: an integer underflow in HTTP/2 send_window accounting, in the codebase since NGINX 1.9.5 in 2015 - more than a decade.
  • CVE-2026-123633: a use-after-free in HTTP/2 request handling when proxying.
  • CVE-2026-123634: a use-after-free in the DNS resolver.
  • CVE-2026-121810: a use-after-free in the HTTP/3 module.
  • CVE-2026-113425: a use-after-free in the mail proxy.

The researchers who found them rate all five at CVSS 4.0 9.2 (8.1 under CVSS 3.1). F5's own advisories score the same bugs lower - "major" and "medium" on NGINX's internal scale. The discrepancy is worth stating plainly rather than averaging away: researchers are scoring the worst case reachable by a network attacker, while the vendor factors in configuration constraints and exploit reliability. Which number matters depends on which side of the patch window you sit.

Fixes are in NGINX open source 1.30.7 and 1.31.9, and in NGINX Plus 37.0.8.1 and 37.3.0.1 (CVE-2026-123633 is additionally fixed in 37.0.7.1 and 37.2.0.1).

Who found what

The discovery credits read like a slow year compressed into five weeks. Theori researcher Xint found the HTTP/2 underflow on October 11 and reported it through HackerOne on October 16. Wiz reported two of the flaws, on October 21 and November 16. The remaining bugs are credited to Ikotas Labs, Metnir, and Rapid7. F5 coordinated a single release for the whole set, which is why ten CVEs share one publication day.

Public exploits within hours

Two proof-of-concept exploits for CVE-2026-123007 were public on release day. Rapid7's Stephen Fewer published a full exploit, but with real-world friction: it needs roughly 128 GB of traffic and anywhere from two minutes to an hour per attempt, and it assumes an x86-64 glibc target. Censys released a second, leaner PoC that works over cleartext HTTP/2 without the large traffic volume. Wiz says it is withholding its own exploit code to give defenders time. The practical reading: exploitation is proven but not yet push-button, and the constraint that matters most is whether HTTP/2 is exposed to untrusted clients at all.

The kernel half nobody patched

The more uncomfortable finding is not in NGINX at all. The NGINX2Root write-ups describe a use-after-free and double-free in the Linux kernel's SCTP implementation, present since Linux 4.15 in 2018. It has no CVE and no patch. Because the sctp kernel module auto-loads the moment any process creates an SCTP socket, an attacker with code execution inside an NGINX container can trigger it and escape to root on the host. The suggested mitigation is blunt but effective: block the module from loading with install sctp /bin/false in the modprobe configuration. The episode is a reminder that a container boundary is only as strong as the kernel underneath it.

How much of the internet is exposed

The numbers vary by who counted and when, and each deserves its attribution. Xint's November 10 scan put 56.7 million hosts in the vulnerable version range - 98.9% of the NGINX deployments it observed. Censys counted between 8.8 and 9 million hosts actually exposing the vulnerable HTTP/2 service for CVE-2026-123007, with about 511,000 already patched and roughly 103 to 104 million more not disclosing a version at all. Florian Roth added a caveat that should travel with every one of these figures: only about 66% of the HTTPS endpoints he sampled support HTTP/2 in the first place, and version-string counting overstates real exposure because distributions backport fixes without changing version numbers. Exposure estimates are maps, not territory.

The ecosystem lag

Two structural problems make this patch cycle harder than the CVE list suggests. First, ingress-nginx - the Kubernetes ingress controller built on NGINX - was archived on March 24, 2026, and will not ship fixes. Every cluster still running it stays vulnerable until it migrates to a maintained alternative. Second, as of November 18 neither Debian nor Ubuntu had a fixed package in their repositories; every supported release of both distributions was listed as vulnerable. Distribution fixes will arrive as package revisions, which is exactly the case where the version string does not change - another reason version-based counting misleads.

What defenders can do now

The practical list, assembled from the advisories and the researcher write-ups:

  • Upgrade to a fixed NGINX build: 1.30.7 or 1.31.9 for open source, or the patched Plus releases.
  • Where HTTP/2 is not needed, turn it off: http2 off; on 1.25.1 and later, or remove http2 from the listen directive on older builds. Restart the server rather than reloading it, so no worker keeps running the old code.
  • Disable HTTP/3 where it is enabled and not required.
  • Prefer static, fixed values in proxy_pass, resolver, mirror and SSI directives; variable-driven resolution is where several of the dangerous code paths live.
  • Block SCTP autoloading (install sctp /bin/false) on hosts and in container base images.
  • Rebuild container images against fixed bases; patching the host alone does not fix what runs inside.
  • Watch for the symptoms the write-ups list: crashing or respawning NGINX workers, unexpected commands in worker context, and unusually long-lived HTTP/2 connections.

The takeaway

NGINX2Root is not the first time a decade-old line of code turned out to be load-bearing for a large share of the internet, and it will not be the last. The instructive part is the shape of the event: the worst bug was found in mid-October, the fix shipped as an exceptional same-day release, the exploits went public within hours, and the slowest link - distribution packages and an archived ingress controller - is what most deployments actually run. The window between "a PoC exists" and "your package manager has the fix" is where all the risk lives. Closing that window is a supply-chain problem, and it does not fit inside a CVSS score.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.