- What: F5 disclosed CVE-2026-123007 (nginx2root), a memory flaw in NGINX's HTTP/2 code present in every open-source release from 1.9.5 through 1.31.8, plus four more code execution flaws.
- Impact: An unauthenticated attacker with a reachable HTTP/2 port can run code as the NGINX worker process, and Xint chained it with a Linux kernel SCTP bug to get root on the host, even from inside a Docker container.
- Fix / mitigation: Upgrade to NGINX Open Source 1.31.9 or 1.30.7, or NGINX Plus 37.3.0.1 or 37.0.8.1, then fully restart; CIRCL lists disabling HTTP/2 as a temporary mitigation for the two HTTP/2 flaws only.
- Who's at risk: Any organization running NGINX or NGINX Plus with HTTP/2 enabled, especially Ubuntu and Debian hosts whose distribution packages have no fix yet and NGINX Plus R34 to R36 customers with no fix listed.
A flaw that has been in NGINX's HTTP/2 code since September 22, 2015 now gives unauthenticated attackers a path to code execution on web servers. CVE-2026-123007, dubbed nginx2root, affects every NGINX open-source release from 1.9.5 through 1.31.8. F5 released fixed versions 1.31.9 and 1.30.7 on November 18, 2026, along with fixes for four more flaws that it rates 9.2 out of 10 on CVSS 4.0. Researchers at Wiz and Xint both report running code through the bug, and public proof-of-concept code is on GitHub.
What the Flaw Does
The only condition for CVE-2026-123007 is HTTP/2 turned on. No login is needed. F5 says attackers can run code where ASLR is turned off or can be bypassed. Wiz says it bypassed ASLR and ran code in its tests. Xint says it ran code with nothing more than a reachable HTTP/2 port.
Code that runs this way gets the permissions of an NGINX worker process. CIRCL, the Computer Incident Response Center Luxembourg, notes that getting root takes a second privilege escalation. Xint supplied one: in its tests, code running as a worker reached root through a separate bug in the Linux kernel's SCTP code, even when NGINX ran inside a Docker container. Containerization did not contain it.
Xint chained nginx2root with an unpatched Linux kernel SCTP flaw to escalate from the NGINX worker to root on the host, with NGINX running inside Docker. Teams that count on containers to limit web tier compromise should not assume that boundary holds here.
Scope: Five Flaws, Five Configurations
The November 18 releases fix five flaws. F5 scores all of them 9.2 and says each can lead to code execution under the same ASLR condition. Each one is reachable only in a specific setup:
- CVE-2026-123007 (nginx2root): HTTP/2 turned on.
- CVE-2026-123633: HTTP/2, with requests passed to a backend server while the worker is running low on worker_connections.
- CVE-2026-123634: a proxied server name built from variables, plus a server-side include (SSI) or mirror subrequest to a location that looks up the same name.
- CVE-2026-121810: HTTP/3 turned on.
- CVE-2026-113425: NGINX used as a mail proxy.
Exposure covers far more than end-of-life builds. The October 14 security releases 1.31.7 and 1.30.6 are vulnerable, and so is 1.31.8 from October 21. Older branches such as 1.28 are affected too. NGINX Plus is affected in 37.x releases before the fixed builds and in R34 to R36. Behind a CDN, what matters is whether the NGINX origin itself accepts HTTP/2. An origin that accepts only HTTP/1.1 cannot be attacked through HTTP/2.
Exploitation Status and Attack Cost
As of November 18, neither F5 nor the Wiz and Rapid7 researchers have reported any in-the-wild exploitation, and the flaw is not in CISA's Known Exploited Vulnerabilities catalog. The public proof of concept on GitHub says it is suitable only for demonstrating the attack against a freshly started Docker container on the same machine.
Rapid7's technical analysis shows how much the attack costs. Its method needs about 128 GB of HTTP/2 traffic and took two and a half minutes in its demonstration. On links slower than about 300 Mbit/s, the transfer would run past NGINX's default one-hour connection limit and the attack would fail. The method also depends on a glibc memory-copy routine that is selected on x86-64 CPUs with the avx512f and avx512vl features. These limits apply to Rapid7's method. Wiz and Xint used their own approaches, and attackers will optimize.
Vendor Fixes and the Distribution Gap
F5 lists these fixed versions for CVE-2026-123007:
- NGINX Open Source 1.31.0 to 1.31.8 (mainline): fixed in 1.31.9.
- NGINX Open Source 1.9.5 to 1.30.6, including older branches such as 1.28: fixed in 1.30.7.
- NGINX Plus 37.1.0 to 37.2.0.1: fixed in 37.3.0.1.
- NGINX Plus 37.0.0 to 37.0.7.1: fixed in 37.0.8.1.
- NGINX Plus R34 to R36: no fix listed.
NGINX's release notes say 1.31.9 and 1.30.7 fix all five flaws. On NGINX Plus, 37.3.0.1 and 37.0.8.1 fix four of them. The CVE-2026-123633 fix shipped earlier, in 37.2.0.1 and 37.0.7.1. Official NGINX Docker images carried 1.31.9 and 1.30.7 tags on release day, including Alpine variants.
The gap is in Linux distributions. Ubuntu's CVE tracker listed nginx as vulnerable on releases from 16.04 to 26.10, and Debian's tracker listed bookworm, trixie, forky and sid as vulnerable, with no fixed packages from either as of November 18. Distribution packages can carry backported fixes without changing the version number NGINX reports, so the distribution's own advisory is the authority on patch status. F5 also notes that a configuration reload alone leaves the old executable running, so a patched binary needs a full restart or a live binary upgrade before it protects anything.
F5 lists workarounds for only two of the five flaws: avoiding SSI, mirror directives and variable-built proxied server names for CVE-2026-123634, and restricting mail proxy listeners to trusted networks for CVE-2026-113425. CIRCL lists disabling HTTP/2 as a temporary measure, which blocks CVE-2026-123007 and CVE-2026-123633 but not the other three. Disabling HTTP/3 removes the condition for CVE-2026-121810. CIRCL also advises blocking the SCTP kernel module on hosts that do not use it until the kernel bug is fixed.
Detection Is Difficult
A clean access log does not rule out an attack. HTTP/2 control frames do not normally appear there as requests, so exploitation traffic can leave almost no trace in the logs most teams watch. CIRCL's indicators include NGINX workers that start unexpected programs, workers that exit abnormally or leave core dumps, long-lived HTTP/2 connections with heavy incoming traffic but few logged requests, and unusual outbound connections or DNS lookups from the web tier. None of these is proof on its own. Together they call for an investigation, especially on hosts where HTTP/2 was exposed during the disclosure window.
RedEye Assessment
nginx2root is serious because of where NGINX sits: at the internet edge, in front of applications, often on hosts that also hold credentials and internal network access. Public PoC code, three independent research teams reporting code execution, and a demonstrated container escape chain mean the window before weaponized exploitation is likely short. The 128 GB traffic requirement and the CPU dependency in Rapid7's method are real limits, but other teams did not report those limits for their methods. Unpatched Ubuntu and Debian fleets and NGINX Plus R34 to R36 deployments are the highest-risk groups right now. Organizations should inventory every NGINX instance, including origins behind CDNs and NGINX inside container images, confirm which have HTTP/2, HTTP/3 or mail proxying enabled, and track vendor and distribution advisories until each one is on a fixed build.
Questions about your exposure?
RedEye Security provides assessments for organizations that need to understand their real risk.
Talk to us