NGINX shipped fixes on 18 November for a memory-corruption bug in its HTTP/2 flow control that has been in the open-source code since September 2015. The flaw, tracked as CVE-2026-123007 and nicknamed “nginx2root,” affects every release from 1.9.5 through 1.31.8, including the October security builds. Fixed versions are 1.31.9 on the mainline branch and 1.30.7 on stable.
What the bug actually does
According to F5’s advisory, a server processing undisclosed HTTP/2 traffic can let the flow-control window exceed its allowed threshold. That turns into a heap buffer overflow, and in practice the worker process usually dies with a segmentation fault. The interesting part is what happens when address space layout randomization is off, or when an attacker can get around it: the overflow becomes a path to remote code execution.
Wiz researchers said they bypassed ASLR and ran code through the flaw in their tests. Xint reported running code against servers on AWS, GCP and Azure using nothing more than a reachable HTTP/2 port. The code runs with the privileges of an NGINX worker, not root. In Xint’s setup, a second bug in the Linux kernel’s SCTP stack took the worker to root, even with NGINX inside a Docker container. CIRCL’s guidance treats that escalation as a separate step, and the SCTP module is worth blocking on hosts that do not use it.
Four more flaws in the same release
The same update closes four other issues: CVE-2026-123633 (HTTP/2, triggered when requests are proxied to a backend while the worker is low on worker_connections), CVE-2026-123634 (a proxied server name built from variables, combined with SSI or a mirror subrequest), CVE-2026-121810 (HTTP/3) and CVE-2026-113425 (the mail proxy module). The Hacker News reports that F5 scores all five 9.2 out of 10 under CVSS 4.0. NGINX’s own advisory page rates CVE-2026-123007 as “major.” Neither F5 nor the researching firms have reported exploitation in the wild, and the vulnerability was not in CISA’s Known Exploited Vulnerabilities catalog at publication time.
Who is exposed
HTTP/2 is enabled when a listen directive includes http2, or when http2 on; is set. If you terminate TLS at a CDN and your origin only speaks HTTP/1.1, the origin cannot be reached over HTTP/2 and this particular flaw does not apply to it. Check with:
openssl s_client -connect example.com:443 -alpn h2 </dev/null 2>/dev/null | grep ALPN
If the output says ALPN protocol: h2, the server accepts HTTP/2.
Ubuntu and Debian had not published fixed packages as of 18 November; both trackers listed nginx as vulnerable. A distribution backport can carry the fix without changing the version string nginx -v prints, so compare the full package version against the distribution advisory rather than the upstream number. Official Docker images for 1.31.9 and 1.30.7, including Alpine variants, were available the same day. NGINX Plus is affected in 37.x builds before 37.3.0.1 / 37.0.8.1, and in R34 through R36, for which no fixed build was listed.
What to do next
- Upgrade to 1.31.9 or 1.30.7, or the matching Plus build.
- Restart NGINX fully, or use the live binary upgrade. A reload keeps the old executable running and can leave existing connections on old workers.
- If you cannot upgrade today, turn HTTP/2 off. On 1.25.1 and later set
http2 off;and drophttp2fromlisten; on older builds remove it from thelistenline. Re-run the openssl check to confirm. - Where HTTP/3 is enabled and unused, disable it to remove the setup CVE-2026-121810 needs.
- Avoid SSI and
mirror, and do not build proxied server names from variables, until you are patched. - Block the SCTP kernel module on hosts that do not need it:
install sctp /bin/falsein/etc/modprobe.d/.
Signs of an attack
CIRCL suggests watching for workers that spawn unexpected programs, workers that exit abnormally or leave core dumps, long-lived HTTP/2 connections with heavy inbound traffic but few logged requests, and unusual outbound connections or DNS lookups. A clean access log proves nothing on its own, because HTTP/2 control frames do not appear there as requests. Rapid7’s reproduction needed roughly 128 GB of HTTP/2 traffic and about two and a half minutes, which means links slower than roughly 300 Mbit/s would run past NGINX’s default one-hour connection limit and the attempt would fail. The technique also depends on a glibc memory-copy path selected on x86-64 CPUs with AVX-512 support.
Sources: The Hacker News, nginx.org security advisories, NVD, F5, Wiz, Xint, CIRCL and Rapid7.



0 Comments