Nginx2Root Opens Every nginx Release Since 2015 to Remote Code Execution, Patch Now
nginx shipped 1.30.7 (stable) and 1.31.9 (mainline) on November 18, and the release notes are long. Ten CVEs in all. Five of them are memory corruption bugs in the worker processes that the security firm Xint grouped under one name, Nginx2Root. Wiz found the same bugs on its own, as cybersecuritymarket.com reports on Xint’s AI platform and the five nginx RCE flaws. Xint reported four of the five. F5 scores each of them 9.2 on CVSS 4.0.
The one that matters for almost everyone is CVE-2026-123007, a memory corruption bug in HTTP/2 flow control. It has been in the code since nginx 1.9.5, the first release with HTTP/2, in 2015. When a client cancels a request, nginx hands the unsent part of the flow-control window back to the connection without checking it against the protocol maximum. A client can push the window past the limit, a later length calculation goes wrong, and the worker copies far more memory than it should. Xint says it turned that into heap reads and writes, then into running commands as the nginx worker over a public HTTPS port.
Dane Knecht, Cloudflare’s CTO, put out the short version for Cloudflare customers:
nginx2root. The HTTP/2 bug in nginx2root goes back to nginx 1.9.5. The 5 memory corruption bugs in nginx were discovered by Xint. Depending on config, an attacker can cause a DoS with a a crashed worker or trigger RCE and run code.
— Dane Knecht 🦭 (@dok2001) November 19, 2026
It is fixed in Nginx 1.30.7 and patched in 1.31.9.
Cloudflare's edge was fixed globally before disclosure, so proxied traffic locked down to us needs nothing from you. If you are using as for DNS-only or proxy just TCP with spectrum please patch your origin.
https://t.co/YspFQqe10x
Crash or code execution depends on the box
Every one of the five can crash a worker. Code execution is possible where ASLR is off or where the attacker gets around it. Xint did get around it, leaking memory addresses first. That is the part to take seriously: a default HTTPS server that only serves static files is enough of a target. Chained with a Linux SCTP kernel flaw, the same research reaches root on the host.
The other four need more specific setups:
- CVE-2026-123633, use-after-free in HTTP/2 when requests are proxied to a backend and the worker runs out of
worker_connectionsslots. nginx 1.19.7 onward. - CVE-2026-121810, use-after-free in HTTP/3 stream handling, same low-slots condition. nginx 1.25.0 onward.
- CVE-2026-123634, use-after-free in the resolver, when proxying to server names with variables while an SSI or
mirrorsubrequest resolves the same name. nginx 0.6.18 onward. - CVE-2026-113425, use-after-free in the mail proxy (IMAP, POP3, SMTP). nginx 0.5.33 onward, so only boxes running the mail module.
Who has to do something
If your site sits behind Cloudflare’s proxy and the origin only accepts Cloudflare traffic, the HTTP/2 connection an attacker can reach terminates at Cloudflare’s edge, which was patched before disclosure. Nothing to do. If you use Cloudflare for DNS only, or pass plain TCP through Spectrum, the attacker talks to your nginx directly. Patch the origin.
Everyone else running nginx on a public port: upgrade to 1.30.7 or 1.31.9, or NGINX Plus 37.3.0.1 or 37.0.8.1. The official Docker images already carry the fixed builds, and nginx.org’s own packages do too. Distribution packages are arriving at different speeds, and old streams that are out of support, AlmaLinux’s retired nginx module streams among them, will not get a patch at all.
Where an upgrade has to wait for a change window, turn HTTP/2 off. Remove http2 from the listen lines, or set http2 off; on newer builds, then nginx -t and a reload. It closes the lead bug and CVE-2026-123633. It also breaks gRPC proxying, which needs HTTP/2, so check before flipping it on an API gateway.
The usual inventory problem applies. nginx hides inside appliances, ingress controllers, vendor images and that one VM someone built in 2019. Those copies are the ones that stay unpatched longest, and an internet-facing HTTP/2 listener on any of them is the exposure. A simple rule works here: anything reachable from outside gets patched or has HTTP/2 off this week, internal boxes go into the normal cycle. More on working a patch backlog by exposure, and on what happens to widely used open source once its maintainers step back.