The bug in two sentences
In HTTP/2 flow control, queued data is credited back to the connection’s send window twice: once by aWINDOW_UPDATE, once by stream cleanup after a RST_STREAM. The window then exceeds its allowed limit, an integer underflow (CWE-191) yields a negative length, and a huge memory copy writes past the buffer: the attacker hijacks a function pointer and runs commands with the privileges of the nginx worker.
The flaw was found by xint.io, who named it nginx2root. The name comes from the chain they describe: the nginx worker first, then root through a separate Linux kernel flaw in SCTP, whose details are not public yet.
Affected versions:
- nginx 1.9.5 through 1.30.6, fixed in 1.30.7 (stable branch);
- nginx 1.31.0 through 1.31.8, fixed in 1.31.9 (mainline branch).
AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H) and 9.2 in CVSS 4.0. The NVD entry carries the same scores. Red Hat rates it “Critical”, Ubuntu high priority.
Rapid7 has published a proof of concept that achieves unauthenticated remote code execution against nginx 1.31.8 with HTTP/2 and TLS. The flaw is not in CISA’s KEV catalog (November 20, 2026 version), and no active exploitation is reported. The same update fixes at least eight other nginx CVEs.
Sizing your exposure
The question to ask: does your nginx accept HTTP/2? Red Hat states that exploitation requires a server that accepts HTTP/2 connections. An nginx that only serves HTTP/1.1 is not exposed to this CVE.
To check whether HTTP/2 is on, read the loaded configuration:
listen ... http2 directive or an http2 on; line means HTTP/2 is enabled on that port.
Don’t rely on the version printed by
nginx -v alone. Debian, Ubuntu and AlmaLinux backport the fix without changing the upstream version: Ubuntu 24.04 still reports nginx 1.24.0 once patched. Only the full package version, compared with your distribution’s advisory, is the reference.Updating nginx, distro by distro
The real fix is a package update followed by a restart of nginx: a plainreload keeps the old workers running for as long as they serve connections.
Take a snapshot of your VPS first.
Debian
Ubuntu
Advisory: USN-9159-1, November 19, 2026.
RHEL and AlmaLinux
Red Hat lists RHEL 8 (thenginx:1.24 stream), RHEL 9 and RHEL 10 as affected. As of November 20, 2026, its bulletin RHSB-2026-019 says fixes are on the way, with no erratum released yet: follow the Red Hat CVE page.
AlmaLinux shipped its fixes on November 18, 2026:
For AlmaLinux 8 and the
nginx:1.24 and nginx:1.26 module streams on AlmaLinux 9, check errata.almalinux.org.
Kubernetes
The communityingress-nginx controller is no longer maintained: according to Upwind, every release is affected and none will be fixed. HTTP/2 is on by default. Turn it off in the controller’s ConfigMap:
Disabling HTTP/2 until you can update
If you can’t update right away, Red Hat and xint.io recommend disabling HTTP/2. Clients fall back to HTTP/1.1 on their own. Remove thehttp2 parameter from every listen directive:
http2 on; with:
A gRPC frontend cannot do without HTTP/2:
grpc_pass needs it. There, only the update protects you. Disabling HTTP/2 covers CVE-2026-123007 and CVE-2026-123633; the other flaws in the release, including the mail proxy one (CVE-2026-113425), need the update.Blocking SCTP until the kernel fix
xint.io reaches root by chaining the nginx flaw with a flaw in the kernel’s SCTP module. Until a fixed kernel ships, they recommend preventing the module from loading:Checking the mitigation
grep from “Sizing your exposure” again too: no listen directive should carry http2, and no http2 on; line should remain.
Re-enabling HTTP/2 after the update
Once the fixed package is installed and nginx restarted, bring HTTP/2 back:use-http2 at "false" for as long as the community ingress-nginx controller is in place.

