Live wire

NGINX HTTP/2 Flaw Lets Unauthenticated Attackers Run Code on Servers

The 10-year-old flaw affects releases from 1.9.5 to 1.31.8; fixes are out, a public proof of concept exists, and no attacks have been reported.

A flaw in the HTTP/2 code of the NGINX web server, tracked as CVE-2026-123007 and dubbed nginx2root, lets an unauthenticated attacker run code on servers that have HTTP/2 enabled. It has existed since version 1.9.5, released in September 2015, and affects every open-source release through 1.31.8. Code runs with the permissions of an NGINX worker process, and gaining root needs a second step, for which researchers used a separate Linux kernel bug, even when NGINX ran in a container. The vendor says attacks work where address randomisation is disabled or bypassed, and one security firm bypassed it in tests.

Exploitation and constraints

No attacks had been reported and the flaw was not in the US known-exploited list. A public proof of concept is on GitHub, but its notes say it suits only demonstrations against a freshly started container on the same machine. Another firm's analysis says the exploit needs about 128 GB of HTTP/2 traffic, took two and a half minutes in its demonstration, would fail on links slower than roughly 300 Mbit per second because of NGINX's default one-hour connection limit, and depends on a memory-copy routine chosen on certain x86-64 processors. Four other flaws fixed in the same releases can also lead to code execution under the same condition. All five score 9.2.

Fixes and mitigations

Fixed versions are 1.31.9 for the mainline branch and 1.30.7 for older branches, and the commercial product has fixed builds except for its older release lines, which list no fix. The official container images carry the fixed tags, but Ubuntu and Debian had no fixed packages yet. After upgrading, NGINX must be fully restarted, because a reload leaves the old executable running. Temporary steps are to turn HTTP/2 off, which blocks two of the five flaws, disable HTTP/3 where used, avoid certain server-side include and mirror settings, and block the kernel module the second step relies on. Signs of attack include workers starting unexpected programs, abnormal exits and long-lived HTTP/2 connections with heavy inbound traffic but few logged requests, and a clean access log does not rule an attack out. The report mentions no India or Southeast Asia link.

Topics