# NGINX HTTP/2 RCE Flaws CVE-2026-123633 and CVE-2026-123007 ("nginx2root" Unauthenticated Remote Code Execution)

> F5/NGINX patched two HTTP/2 memory-corruption flaws, a use-after-free (CVE-2026-123633) and a flow-control integer underflow (CVE-2026-123007), that can let unauthenticated remote attackers execute code as the nginx worker on servers with HTTP/2 enabled. Wiz Research confirmed code execution with an ASLR bypass on official 1.31.7/1.31.8 builds and Rapid7 has published a constrained PoC for CVE-2026-123007; no in-the-wild exploitation has been reported and neither CVE is in CISA KEV as of 2026-11-19.

- **Published:** 2026-11-20T00:03:09Z
- **Last reviewed:** 2026-11-20T14:25:42.018Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6514
- **ID:** TL-2026-6514
- **Severity:** CRITICAL (CVSS 9.2)
- **Category:** VULNERABILITY
- **Status:** PATCHED
- **Detections:** 9 (1 shown free below) · **IOCs:** 14 (1 shown free below)
- **CVEs:** CVE-2026-123633, CVE-2026-123007

## Description

On 2026-11-18 F5 released NGINX Open Source 1.30.7 (stable) and 1.31.9 (mainline), plus NGINX Plus 37.0.7.1/37.0.8.1 and 37.2.0.1/37.3.0.1, fixing ten security issues including two HTTP/2 flaws scored CVSS 4.0 9.2 (CVSS 3.1 8.1 per F5 CNA records as reported by secondary sources). CVE-2026-123633 is a use-after-free (CWE-416 per Equixly) affecting NGINX 1.19.7-1.31.8: an HTTP/2 connection paused mid-HEADERS is marked reusable and stays on the reusable-connection list after the request starts. When worker connection slots run low during upstream proxying (proxy_pass, fastcgi_pass, grpc_pass), ngx_drain_connections() can close that connection and free memory the request still uses; Wiz also describes a second route where GOAWAY frame construction fails and the error path frees the connection. Proxy configurations are the highest risk.

CVE-2026-123007 (nicknamed nginx2root; CWE-191 and CWE-787 per F5/Rapid7/Equixly; CVSS 4.0 vector CVSS:4.0/AV:N/AC:H/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N per Rapid7) affects every release from 1.9.5 (2015, first HTTP/2 release) through 1.31.8. When a stream is cancelled, ngx_http_v2_filter_cleanup() refunds unsent DATA credit to h2c->send_window without checking the HTTP/2 maximum (2^31-1). A client that already restored the same credit with WINDOW_UPDATE can push the window past the maximum; the unsigned size_t bounds check wraps, the window is later cast to a signed off_t producing a negative length, and the TLS send path (ngx_ssl_send_chain) performs a memcpy with an enormous length. On x86-64 glibc the copy terminates after limited out-of-bounds reads and writes, giving heap read/write primitives. Code runs with the nginx worker's privileges. Wiz states attackers must leak memory addresses and groom the heap, and confirmed code execution with an ASLR bypass on official 1.31.7 and 1.31.8 builds serving static content over HTTPS on 64-bit systems.

Five teams independently reported the HTTP/2 flaws (Wiz Research, Ikotas Labs, Metnir, Rapid7, Xint); sources disagree on which team found which bug. Xint chained nginx flaws with a separate unpatched Linux SCTP kernel bug (Linux 4.15+, no CVE named) to get root, including out of a Docker container. Rapid7's public PoC (github.com/sfewer-r7/CVE-2026-123007) targets a fresh linux/amd64 Docker container running nginx 1.31.8 with HTTP/2+TLS, needs roughly 128 GB of HTTP/2 traffic on a single connection (about 2 minutes at 10 Gbit/s, about 17 minutes at 1 Gbit/s, impractical below about 300 Mbit/s because of nginx's default one-hour connection lifetime), depends on the glibc AVX512 memcpy path, and is documented as unreliable against long-running workers, concurrent clients, or proxies that alter HTTP/2 frames. Wiz and Xint are withholding working exploits. CDNs/WAFs that terminate HTTP/2 may break parts of an attack, but origins reachable directly remain exposed. Disabling HTTP/2 mitigates only these two CVEs; the same release fixes CVE-2026-123634 (resolver), CVE-2026-121810 (HTTP/3) and CVE-2026-113425 (mail proxy) that are not mitigated by that workaround.

Exposure is broad: Xint's Shodan snapshot (2026-11-10) found 56.7M nginx hosts with 98.9% of version-reporting hosts in range for CVE-2026-123007, and Censys counted about 8.8M services on an affected version with HTTP/2 enabled on 2026-11-19. Downstream impact includes Ubuntu (USN-9159-1), Debian, Alpine, Amazon Linux, AlmaLinux, Synology (SA-26:20) and the retired community ingress-nginx. Detection is difficult because WINDOW_UPDATE and RST_STREAM control frames are not logged as requests; defenders should watch for nginx worker crashes (signal 11), shells spawned by the worker UID, and long-lived HTTP/2 connections with heavy inbound bytes but few logged requests.

## MITRE ATT&CK

- T1190 Exploit Public-Facing Application
- T1059.004 Unix Shell
- T1611 Escape to Host
- T1068 Exploitation for Privilege Escalation
- T1499.004 Endpoint Denial of Service: Application or System Exploitation

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**Shell Spawned by NGINX Worker UID (auditd) — nginx2root Post-Exploitation** (`TL-2026-6514-SPL-002`)

Detects /bin/sh, bash, dash or similar executed under the nginx worker UID using the auditd execve rule recommended by Xint (key nginx_worker_exec). A shell spawned by an nginx worker is the clearest sign of successful code execution via CVE-2026-123007 or CVE-2026-123633.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1059.004, T1190
- **Data sources:** linux auditd (index=linux, sourcetype=linux:audit), linux_syslog (Sentinel Syslog table with AMA auditd)
- **Severity:** critical
- **Confidence:** high

### Splunk SPL

```spl
index=linux sourcetype=linux:audit earliest=-24h "nginx_worker_exec" type=SYSCALL | rex field=_raw "\bexe=\x22(?<exe>[^\x22]+)\x22" | rex field=_raw "\beuid=(?<euid>\d+)" | where match(exe,"/(sh|bash|dash|ash|zsh|ksh)$") | stats count min(_time) as first_seen max(_time) as last_seen values(exe) as shells by host, euid | convert ctime(first_seen) ctime(last_seen)
```

### Microsoft KQL

```kql
Syslog | where TimeGenerated > ago(24h) | where SyslogMessage has "nginx_worker_exec" | where SyslogMessage matches regex @"exe=""(/usr)?/bin/(sh|bash|dash|ash|zsh|ksh)""" | summarize Events = count(), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated) by Computer | project Computer, Events, FirstSeen, LastSeen
```

### Sigma

```yaml
title: Shell Executed by NGINX Worker UID via auditd (nginx2root)
id: 2b9d4e18-5a7c-4f63-8e02-91c4d6a7f302
status: experimental
description: Detects a shell executed under the nginx worker UID, recorded by the auditd rule 'auditctl -a always,exit -F arch=b64 -S execve,execveat -F euid=NGINX_UID -k nginx_worker_exec'. Indicates likely successful exploitation of CVE-2026-123007 or CVE-2026-123633.
references:
    - https://xint.io/blog/nginx2root-remote-code-execution
    - https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities
author: Threadlinqs Team
date: 2026/11/19
tags:
    - attack.execution
    - attack.t1059.004
    - attack.initial-access
    - attack.t1190
    - cve.2026.123007
    - cve.2026.123633
logsource:
    product: linux
    service: auditd
detection:
    selection:
        type: 'SYSCALL'
        key: 'nginx_worker_exec'
        exe|endswith:
            - '/sh'
            - '/bash'
            - '/dash'
            - '/ash'
            - '/zsh'
            - '/ksh'
    condition: selection
falsepositives:
    - Administrators running nginx with a custom wrapper script that execs a shell
    - Misconfigured auditd rule that matches the nginx master (root) rather than the worker UID
level: critical
```

**Known false positives:**

- Custom nginx deployments that invoke shell scripts from the worker (e.g. fcgiwrap/CGI wrappers running as the worker UID)
- Audit rule scoped to euid=0 rather than the worker UID, which would also match legitimate master-process activity

**Tuning notes:** Requires the auditd rule 'auditctl -a always,exit -F arch=b64 -S execve,execveat -F euid=<nginx worker uid> -k nginx_worker_exec' (events are recorded only after installation). Substitute the real worker UID. Exclude known CGI wrapper binaries by exe path if they are legitimate. Confirm the Sentinel Syslog collection includes the auditd facility.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Long-Lived Heavy-Inbound TLS Flow or Single-Source Short-Connection Burst to HTTPS — CVE-2026-123007 HTTP/2 Window-Underflow Exploitation Traffic (locked: unlock with TL Creds)
- NGINX Worker Spawning Download Tools, Interpreters or Reverse-Shell Commands (locked: unlock with TL Creds)
- Root or Container-Escape Activity in NGINX Process Lineage — nginx2root Privilege Escalation (locked: unlock with TL Creds)
- CVE-2026-123007 Public PoC Tooling Execution (Rapid7 exploit.py / [indicator]) (locked: unlock with TL Creds)
- Shell Process with NGINX as Parent (Process Creation) — nginx2root Code Execution (locked: unlock with TL Creds)
- Outbound Connection to Public IP from Shell or Tool Running as Web Worker UID (locked: unlock with TL Creds)
- NGINX Worker Crash / Connection Exhaustion Burst — CVE-2026-123007 & CVE-2026-123633 (nginx2root) Exploitation Attempt (locked: unlock with TL Creds)
- SCTP Socket Created by NGINX Process — Worker-to-Root Kernel Chain Indicator (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (1 of 14)

Defanged, TLP:CLEAR.

- URL: `hxxps://github[.]com/sfewer-r7/CVE-2026-123007`

Full set of 14 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [NGINX Patches CVE-2026-123633 And CVE-2026-123007 RCE Flaws (The Cyber Express)](https://thecyberexpress.com/cve-2026-123633-cve-2026-123007/)
- [Wiz Research: NGINX HTTP/2 RCE vulnerabilities](https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities)
- [nginx security advisories](https://nginx.org/en/security_advisories.html)
- [F5 K000163926 (CVE-2026-123007)](https://my.f5.com/manage/s/article/K000163926)
- [F5 K000163928 (CVE-2026-123633)](https://my.f5.com/manage/s/article/K000163928)
- [Xint: nginx2root remote code execution](https://xint.io/blog/nginx2root-remote-code-execution)
- [Rapid7: CVE-2026-123007 critical NGINX HTTP/2 integer underflow leads to RCE](https://www.rapid7.com/blog/post/ve-cve-2026-123007-critical-nginx-http-2-integer-underflow-leads-to-rce-fixed/)
- [Rapid7 PoC repository for CVE-2026-123007](https://github.com/sfewer-r7/CVE-2026-123007)
- [The Hacker News: 10-Year-Old NGINX HTTP/2 Flaw Lets Unauthenticated Attackers Execute Code on Servers](https://thehackernews.com/2026/11/10-year-old-nginx-http2-flaw-lets.html)
- [Equixly: nginx2root - Disabling HTTP/2 stops two of five nginx RCE flaws](https://equixly.com/blog/2026/11/18/nginx2root-disabling-http-2-stops-two-of-five-nginx-rce-flaws/)
- [Precursor Security: nginx2root live updates](https://www.precursorsecurity.com/blog/nginx2root-critical-nginx-rce-vulnerabilities)
- [Private DevOps: nginx 1.30.7 and 1.31.9 fix two HTTP/2 code execution bugs](https://privatedevops.com/news/nginx-http2-cve-2026-123007-123633)
- [CVE Record CVE-2026-123007](https://www.cve.org/CVERecord?id=CVE-2026-123007)
- [CVE Record CVE-2026-123633](https://www.cve.org/CVERecord?id=CVE-2026-123633)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple and Gold in full, 2 calls a day on a free account): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6514
