No items found.
November 18, 2026
·
0
Minutes Read

nginx2root: No-auth Nginx + Linux Kernel Root RCE Chain

Advisory
November 18, 2026
·
0
Minutes Read

nginx2root: No-auth Nginx + Linux Kernel Root RCE Chain

Advisory
November 18, 2026
·
0
Minutes Read
Kudelski Security Team
Find out more
table of contents
Share on
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Summary

Publicly disclosed by Xint on November 18, nginx2root is the collective name for six vulnerabilities: five in nginx itself that each allow unauthenticated remote code execution, plus a separate, still-unpatched Linux kernel privilege-escalation bug that can be chained with any of them to obtain root on the host including escaping a Docker container.

Action required! Patch now:

  • nginx: update to 1.31.9 (mainline) or 1.30.7 (stable), or install your vendor's build containing the fixes; NGINX Plus users need 37.3.0.1 or 37.0.8.1.
  • Restart nginx after updating. A configuration reload (nginx -s reload) does not replace the running binary.
  • Linux kernel: a separate update is required and is not yet available; follow your kernel vendor's advisory and use the temporary module-blocking mitigation in the Mitigation section until the patch lands.

Why this is urgent rather than routine:

  • The exposure is old. The most severe flaw has been present in every official nginx release from 1.9.5 (2015) through 1.31.8, including the master branch as of 10 Nov 2026; roughly a decade, not a new bug in a recent release.
  • The bugs are numerous and independent. There are five separate attack paths. Any one of them is sufficient for command execution; they do not need to be combined, and any one of them can be paired with the kernel bug.
  • The impact is not limited to the web process. Worker-level code execution gives access to application data, credentials, TLS keys and internal services; with the kernel bug it becomes full host compromise and escape from containers.
  • Edge protections do not substitute for patching. These flaws live in HTTP/2, HTTP/3, resolver and mail session state machines, so WAF rules that inspect URLs or request bodies cannot reliably block them.
  • No exploitation is known, but the code has been around for years. The researchers report no evidence of in-the-wild exploitation and deliberately withheld their exploit code, which means any third-party "nginx2root exploit" or "scanner" you find online is not from the research team and should be treated as untrusted.

Affected Systems and/or Applications

Products and branches

Product / branch Known vulnerable Fixed in
NGINX Open Source 1.x 1.31.0–1.31.8, and 1.9.5–1.30.6 1.31.9 or 1.30.7
NGINX Open Source 0.x 0.6.18–0.8.55 (resolver path); 0.5.33–0.8.55 (mail path) No fix listed; migrate to a supported branch
NGINX Plus 37.x 37.1.0–37.2.0.1; 37.0.0–37.0.7.1 37.3.0.1 or 37.0.8.1
Linux kernel (SCTP) Depends on kernel build, loaded modules and container restrictions Not currently known; patch pending

How exposure is determined

Affected versions alone do not mean you are exploitable. Each of the five nginx flaws requires a particular feature to be enabled in the running configuration:

Configuration feature in use Reachable path
HTTP/2 enabled on an externally reachable listener CVE-2026-123007 (and CVE-2026-123633 if requests are proxied upstream)
HTTP/3 (QUIC) enabled with requests forwarded to an upstream CVE-2026-121810
resolver + a request-derived upstream name + mirror or another subrequest feature CVE-2026-123634
Mail proxy (proxy on) with auth_http CVE-2026-113425

Practical notes:

  • Distribution packages, appliances and container images frequently backport the fix while retaining an older version string. Confirm patch status from the vendor advisory rather than from nginx -v alone.
  • The affected ranges differ per CVE. Use the nginx.org advisories as authoritative; the researchers' "first affected" figures differ for the resolver and mail paths (they date them from the features they tested, 1.13.4 and 0.5.15) and reflect their specific test configurations rather than the full code history.
  • The kernel SCTP issue is a separate exposure question: on many distributions the module is installed by default and auto-loaded the first time any program opens an SCTP socket, and whether it is reachable depends on the kernel build, available modules and container restrictions.

Technical Details

The nginx vulnerabilities

CVE Component Type Exposure condition Affected (nginx.org) Severity
CVE-2026-123007 ngx_http_v2_module Memory corruption / heap out-of-bounds write in HTTP/2 flow-control handling HTTP/2 enabled 1.9.5–1.31.8 Major
CVE-2026-123633 ngx_http_v2_module Use-after-free in HTTP/2 upstream connection draining HTTP/2 and upstream proxying 1.19.7–1.31.8 Medium
CVE-2026-121810 ngx_http_v3_module Use-after-free in HTTP/3 connection draining HTTP/3 and upstream proxying 1.25.0–1.31.8 Medium
CVE-2026-123634 Resolver Use-after-free during response processing resolver, a request-derived proxy_pass destination, and mirror with mirror_request_body off 0.6.18–1.31.8 Medium
CVE-2026-113425 ngx_mail_proxy_module Use-after-free in the mail proxy auth_http path Mail proxy proxy on + auth_http 0.5.33–1.31.8 Major

Full technical write-ups are available in the references.

The kernel component (host-root stage)

The nginx flaws yield code execution with nginx worker privileges, which is normally constrained. The researchers found a separate zero-day in the Linux kernel's SCTP subsystem that chains with any of them to give host root.

  • It was demonstrated to produce a root shell on the host even when nginx ran inside a Docker container. Because containers share the host kernel, this compromises every container on that host.
  • The researchers are withholding most reproduction details until a kernel patch is available; no CVE identifier and no public patch are currently known, with the fix expected "in the coming weeks".

Exploitation status and publication posture

The Xint researchers validated working exploits against real deployments, including from inside the target server, across private VPC/VNet networks, and over public HTTPS endpoints on AWS, GCP, Azure and bare metal. No exploitation in the wild is reported. Given the age of the affected code, that should be treated as "no evidence of exploitation," not as assurance.

Mitigation

Priority 1: Patch

  1. Install nginx 1.31.9 or 1.30.7, or your vendor's security update; NGINX Plus: 37.3.0.1 or 37.0.8.1.
  2. Restart nginx; do not rely on a reload. nginx -s reload refreshes configuration but leaves the old vulnerable binary running; use your service manager, rolling replacement, or binary upgrade procedure, and confirm the old worker processes have exited.
  3. Confirm patch status at the package level, because distributions and appliances backport fixes without changing the upstream version string.
  4. Patch the kernel separately using your kernel vendor's advisory.
  5. Apply the same release across the estate. Partial upgrades leave the estate's exposed surface unchanged.

Priority 2: Temporary mitigations if you cannot patch immediately

These block the currently known paths; they are not a substitute for the patch.

Mitigation Blocks
Disable HTTP/2 on all affected listeners CVE-2026-123007, CVE-2026-123633
Use fixed upstream groups instead of request-time DNS resolution CVE-2026-123634
Remove QUIC listeners, set http3 off, clear Alt-Svc CVE-2026-121810
Restrict mail proxy to loopback or trusted sources CVE-2026-113425
Force-disconnect existing connections via restart All of the above, for sessions already open
Block SCTP module loading (escalation stage only) Host-root stage, not the nginx CVEs

For further details on these mitigations, see Xint's Emergency Mitigations Before Patching section of their nginx2root blog post.

Detection and incident response

Watch for:

  • Unexpected child processes or command execution spawned by nginx workers.
  • Repeated worker crashes or restarts: several of these paths terminate workers.
  • Unusual outbound network activity from the host.
  • Error logs, process-execution audit records (auditd, EDR) and network captures.

Ordinary access logs cannot rule out exploitation, because HTTP/2 control frames such as WINDOW_UPDATE and RST_STREAM are not fully recorded. Repeated crashes or processes running under the nginx account are not proof of exploitation on their own, but warrant investigation.

If compromise is suspected or confirmed: isolate the host, preserve evidence, inventory everything the compromised worker could read (application data, credentials, TLS keys, internal service tokens), rotate exposed secrets, assess the host and neighbouring containers for the SCTP escalation, and rebuild from a trusted image; patching alone does not undo attacker changes.

What the CFC is Doing

The CFC is monitoring the situation and analyzing the case to identify potential threat-hunting campaigns. This advisory will be updated if required.

References

Related Post