FR
live

ingress-nginx gets a CVE Kubernetes will not fix

On November 5, 2026, the Kubernetes Security Response Committee published CVE-2026-116472, a flaw that lets any user allowed to create an Ingress take down all ingress-nginx traffic through the mirror-target annotation, with no upstream fix because the project’s maintenance ended in March 2026. If your cluster still runs ingress-nginx, plan the move to a maintained controller and tighten your admission rules now.

An Ethernet cable bent into a closed loop and plugged back into its own port on a dark network switch, the connector catching an amber glint.

November 5, 2026. The Kubernetes Security Response Committee publishes CVE-2026-116472, an ingress-nginx flaw that lets any user allowed to create an Ingress cause a denial of service across the entire data plane. March 2026. The ingress-nginx project officially stopped maintenance, an ending announced as early as November 2025 by the SIG Network. 1.15.2-prime1. The only patched version released so far, shipped by SUSE in its fork for Rancher Prime RKE2. Why it matters: eight months after end of life, ingress-nginx gets its first CVE with no upstream fix — and thousands of clusters still run it in production.

The flaw: a mirror that points traffic back at the controller itself

The issue comes down to a missing validation in the mirror-target annotation. That annotation, designed to duplicate an incoming request to a test or observability service, accepts any target — including the controller’s own internal configuration API, which is unauthenticated.

A user with permission to create or modify an Ingress can therefore submit, through this annotation, write commands to that API. By corrupting the data it holds, they take down all traffic passing through that ingress-nginx instance, until a configuration change or a controller restart rebuilds the state correctly.

The flaw carries two classifications: CWE-20 (improper input validation) for the missing target check, and CWE-918 (SSRF) because the annotation acts as an internal bounce vector. In principle, the write access could also redirect traffic from other Ingresses served by the same instance — but the committee judges that risk negligible in practice: it would require detailed knowledge of the controller’s internal state, and a user who can create an Ingress can already redirect traffic through that path.

On the scoring side: CVSS 3.1 of 2.7, rated low — the vector AV:N/AC:L/PR:H/UI:N/S:U/C:N/I:N/A:L describes a network attack, trivial to execute, but requiring high privileges and affecting only availability. Do not use that score to downplay it: in an organization where dozens of teams create Ingresses, the “high privilege” prerequisite is often a formality.

A fix that will never come from upstream

The sentence that changes everything is in the official advisory itself: “Kubernetes will not be issuing a fix for this CVE, because ingress-nginx maintenance ended in March 2026.” The project is end of life, announced by the SIG Network in November 2025 with a best-effort maintenance window running until March 2026. Past that point: no more releases, no bug fixes, no security patches.

The one exception in the ecosystem is SUSE, which maintains its own fork and released version 1.15.2-prime1 for Rancher Prime RKE2. Every other deployment — the community image registry.k8s.io/ingress-nginx/controller, the official Helm chart, and packages shipped by other vendors — remains exposed without a fix.

The flaw was reported by Victor Pereira. It exposes a structural problem the cloud-native world knows well: a component that became a de facto standard by being the “default” controller in so many tutorials and platforms, but whose governance has walked away. The lesson goes beyond ingress-nginx: any component without an active maintainer eventually accumulates CVEs that nobody closes.

Protecting yourself while you migrate

Until you migrate, two official mitigations shrink the exposure window. The first is to reject, via an admission controller, any Ingress carrying the mirror-target annotation — an OPA Gatekeeper or Kyverno rule of a few lines. The second goes through the controller’s ConfigMap: set the annotations-risk-level parameter to “Medium” or “Low” to neutralize risky annotations.

The committee also ships a detection command. A localhost address in the mirror-target annotation is a likely sign of exploitation.

bash
# 1. Check whether ingress-nginx is deployed on the cluster
kubectl get pods --all-namespaces \
  --selector app.kubernetes.io/name=ingress-nginx

# 2. Detect exploitation: mirror-target pointing at localhost
kubectl get ingress --all-namespaces -o json | jq -r '
  .items[]
  | select(.metadata.annotations["nginx.ingress.kubernetes.io/mirror-target"]
      | test("127\\.0\\.0\\.1|localhost|::1"))
  | .metadata.namespace + "/" + .metadata.name'

These measures are band-aids, not cures. They do not protect against the next CVE, and the absence of a maintainer means no patch will ever close it. The only durable fix lies elsewhere.

Moving to a maintained controller: the only durable fix

The official path is Gateway API. Unlike the Ingress API — frozen and tied to aging implementations — Gateway API is an active standard, stewarded by the SIG Network, with living, maintained implementations. Three destinations dominate the migration write-ups:

  • Envoy Gateway — the Gateway API implementation built on Envoy, now the reference migration target (the CNCF published a zero-downtime migration guide from ingress-nginx in May 2026).
  • Traefik — a mature controller with native Gateway API and middleware support, often the fastest replacement for simple cases.
  • NGINX Gateway Fabric — the unofficial successor to ingress-nginx on the F5 NGINX side, also built on Gateway API.

The Ingress2Gateway tool automates part of the conversion of annotations into Gateway API resources — but beware the black box: it converts what it knows how to map, and uncovered annotations are silently dropped. For a multi-tenant fleet, that is exactly where the risk hides. Every mirror, rewrite, or auth annotation has to be verified by hand.

Migration is not a simple image swap. It touches the data plane, TLS, rewrite rules, authentication, and telemetry. Plan a canary or DNS-weight cutover, never a hard switch.

Verdict

CVE-2026-116472 is not an emergency in terms of criticality — it is a signal. An end-of-life component just proved it will never receive another fix, and it is still everywhere.

If you are on Rancher Prime RKE2, upgrade to 1.15.2-prime1 now and stay on the maintained SUSE fork. If you run the community image or the official Helm chart, start the move to Gateway API this quarter — Envoy Gateway if you are starting fresh, NGINX Gateway Fabric if you want to minimize configuration changes. In the meantime, deploy an admission rule that rejects mirror-target and set annotations-risk-level to “Medium” — that closes today’s flaw without waiting. One rule to remember: a component without a maintainer is no longer a piece of infrastructure — it is debt quietly accumulating.

References

The cyber brief, every Tuesday

The flaws that matter and the patches to apply, in a ten-minute read.

No spam. One-click unsubscribe.
read next

On the same topic

← Back to the feed

Type at least two characters.

↑ ↓ navigate ↵ open esc dismiss