Back to posts

azure

What NGINX2Shell and NGINX2Root mean for ingress-nginx on AKS

I’ve worked in environments that use open-source ingress-NGINX in front of Azure Kubernetes Service (AKS). And today, i’m extra happy we made the transition to GatewayAPI months ago. When NGINX shipped security fixes on November 18, 2026, the question for anyone running ingress-nginx was simple: does that NGINX advisory apply to my Kubernetes ingress too? It can. The controller runs a copy of the NGINX web server inside its pods.

Is ingress-nginx the same as the NGINX web server?

Not quite. NGINX Open Source is the web server and reverse proxy. You can run it on a VM or in a container and give it an nginx.conf. A Kubernetes Ingress is a set of routing rules, not a proxy. It can say, for example, that requests for app.example.com should go to a particular Service.

ingress-nginx is the community controller that connects the two. It watches Kubernetes resources and builds the configuration for the NGINX it ships. The controller turns the rules into configuration; the NGINX workers handle the actual requests. A common AKS path is:

Client → Azure Load Balancer → NGINX in the ingress-nginx pod → application pods

That’s why a bug in the NGINX web server can affect an ingress-nginx deployment even if you never installed NGINX separately or wrote an nginx.conf. The CVEs discussed here are in the bundled NGINX server, not in Kubernetes’ Ingress API. And a new NGINX release doesn’t update the copy inside a controller image.

What the November fixes mean for ingress-nginx

Start with the two HTTP/2 fixes. If your ingress-nginx image contains an affected NGINX build and its listener receives HTTP/2, that’s the combination to check first. The other vulnerabilities depend on different features; don’t assume all five apply just because you run the controller.

NGINX2Shell names those two HTTP/2 flaws. NGINX2Root groups them with three others. That’s not the full November 18 release: the mainline changelog lists ten security fixes and the stable changelog lists nine.

  • CVE-2026-123007 is an HTTP/2 flow-control memory-corruption flaw. An HTTP/2 listener is the key prerequisite.
  • CVE-2026-123633 is a separate HTTP/2 use-after-free issue affecting certain reverse-proxy connections.
  • CVE-2026-121810 affects HTTP/3 (QUIC) proxy connections.
  • CVE-2026-123634 requires a particular combination of resolver, proxy, and subrequest configuration.
  • CVE-2026-113425 affects NGINX mail-proxy configurations using auth_http.

F5’s security advisories list the affected ranges. All five have fixes in NGINX Open Source 1.30.7 (stable) and 1.31.9 (mainline). Those are NGINX versions, not ingress-nginx versions. An image publisher can backport a fix without changing the upstream version string, so check what fixes your running image actually contains.

Researchers demonstrated remote code execution for CVE-2026-123007 in a test environment. Depending on the circumstances, an attack can crash a worker or run code with that worker’s permissions. Despite the name NGINX2Root, that doesn’t automatically give anyone root. They’d need another weakness or a worker already running with elevated privileges. CERT-EU’s initial advisory reported no known exploitation in the wild at the time.

Who patches the copy running on AKS?

What handles traffic?Where to verify the fix
Standalone NGINX running in front of AKS or inside itCheck F5’s advisory or the package or image publisher’s confirmed backport.
Self-managed community ingress-nginx, often installed with HelmThere is no official patched community image to upgrade to. The upstream version table lists NGINX 1.27.1 for the final controller release, v1.15.1. That’s in the affected upstream range for CVE-2026-123007; check which image is running and whether HTTP/2 reaches it.
AKS application routing NGINX add-onCheck the running image digest and Microsoft’s guidance for the add-on. The default IngressClass is webapprouting.kubernetes.azure.com, but the add-on can manage additional classes.

Kubernetes retired the community controller in March 2026, not the Ingress API, NGINX Open Source, or other controllers that use NGINX. Retirement didn’t turn off running controllers or remove the old images. I don’t know how many AKS clusters still run ingress-nginx. In January 2026, the Kubernetes Steering and Security Response Committees cited Datadog research estimating that about half of cloud-native environments relied on it. That’s a pre-retirement estimate, not a current count or an AKS figure.

If you installed the community image yourself, reinstalling the chart or upgrading AKS won’t patch the NGINX binary inside it. You can build and test your own patched controller image, or use a maintained vendor build if its publisher confirms the fixes. A custom build means you own the testing and future security updates. It isn’t the same as swapping in a standard NGINX image.

The AKS-managed application routing add-on is different: Microsoft owns its image and documents critical security patch support through November 2026. That support doesn’t cover a community Helm release you installed yourself.

As of November 19, Azure/aks-app-routing-operator#623 reports NGINX 1.27.1 and HTTP/2 enabled on a managed deployment. I didn’t find a public Microsoft advisory confirming affected digests or a patch rollout for these CVEs. The support window isn’t evidence that a specific image has been patched, and the issue isn’t evidence that every cluster is exposed. Verify your image with Azure Support if you can’t establish patch status; Microsoft documents that editing the managed controller ConfigMap is unsupported.

What I’d check first

  1. Inventory every NGINX entry point. Start with kubectl get ingressclass and kubectl get ingress -A. Match the classes to the controllers and record the actual image digest from the running pods, not just the tag in the Deployment. Include private listeners and standalone proxies outside the cluster. Trace where HTTP/2 terminates: an Azure Load Balancer forwards traffic at layer 4, so it doesn’t terminate HTTP/2 for NGINX.
  2. Patch where there is a supported update. For standalone NGINX, use a fixed release or a confirmed vendor backport. For the AKS-managed add-on, follow Microsoft’s release guidance and verify the image running in your pods.
  3. Migrate off the retired community image. There is no official patch to wait for. Keep HTTP/2 from reaching the controller where you can while you move to a maintained controller or Gateway API implementation. Test your existing routes and annotations before moving traffic.
  4. Keep HTTP/2 off the path to unpatched NGINX where possible. Turn it off on a controller you manage, or use a layer 7 proxy that talks HTTP/1.1 to NGINX and block direct access. Check which workloads need native gRPC before making either change.
  5. Watch for updates and investigate anomalies. Prioritize internet-facing listeners, review controller crashes and unexpected process activity, and monitor F5, Microsoft, and your package or image publisher for updated advisories.

Can I turn off HTTP/2 without a WAF?

Migrating to Gateway API isn’t something I’d expect to do in an hour. If you run the community controller yourself, you can turn off HTTP/2 while you work on the migration. In the final community release, v1.15.1, use-http2 defaults to true. If you installed it with Helm, add this to your release values:

controller:
  config:
    use-http2: "false"

If you manage the manifests directly, set use-http2: "false" under data in the ConfigMap named by the controller’s --configmap argument. This is a controller-wide change, not an annotation on a single Ingress. HTTPS still works over HTTP/1.1, but native gRPC on that listener needs HTTP/2. CERT-EU recommends HTTP/2 disablement as a temporary mitigation for the two HTTP/2 CVEs, not the other issues in the release.

What about the AKS-managed ConfigMap?

Microsoft doesn’t support editing the managed ConfigMap, and the documented NginxIngressController settings don’t include an HTTP/2 switch. httpDisabled isn’t that switch: it disables plain HTTP, not HTTP/2 over HTTPS.

Does Front Door or Application Gateway buy you time?

Yes, if all internet traffic reaches ingress-nginx through a proxy that terminates HTTP/2 and uses HTTP/1.1 to talk to NGINX. You don’t need to turn off HTTP/2 for your users or give up HTTPS to the backend. Check what protocol reaches NGINX, not just whether you have a WAF.

What sits in front of ingress-nginx?What to verify
Azure Front DoorRoutes without gRPC enabled use HTTP/1.1 to the origin. On Standard and Premium, routes using the gRPC preview use HTTPS with HTTP/2 on both hops. Don’t count those routes as mitigated if they still deliver HTTP/2 to unpatched NGINX. If native gRPC is required, move it to a patched or maintained gateway rather than forcing HTTP/1.1.
Azure Application GatewayMicrosoft documents that layer 7 HTTP(S) listeners use HTTP/1.1 to backend pools, even when clients use HTTP/2. Don’t assume the same for its layer 4 TCP/TLS proxy, which doesn’t translate HTTP versions, or for Application Gateway for Containers, a separate product with gRPC support.

No custom WAF rule is needed for the HTTP/1.1 path. The rule worth adding is one that prevents clients from reaching the ingress directly. Prefer a private ingress reachable only via the gateway. For Front Door, use Premium Private Link where supported; for a public origin, follow Microsoft’s guidance to restrict network access with AzureFrontDoor.Backend and validate your profile’s X-Azure-FDID at the origin. A header check alone isn’t a network restriction. For Application Gateway, restrict ingress reachability to the gateway’s backend traffic path.

I’d send a normal request through each route and look for HTTP/1.1 in the ingress access logs. A browser or curl pointed at the public hostname only tells you what the edge negotiated. Check whether other public listeners or internal clients can still reach NGINX directly. This buys you time for CVE-2026-123007 and CVE-2026-123633 on the paths you checked. It doesn’t patch the controller or cover the other CVEs.

Migration is more than rewriting the manifests

If you need to keep your Ingress resources, Microsoft lists Application Gateway for Containers, which supports both Ingress and Gateway API. That doesn’t make it a drop-in replacement for ingress-nginx annotations. If you’re moving to the AKS application routing Gateway API implementation, Microsoft documents running the old and new data planes side by side, on separate IPs.

Before moving DNS or changing a Front Door origin, validate certificates, rewrites, auth, client IP handling, and any streaming or gRPC routes you depend on. Check feature parity for controller-specific annotations and snippets. Keep the old path available for a short rollback window, then remove the legacy controller and its listeners after traffic drains. Moving DNS alone doesn’t take an old public endpoint offline.

Further reading