CVE-2026-123007 and CVE-2026-123633: Your BIG-IP May Not Be Vulnerable. What About the NGINX Behind It?

Based on advisories available on November 18, 2026. Product assessments may change as investigations continue.

The first question many administrators ask about NGINX2Root is: “Does this affect my F5?”

Answering that requires identifying the product, its installed version, and which component actually processes the vulnerable protocol.

A BIG-IP in front of an application and an NGINX instance running as a reverse proxy or ingress are separate components. The security status of one does not determine the status of the other.

For the two HTTP/2 vulnerabilities, CVE-2026-123007 and CVE-2026-123633, F5 lists BIG-IP, all modules, as not vulnerable within the evaluated software scope. The same advisories identify vulnerable configurations of NGINX Open Source and NGINX Plus.

What NGINX2Root actually means

NGINX2Root is the name given by Theori’s Xint team to research involving five NGINX vulnerabilities and a separate Linux kernel vulnerability.

These represent two distinct stages:

  • The NGINX vulnerabilities can allow code execution with NGINX worker privileges.
  • Obtaining root on the host requires an additional privilege-escalation step and the conditions needed to exploit it.

A vulnerable NGINX instance therefore does not automatically mean direct execution as root. Compromising the worker is still serious: it can expose data, credentials, and internal services accessible to that process.

The five CVEs have different prerequisites

CVE Component Configuration or condition to investigate
CVE-2026-123007 HTTP/2 flow control A vulnerable NGINX instance accepting HTTP/2 connections
CVE-2026-123633 HTTP/2 Client HTTP/2 connections proxied to a backend under conditions where available worker_connections are running low
CVE-2026-121810 HTTP/3 HTTP/3 enabled, with additional runtime conditions
CVE-2026-123634 Resolver Proxy configurations using variables in the upstream server name, with SSI or mirror subrequests resolving the same name
CVE-2026-113425 Mail proxy Use of the vulnerable mail proxy module

The mechanisms include memory corruption and use-after-free. Outcomes can include worker termination, denial of service, and code execution.

F5’s advisories describe ASLR being disabled or bypassed as a condition for code execution. Having ASLR enabled should not be treated as evidence that the deployment is protected.

For HTTP/2, the assessment must cover protocol processing, not just URLs, parameters, and application content. For the resolver vulnerability, finding a resolver directive alone does not establish exposure: the configuration must meet the conditions described in the advisory.

How to interpret the impact on F5 products

The CVE-2026-123007 advisory lists the following assessment:

Product or service Status in the reviewed advisory
NGINX Open Source and NGINX Plus Vulnerable releases identified; exposure depends on configuration
BIG-IP, all modules Not vulnerable
BIG-IQ Centralized Management Not vulnerable
F5OS, F5OS-A, and F5OS-C Not vulnerable
BIG-IP CNF and BIG-IP Next for Kubernetes Not vulnerable
F5 AI Gateway Not vulnerable
F5 Distributed Cloud and NGINX One Console Under investigation

This table applies to the specified CVE, within F5’s evaluated software scope. It is not a blanket statement covering every CVE or unsupported release.

For F5 NGINX Ingress Controller, there is a specific update to review: the 2026-lts-r11 changelog, published on November 18, updates its NGINX Plus dependency to 37.0.8.1. Check the image actually running in your pods, not only the version declared in your deployment repository.

Also distinguish F5 NGINX Ingress Controller from other ingress projects that use NGINX. Each distribution has its own update process.

A BIG-IP in front changes the traffic-path assessment

Consider three deployment patterns:

Architecture What to verify
TLS passthrough to NGINX NGINX may still receive the HTTP/2 connection negotiated directly with the client
HTTP/2 terminates on BIG-IP; the backend receives HTTP/1.1 This path does not deliver the client’s HTTP/2 frames to NGINX; other CVEs and access paths still require assessment
HTTP/2 is also used between the proxy and NGINX Assess that second connection and how incoming traffic can influence it

This is architectural reasoning, not a claim of an F5-validated mitigation.

Check alternative paths as well: a publicly reachable origin, additional ports, IPv6, internal access, and Kubernetes services exposed outside the intended ingress path.

Testing the VIP does not automatically characterize the origin. Likewise, the presence of a WAF does not establish protection against every memory-corruption condition in a downstream server.

Which versions contain the fixes?

For the five CVEs grouped under NGINX2Root, the official NGINX advisory index identifies fixes in Open Source 1.30.7 stable and 1.31.9 mainline.

Distribution Update reference
Upstream NGINX Open Source 1.30.7 stable or 1.31.9 mainline
NGINX Plus LTS track PLS.37.0.8.1
NGINX Plus Continuous Release track PLS.37.3.0.1
Linux distribution package Fixed package build identified in the distribution’s advisory
Container or product embedding NGINX Updated image or release provided by its maintainer

The listed Plus releases incorporate the relevant fixes. One detail matters when reviewing patch coverage: the changelog records the CVE-2026-123633 fix in the earlier 37.0.7.1 LTS and 37.2.0.1 CR releases.

Being patched for that CVE does not mean the installation is patched for the entire set.

Linux distributions may also backport fixes while retaining an older upstream version number. Compare the full package identifier against the relevant security advisory. AlmaLinux, for example, published fixed packages for maintained streams and explicitly distinguished retired streams.

A practical remediation workflow

I would organize the change into four stages.

1. Identify the process actually exposed

On directly managed NGINX installations, these commands help with inventory:

nginx -v
nginx -V
nginx -T

Confirm that the binary you query corresponds to the instance handling traffic. In container environments, run the checks against the correct workload.

The output of nginx -T may contain sensitive information. Keep configuration evidence internally and remove secrets before sharing it.

2. Select the correct package or image

Record the package source, full version, architecture, additional modules, and supported upgrade path.

If remediation requires moving to another NGINX Plus release track or generation, review the upgrade notes. The documentation identifies behavior changes involving upstream keepalive, the HTTP version used for upstream communication, and shared-memory-zone sizing.

A security update may require functional validation beyond a configuration syntax check.

3. Verify that the fixed version is running

Installing a package does not prove that old processes have stopped handling connections.

Follow the vendor’s update procedure and confirm that the relevant processes or pods have been replaced. A configuration reload alone should not be accepted as proof that the vulnerable executable has been replaced.

4. Validate service availability and patch coverage

Record:

  • The installed package version or deployed image digest.
  • The instances and replicas actually replaced.
  • Application test results.
  • Protocols still enabled on each listener.
  • Outstanding exceptions, their owners, and remediation deadlines.

Someone else should be able to verify the evidence used to close the change.

If you cannot patch immediately

Xint describes disabling HTTP/2 as a temporary measure for the two HTTP/2 vulnerabilities, provided the change covers the relevant listeners and existing HTTP/2 connections are no longer being processed.

This does not address the HTTP/3, resolver, or mail proxy vulnerabilities.

Assess application dependencies such as gRPC before applying that measure.

For CVE-2026-123634, F5 recommends avoiding SSI or mirror directives and avoiding variables in proxied server names. Evaluate the functional impact before changing those configurations.

Increasing worker_connections should not be treated as a fix for the underlying memory-safety issue. Nor should a mitigation for one CVE become a statement that “NGINX2Root is resolved.”

Patching does not close a compromise investigation

According to the researchers, ordinary access logs do not fully record the HTTP/2 control frames involved. The absence of suspicious requests in those logs does not rule out exploitation.

Correlate worker failures with process execution, outbound connections, and host telemetry. A crash alone does not establish an attack.

If compromise occurred, patching does not remove persistence or revoke exposed credentials. The separate kernel vulnerability described in the chain also requires its own assessment and remediation.

Where does HTTP/2 actually terminate in your environment?

On BIG-IP, on the ingress controller, or on the application’s NGINX instance?

That answer changes prioritization. Share your architecture and the main obstacle you encountered during remediation: gRPC dependencies, third-party images, a Plus release migration, or difficulty finding every running instance.

Those operational details can help another administrator make the right change before their maintenance window closes.

References