Which NGINX advisories can this checker review?
The NGINX Vulnerability Checker compares the version and configuration from one deployment with two published HTTP/2 advisories. Each advisory gets its own result, so the same inputs can produce different decisions. The table describes the observable evidence the checker uses. The vendor advisory remains the reference for fixes, support status and operational conditions.
| Advisory | Configuration evidence | Official Open Source fixes |
|---|---|---|
| CVE-2026-123007 | Incoming HTTP/2 enablement | 1.30.7 / 1.31.9 |
| CVE-2026-123633 | Incoming HTTP/2 and backend proxying in the same server scope | 1.30.7 / 1.31.9 |
Why the same deployment can get different results
CVE-2026-123007 concerns HTTP/2 flow control. CVE-2026-123633 concerns a worker handling client HTTP/2 and proxying requests to a backend under specific runtime conditions. The checker can recognize a proxy_pass directive in the same server scope, including in a location block. The pasted text cannot establish whether the listener is reachable, what requests it serves, how many worker connections are available or whether exploitation would succeed. A matched indicator is a reason to review the specific advisory and deployed package.
F5 publishes separate NGINX Plus updates for these findings. For CVE-2026-123007, the listed fixes are 37.0.8.1 on the long-term support line and 37.3.0.1 on the continuous release line. For CVE-2026-123633, the corresponding thresholds are 37.0.7.1 and 37.2.0.1. If you need to address both records, choose a supported release that covers both. Use the F5 advisory for older Plus branches and managed products.
Collect the inputs from the running instance
Identify the binary and its package source
Collect nginx -V where the service runs. If your package manager installed one binary and your service launches another, first locate the executable the service uses. In a container deployment, check a running replica rather than the host or a local copy of the image. Record the source of the binary and the exact package release in your maintenance ticket. If replicas are on different versions during a rollout, assess each group separately.
Select an official release only after confirming that the package came from the corresponding upstream channel. A distribution or a hosting vendor can ship a security fix under the same upstream version number. Check its package advisory for the operating system release, package name, repository and complete version, including any suffix. For NGINX Plus, collect the actual Plus release identifier. The Open Source base version in a banner does not identify the Plus update.
Keep the configuration relevant to each listener
Collect the effective configuration from the same instance, usually starting with nginx -T. Remove credentials and unrelated environment details before pasting. Keep the relevant http and server blocks, listener options and any location blocks with proxy directives. If the listener settings live in another file, include the relevant text from that file. The assessment can only read the material you supply.
Check both forms of HTTP/2 enablement when reviewing a mixed fleet. Newer NGINX versions can use http2 on; at HTTP or server scope. Older configurations can use listen 443 ssl http2;. A commented line does not provide enablement evidence. An explicit http2 off; in a server block can override the HTTP-level setting for that server. Also review legacy listener options and shared listeners separately. Pasted excerpts leave the state of omitted files and sockets unresolved.
How to act on the results
Follow the evidence for each advisory
Potentially affected means the version and configuration evidence match the rule. Start with the running package and the listeners used by the services you operate. For a downstream build, take the full package release to the distributor's advisory before deciding whether a backport has already fixed the issue. If the package is affected, identify the supported update for the branch you deploy and plan the rollout.
Matches an upstream fix means the version from a selected official release line meets the fix threshold for that row. Confirm that the running service was restarted or replaced with the updated binary and that old replicas have drained. That outcome is specific to the listed advisory. Assess other packages and modules against their own security updates.
More evidence needed tells you what information is missing or needs independent confirmation. A version alone can be enough to match an official fix. It cannot establish whether HTTP/2 is enabled on an affected release. Similarly, an HTTP/2 directive can identify configuration evidence while the package version and vendor patch remain unknown.
Examples for a reverse proxy and a mixed fleet
Review a reverse proxy with both indicators
Select Load example for an official 1.31.8 configuration with http2 on; and a proxy_pass in a location inside the same server. Both reports can show Potentially affected. If you remove the proxy directive from that example, the flow-control advisory can still match, while the second report requests evidence of backend proxying. This illustrates why configuration scope matters even when the version is identical.
Account for a frontend and its origin separately
A CDN or load balancer can accept HTTP/2 from clients and connect to the origin using HTTP/1.1. Identify the system that terminates HTTP/2 for each hop. Review the NGINX instance on the selected hop using its own configuration, not the edge's protocol indicator. If the origin also accepts HTTP/2 directly on a second listener, gather that listener's evidence too.
When an update will be staged across several hosts, keep a list of the package source, installed release, HTTP/2 settings and owner for each group. Record the specific CVE and vendor update reference in the change ticket. After deployment, compare the running package and image digest again. If you need the impact and temporary mitigation for just one record, use the CVE-2026-123007 bulletin.
Frequently asked questions
Can I check all NGINX vulnerabilities here?
This version covers CVE-2026-123007 and CVE-2026-123633. Review the full NGINX security advisory index for other vulnerabilities. For distro builds, use the distro advisory for the actual package. Updates for NGINX Plus and the NGINX Ingress Controller have their own vendor information.
Why can one advisory match while the other needs more evidence?
The HTTP/2 flow-control advisory can match with HTTP/2 enablement alone. The other advisory also concerns a backend proxy. The report looks for proxy_pass within the same server scope and leaves runtime conditions for the operator to verify. Each advisory has a separate result, even when the version inputs are identical.
What should I do about an Ubuntu or Debian version string?
Check the full package identifier and the OS release against the distributor advisory. A package may carry a backported update even when nginx -V prints an older Open Source version. Record the published package revision and repeat the check on the currently running instance after the update.
Can I use a domain, IP address or URL as an input?
The checker takes deployment evidence collected by the operator: a version, optional build flags and pasted configuration. It does not connect to a target. To establish reachability, consult the applicable listener records, routing rules and deployment evidence for the service you administer.
Does the checker store what I paste?
The checker processes the input in its browser frame. It has no code to upload or persist the inputs or verdicts. Clear removes your input and returns the result to its initial state. Use redacted excerpts where possible. See the Privacy Policy for how the surrounding site handles ordinary page requests.
How do I document the result after an upgrade?
Record the advisory identifier, the published fix threshold, the full running package version and the listener reviewed. If the installation is containerized, also record the running image digest. Confirm that the updated binaries have replaced the affected ones before closing the maintenance ticket.