CVE-2026-123634 decision guide for reverse-proxy operators
CVE-2026-123634 decision guide for reverse-proxy operators The right response to CVE-2026-123634 depends on whether an NGINX deployment performs request-time DNS resolution and creates a compatible subrequest. Version
CVE-2026-123634 decision guide for reverse-proxy operators
The right response to CVE-2026-123634 depends on whether an NGINX deployment performs request-time DNS resolution and creates a compatible subrequest. Version alone does not answer that question. A short decision path can separate systems that need immediate patch verification from those with an additional emergency configuration problem.
Vulnerability overview
Official sources identify a resolver use-after-free in NGINX Open Source 0.6.18 through 1.31.8, fixed in 1.31.9 and 1.30.7. CVE-2026-123634 has a CVSS v4.0 score of 9.2 and CWE classification CWE-416. The potential impact includes worker crashes and remote code execution with worker privileges.
The flaw is separate from the HTTP/2 and HTTP/3 CVEs announced at the same time. It also differs operationally: request filtering for HTTP/2 cannot reduce this specific path.
Mechanism and exploitation conditions
Nginx uses resolver contexts for asynchronous DNS lookups. A main request and a mirror subrequest can wait on the same name. When an error path terminates the main request, cleanup can release the mirror context that the resolver still plans to process. The resumption can write through and call through freed memory.
The attacker needs a reachable route with a request-derived upstream name, resolver use, and a compatible subrequest. Xint's tested setup used mirror with mirror_request_body off; SSI may provide another subrequest route. The attacker-controlled DNS answer path is part of the tested impact chain. Without those conditions, the server may still be unpatched but the demonstrated path is not present.
Impact
The most urgent scenario is an internet-facing reverse proxy that accepts client-controlled hostnames and mirrors requests. In that design, the two sides of the trigger can be reached without authentication. A worker compromise can expose upstream secrets and internal network access.
The second scenario is a fixed-upstream service with no dynamic resolver use. It still needs patch tracking, but its immediate configuration risk is lower. The third scenario is uncertainty: the team knows the version but cannot prove how the configuration is generated. Treat uncertainty as a reason to inspect the deployment, not as evidence that it is safe.
Affected products and scope
Use 1.31.9, 1.30.7, or a later maintained release as the upstream fix baseline. Confirm distribution and appliance backports with the responsible vendor. Check all configuration sources, including generated routes, container images, sidecars, and test services that receive mirrored production traffic.
The scope review should answer four questions:
- Is the running build vulnerable after accounting for backports?
- Does any externally reachable route use request-time DNS resolution?
- Can a client influence the name or the resolver response?
- Does mirror, SSI, or another subrequest create a same-name waiting list?
Exposure context
ZoomEye Search app="NGINX" returned 316,220,816 product fingerprints. The CVE Filter returned 0. Exposure is 316,220,816 under the selected-query maximum rule. This number shows broad NGINX product presence, not confirmed vulnerable endpoints. Configuration and DNS egress evidence are required for a defensible risk decision.
Remediation and mitigations
Patch first. If a fixed release cannot be deployed immediately, replace request-time DNS names with fixed upstream groups for the main route and the mirror. Remove mirroring when it has no clear owner or purpose. Do not use HTTP/2 shutdown as coverage for CVE-2026-123634. Do not rely on mirror_request_body on as a standalone fix.
After a change, run configuration validation, reload or restart, and confirm that old workers have exited. Test routing, resolver-independent health checks, TLS termination, upstream failure responses, and mirror behavior. Record the result as package evidence and configuration evidence, with a named owner and a date for rechecking. This closes the operational question without overstating what a public product count can prove.
References
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.