When does NGINX HTTP/2 memory corruption become code execution?
When does NGINX HTTP/2 memory corruption become code execution? A memory-safety flaw can be serious without making remote code execution automatic. The useful question for CVE-2026-123633 is which outcome the evidence
When does NGINX HTTP/2 memory corruption become code execution?
A memory-safety flaw can be serious without making remote code execution automatic. The useful question for CVE-2026-123633 is which outcome the evidence supports, on which configuration, and under what additional memory-layout condition.
Vulnerability overview
F5's advisory K000163928 describes CVE-2026-123633 as a use-after-free in ngx_http_v2_module. The vendor classifies it as CWE-416 and assigns both a CVSS v3.1 score of 8.1 and a CVSS v4.0 score of 9.2. The disclosed outcomes are denial of service and possible code execution in an NGINX worker.
The vulnerable behavior occurs when NGINX accepts HTTP/2 client connections and proxies requests to a backend. The worker must also be low on worker_connections. F5 does not describe a universal path that ignores those conditions.
Mechanism and exploitation conditions
Connection draining can close and free a connection while a forwarded request still refers to it. The later use of that expired reference creates the use-after-free. The immediate consequence is memory access against an object whose lifetime has ended.
A crash is the simplest observable result. Code execution needs the attacker to influence memory after the free and place controlled data where the old reference is used. F5 adds a further condition: ASLR must be disabled or bypassed. That statement explains why the same defect can be a denial-of-service issue on one deployment and a code-execution risk on another.
The public vendor material does not include a complete exploit chain. Claims made from the CVSS score alone should be narrowed. A score summarizes severity, while the advisory text defines the supported outcome and prerequisites.
Impact
The availability impact is direct. A failed worker can interrupt active requests and force process replacement. Repeated failures may produce unstable upstream routing and leave gaps in logs.
Code execution would inherit the worker's privileges. That is not automatically root on the host. Container boundaries, worker user settings, seccomp, filesystem permissions, and upstream isolation still determine what an attacker can reach. Those controls do not make the memory flaw safe, but they affect the consequences after compromise.
Organizations should also avoid treating a known ASLR bypass in another component as a separate issue. If an attacker can combine the NGINX flaw with a reliable memory-layout bypass, the lower-probability path becomes more relevant to risk planning.
Affected products and scope
NGINX Open Source 1.19.7 through 1.30.6 and 1.31.0 through 1.31.8 are listed as vulnerable. Fixed releases are 1.30.7 and 1.31.9. NGINX Plus branches have separate vulnerable and fixed point releases in K000163928, including 37.0.0 through 37.0.6.1 to 37.0.7.1 and 37.1.0 through 37.1.1 to 37.2.0.1.
The affected package may be embedded in an ingress controller, WAF bundle, or managed appliance. Confirm the component version from the product's own advisory. Do not infer a fixed status from a recent container tag if the image still contains an older worker binary.
Exposure context
On November 20, 2026, ZoomEye Search app="NGINX" returned 316,211,831 product fingerprints. The CVE Filter vul.cve="CVE-2026-123633" returned 0. The successful Search supplies the Exposure figure, but not a vulnerability verdict.
A useful triage list starts with that inventory, then subtracts assets that do not expose HTTP/2, do not proxy to a backend, run a fixed binary, or cannot reach the required connection-pressure state. Each subtraction needs evidence from configuration or telemetry. Product counts alone cannot support a claim of widespread exploitation.
Remediation and mitigations
The primary action is to install a fixed NGINX release or the vendor's fixed package for the deployed product. Reboot or restart the relevant workers because replacing files without loading the new binary does not remove the vulnerable code.
F5 lists no mitigation. Temporarily disabling HTTP/2 can reduce reachability but is an operational change, not a fix. Capacity increases may make connection pressure harder to reach without addressing the use-after-free.
After upgrading, verify the loaded binary, worker start time, image digest, and package source. Test proxied HTTP/2 traffic and review crash telemetry. Keep the distinction between denial of service and code execution explicit in tickets and executive reporting so the remediation priority reflects the actual deployment conditions.
References
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.