NGINX HTTP/2: prove the fix on every instance
Two HTTP/2 flaws require checking the configuration and NGINX version actually handling traffic. Remediation should end with evidence for each instance, including old processes and replicas.
Читати португальськоюBy Pentesty
Identify which conditions exist in your environment
The F5 record for CVE-2026-123007 describes memory corruption and a possible worker crash when HTTP/2 is enabled. Remote code execution depends on ASLR being disabled or bypassable. Your inventory needs to identify reachable endpoints and their effective configuration.
CVE-2026-123633 is separate: it involves HTTP/2 clients, forwarding requests to a backend and a worker running low on available worker_connections. F5 also conditions potential code execution on ASLR. Track the two identifiers separately in the remediation ticket.
Worker credentials depend on configuration. Consequently, code execution in a worker does not by itself establish root privileges on the host. Impact assessment should consider permissions, accessible secrets and separation between processes.
Map exposure by service and owner, including environments serving partners or internal networks. Priorities should reflect observed reachability and the business impact of disruption. This classification guides the maintenance order and the review of exceptions that require a separate remediation decision.
Select the fix by product and vulnerability
The official NGINX list identifies 1.31.9 and 1.30.7 as fixed versions for both flaws. The affected starting points differ: 1.9.5 for CVE-2026-123007 and 1.19.7 for CVE-2026-123633. Do not apply one historical version range to both issues.
For Plus, the official release notes require distinguishing an earlier fix from public disclosure. Use complete build identifiers to select the corresponding release line:
- Open Source: mainline 1.31.9 or stable 1.30.7, released on November 18.
- Plus, CVE-2026-123007: 37.3.0.1 CR or 37.0.8.1 LTS, released on November 18.
- Plus, CVE-2026-123633: 37.2.0.1 CR or 37.0.7.1 LTS, documented on October 14.
- The earlier builds that fixed only CVE-2026-123633 do not automatically close CVE-2026-123007.
Accept the change against the process serving traffic
The control documentation distinguishes configuration reload from executable replacement. Old workers can continue serving existing clients. Installing a package and recording a successful reload leaves a question open: which binary serves each connection?
Define a restart or binary upgrade procedure compatible with service availability. Afterwards, connect each instance, process and executable to the fixed version. For containers, record the deployed digest and confirm replica replacement. These acceptance criteria are an operational proposal from Pentesty grounded in the documented process lifecycle.
For deployments behind a load balancer, plan validation for each pool member while preserving the capacity needed during replacement. Compare the approved inventory against the final observed deployment. If an instance cannot be identified or validated, keep its exception explicit rather than treating the whole service as fixed on the strength of a favorable sample.
Collect evidence that supports closing the ticket
Organize the ticket around three reproducible checks. Preserve timestamps and ownership to connect exposure, change and subsequent validation.
Do not close tracking with a single version screenshot. Collect the instance list, the origin of the approved artifact, test results and decisions about exceptions. That evidence allows another person to check the acceptance criteria and helps prevent a later restoration from reintroducing an old deployment without a fresh assessment of its exposure.
- Exposure: identify where HTTP/2 terminates, which origins are reachable and which proxy configurations meet the conditions for CVE-2026-123633.
- Execution: compare the approved package or image against active processes and confirm that old workers have stopped serving traffic.
- Continuity: validate protocol negotiation, representative responses, errors and capacity after the change. Agree rollback criteria before the maintenance window.
Beyond HTTP/2: check resolver, HTTP/3 and mail-proxy use
Editorial update: November 19, 2026
The F5 records published on November 18 also describe three additional conditions beyond those examined above. CVE-2026-123634 involves a proxy whose upstream name contains variables and an SSI or mirror subrequest resolving that same name. The configuration with mirror_request_body off is one research-tested path, not a universal prerequisite; changing that setting alone does not establish remediation.
CVE-2026-121810 affects HTTP/3 configurations, with additional exploitation conditions beyond the attacker’s control. CVE-2026-113425 requires checking mail-proxy use; the researchers evaluated a configuration with auth_http. For all three, F5 describes memory corruption in the worker and conditions potential code execution on ASLR being disabled or bypassed. Disabling HTTP/2 alone does not assess these other paths.
For these three flaws, the official list identifies Open Source 1.31.9 and 1.30.7 as fixed; the CNA records identify the upper boundaries of the affected Plus ranges as 37.3.0.1 and 37.0.8.1. The October builds cited for CVE-2026-123633 do not close this rollout. Record enabled functions for each service, connect effective configuration to the deployed package and confirm process replacement across every instance.
Xint describes a separate laboratory chain involving the SCTP kernel subsystem to obtain root, including one container setup; the result depends on the environment and does not establish an incident. Keep host verification separate from NGINX patch acceptance. Retest a legitimate operation for each enabled function and record any instances that remain unverified.
Frequently asked questions
Does disabling HTTP/2 equal applying a patch?
They are different actions. Removing the HTTP/2 condition may temporarily reduce exposure to these two flaws; validate the resulting configuration and traffic. It does not demonstrate remediation of other NGINX issues.
Does an installed version prove remediation?
It provides partial evidence. Connect the package or image to the executable used by active processes and check every instance. An old replica still requires attention.
Does the October Plus fix resolve both CVEs?
No. The October builds cited here fix CVE-2026-123633. For CVE-2026-123007, the release notes reviewed identify the November builds listed in this article.
Do worker crashes prove an attack?
Not in isolation. Preserve logs and context for investigation; process failures can have other causes. The technical conditions described by F5 do not constitute proof of exploitation at a particular company.
Sources and references
- F5: CVE-2026-123007 CNA record
- F5: CVE-2026-123633 CNA record
- NGINX: official security advisories
- NGINX Plus: official release notes
- NGINX: worker process credentials
- NGINX: reload and executable upgrade
- The Hacker News: HTTP/2 news report reviewed
- Cibersegurança.PT: NGINX news report reviewed
- NGINX: official 2026 release calendar
Original Pentesty analysis, published November 18, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.