CVE-2026-100970 in OpenShellinfo

Summary

by MITRE • 11/17/2026

NVIDIA OpenShell for Linux contains a vulnerability in the network policy hot-reload component where a local attacker with low privileges could bypass updated network policy restrictions through existing L7-inspected tunnels that continue to use the previous policy after a reload. A successful exploit of this vulnerability might lead to information disclosure and data tampering.

If you want to get the best quality for vulnerability data then you always have to consider VulDB.

Analysis

by VulDB Data Team • 11/17/2026

The NVIDIA OpenShell for Linux platform, designed to facilitate secure container networking and orchestration within high-performance computing environments, contains a critical flaw in its network policy hot-reload mechanism. This component is responsible for dynamically updating security policies without requiring service restarts or pod rescheduling, a feature essential for maintaining continuous availability while enforcing strict access controls. The vulnerability arises from an inconsistency in how the system handles state transitions during these dynamic updates. Specifically, when new network policies are pushed to the runtime environment, the hot-reload process fails to properly terminate or rebind existing Layer 7 inspected tunnels that were established under previous policy configurations. This architectural oversight creates a window of opportunity where legacy connections remain active and operational despite having been explicitly restricted by the newly applied security rules.

From a technical perspective, this flaw represents a race condition combined with improper state management within the network stack. When an administrator or automated system applies a new set of network policies intended to block specific traffic flows or restrict access to certain resources, the OpenShell daemon updates its internal configuration tables but does not forcefully dismantle existing L7-inspected tunnels that match the old criteria. These tunnels continue to utilize the previous policy context for packet inspection and filtering decisions. Consequently, any data transmitted through these lingering connections bypasses the newly enforced restrictions. This behavior effectively nullifies the intended security controls, allowing traffic that should be denied by the updated policy to pass through unimpeded as if it were still governed by the older, potentially more permissive ruleset.

The operational impact of this vulnerability is significant for organizations relying on NVIDIA OpenShell for strict network segmentation and data isolation. A local attacker with low privileges can exploit this state inconsistency to bypass critical security boundaries. By initiating connections before a policy update and maintaining them through the reload process, an adversary can access resources that should be inaccessible under the new configuration. This capability directly facilitates information disclosure, as sensitive internal services or data stores may become reachable from unauthorized network segments or user namespaces. Furthermore, it enables data tampering, allowing malicious actors to inject or modify traffic destined for protected endpoints without triggering security alerts associated with policy violations. The persistence of these tunnels means that the attack vector remains viable until the affected processes are manually restarted or the system is rebooted, extending the window of exposure significantly beyond the initial moment of exploitation.

This vulnerability aligns closely with CWE-613, which describes insufficient session expiration, and more broadly with CWE-284, improper access control, as it involves a failure to enforce intended security policies due to flawed state management. In terms of offensive tactics, this scenario maps to the MITRE ATT&CK technique T1071, Application Layer Protocol, specifically where an attacker uses existing communication channels to evade detection and policy enforcement. It also relates to T1530, Data from Cloud Storage Object Dumps, if the bypassed policies protect specific data repositories. The core issue is a lack of atomicity in applying network rules across active connections, leading to a temporary but exploitable divergence between configured intent and actual runtime behavior.

To mitigate this risk, immediate action should be taken by NVIDIA users deploying OpenShell for Linux. Administrators must ensure that their environments are updated with the latest patches released by NVIDIA that address this hot-reload inconsistency. Until such updates can be applied in production systems, a recommended workaround involves implementing manual restarts of relevant network services or pods following any significant policy changes to force the teardown and recreation of all tunnels under the new security context. Additionally, monitoring tools should be configured to detect anomalies in tunnel lifecycles that do not align with policy change events. Long-term remediation requires architectural improvements to ensure that hot-reload mechanisms atomically terminate existing connections or seamlessly migrate them to the new policy state without allowing legacy rules to persist on active streams. This ensures that network segmentation remains robust and that security policies are enforced consistently across all traffic flows regardless of their establishment time relative to configuration updates.

Responsible

Nvidia

Reservation

09/27/2026

Disclosure

11/17/2026

Moderation

accepted

CPE

ready

EPSS

0.00000

KEV

no

Activities

very low

Sources

Do you want to use VulDB in your project?

Use the official API to access entries easily!