Linux Kernel NSM CVE-2026-68179

Exploitation requires: (1) the attacker has code execution inside an AWS Nitro Enclave VM - a strongly bounded execution environment; (2) the running Linux kernel has the NSM miscdevice driver compiled in (CONFIG_NSM) and the /dev/nsm device node is accessible to the process; (3) the kernel version predates the fix commits (pre-6.12.101, pre-6.18.42, pre-7.1.6, or pre-7.2-rc5). No network access is required. The CVSS PR:N metric reflects that within the Nitro Enclave context, NSM device access may not require elevated privileges. Exploitation is limited to: workloads deployed in AWS Nitro Enclaves specifically (not standard EC2 instances), requires local code execution within the enclave boundary, and achieving impact beyond a kernel crash requires further exploitation complexity beyond the documented PoC.

The NVD-assigned CVSS of 8.4 (AV:L/AC:L/PR:N/UI:N/C:H/I:H/A:H) warrants scrutiny. AV:L is correct - the attack requires local execution inside a Nitro Enclave. PR:N is plausible in the Nitro Enclave context where any process running inside the enclave can open /dev/nsm. However, AC:L appears optimistic: while triggering the Lockdep warning is trivial (a single ioctl with an invalid pointer), converting the mutex state corruption into a full C:H/I:H impact beyond a kernel crash requires additional exploitation complexity. EPSS at 0.20% (10th percentile) and the absence of a CISA KEV listing both indicate no observed widespread exploitation. The PoC described in the commit demonstrates a clean crash trigger but not privilege escalation. Real-world impact is most credibly A:H (kernel crash / DoS within the enclave) with C:H/I:H requiring elevated attacker sophistication. The niche deployment context - only AWS Nitro Enclave workloads run this module - substantially narrows the exposed population despite the high severity label. No conflicting vendor data exists; the upstream Linux kernel team is the reporting and patching authority.

An unprivileged process running inside an AWS Nitro Enclave opens /dev/nsm and issues NSM_IOCTL_RAW with a deliberately invalid or unmapped user-space pointer as the argument. The copy_from_user() call fails and control transfers to the shared out label, which calls mutex_unlock() on a mutex that was never locked, corrupting the kernel's locking state. A PoC demonstrating this exact sequence and producing the Lockdep warning is described in the upstream fix commit; reliable escalation beyond a kernel crash would require additional exploitation work.