Linux Kernel CVE-2025-38369

Requires local low-privileged access to run Intel IDXD (DSA) workqueue workloads inside a container with the host /dev directory mounted, specifically under a Docker configuration where the mount replica is not propagated back to the original mount point. The use-after-free is triggered when the container's parent process is terminated, causing the user driver to detach and destroy_workqueue() to run against an already-destroyed WQ. Limiting factors: it is not remotely reachable (local only), it requires Intel DSA/IDXD accelerator hardware with the idxd wq driver in use, and it depends on a specific non-default Docker mount-propagation condition plus the parent-termination event - so default or hardware-less deployments are not exposed.

Signals are consistent and point to a moderate, non-urgent local issue rather than a widespread threat. The CVSS 3.1 vector (AV:L/AC:L/PR:L/UI:N) describes a local, low-privileged flaw with no remote vector, and the 7.8 base score is driven by full C/I/A impact typical of a kernel use-after-free (CWE-416). The real-world signals temper this: EPSS is very low at 0.19% (8th percentile), there is no CISA KEV listing, and no PoC or exploit reference appears in the input (all four references are patch commits). The vuln.today priority is 'this_week', not act_now/emergency. These agree with each other - a patchable local UAF with no evidence of exploitation. The practical impact described in the source is a call trace or kernel panic (denial of service), and the trigger is tied to a specific Docker mount-propagation misconfiguration plus termination of the container's parent process, which narrows real exposure well below what the raw 7.8 suggests. One tension worth noting: NVD rates C:H/I:H (full confidentiality/integrity impact) but the commit text only describes a crash/panic, so the higher confidentiality/integrity rating is inferred from the UAF class rather than demonstrated.

A local, low-privileged user runs Intel IDXD (DSA) workqueue workloads inside a container that has the host /dev directory mounted. Under a Docker configuration where the mount replica is not propagated back to the original mount point, terminating the container's parent process causes the user driver to detach and the WQ to be destroyed, yet destroy_workqueue() is still called on the freed wq->wq, triggering a use-after-free that produces a kernel call trace or panic. No public exploit is identified at time of analysis; the available references are patch commits only.