GPU VulnDB

Database/Container, Kubernetes & orchestration

containers/storage: leaked file descriptor on corrupt layers can exhaust a container runtime's fd limit

CVSS 3.3CVE-2026-114236Container, Kubernetes & orchestrationcurated

Impact

ApplyLayer and ApplyUncompressedLayer in containers/storage leak a file descriptor when handed a corrupt or malicious layer tarball. Repeated unbounded calls exhaust the process's file descriptor limit, which for Podman, Buildah, CRI-O or any image-pull path built on this library means the component stops being able to open files - no further image pulls or container starts on that node until it is restarted. On a GPU node this is a self-inflicted-by-pull denial of service rather than a path to code execution or data access: an operator sees pods stuck in image-pull or container-create on one node. The record scores it 3.3, local, availability-only, and needing user interaction, i.e. somebody has to pull or apply the hostile image.

Who can reach it

Local, via whatever can cause layers to be applied on the node - a user or workload that can trigger an image pull or layer unpack of an image they control. Needs the hostile layer to be pulled repeatedly to reach fd exhaustion; no authentication to the library itself.

What to do

Update the containers/storage-derived packages on affected hosts and in affected container images and restart the runtime that embeds them (Podman/Buildah/CRI-O); Red Hat tracks the per-product fixed versions at access.redhat.com/security/cve/CVE-2026-114236 - this record names no version. Ordinary runtime patching and daemon restart, not a node drain. Restricting which registries nodes may pull from limits who can feed in corrupt layers in the meantime. Note the Red Hat product list includes NVIDIA-adjacent components (DPU kit for NVIDIA, JetPack 6 for RHEL) only because they ship the library downstream; the flaw is in containers/storage.

References

Related entries

All Container, Kubernetes & orchestration entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.