Database/Container, Kubernetes & orchestration
BuildKit: build requesting a CDI device panics a daemon started with --cdi-disabled
Impact
A buildkitd started with --cdi-disabled panics when a build tries to use a CDI device. CDI is the mechanism GPUs are handed to containers, so on a GPU build host this is reachable by anything that asks for a GPU in a build - deliberately or by a stale Dockerfile. An authenticated user who can submit builds can take down the shared daemon and every concurrent build on it; the daemon has to be restarted. Impact is availability only, with no disclosure or tampering per the vendor's scoring.
Who can reach it
Any authenticated user who can submit a build to a buildkitd instance running with --cdi-disabled, requesting a CDI device.
What to do
Upgrade BuildKit to v0.33.1 and restart buildkitd. As a mitigation, stop running with --cdi-disabled (leave CDI enabled) or block CDI device requests at the build-submission layer.
References
Related entries
- containerd: malformed subject-annotated manifest digest panics the daemon on image pullCVE-2026-95839 · containerd (transfer service image manifest processing)High
- runc: Volume-mount race gives incorrect access control and privilege escalation to hostCVE-2019-19921 · runcHigh
- Podman: File permissions not checked for non-root users in a privileged containerCVE-2021-20188 · PodmanHigh
- runc: Regression of CVE-2019-19921: incorrect access control leading to privilege escalation via volume mountsCVE-2023-27561 · runcHigh
- Slurm: Filesystem race conditions allow gaining ownership of, overwriting, or deleting filesCVE-2023-41914 · SlurmHigh
- Traefik: ssl-passthrough Ingress on a non-TLS entry point drops all access-control annotationsCVE-2026-114636 · Traefik (Kubernetes Ingress NGINX provider, ssl-passthrough HTTP router)High
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.