GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF: BPF_PROG_QUERY writes the revision field past a short bpf_attr from userspace

CVSS 7.8CVE-2026-74371Kernel, userspace & hypervisorcurated

Impact

BPF_PROG_QUERY unconditionally writes back query.revision even when userspace supplied a smaller bpf_attr - the 40-byte layout that predates the field - producing an out-of-bounds write into the caller's own memory. The fix propagates the user-supplied attribute size into the cgroup query handlers and skips the write when the buffer is too small. Practical impact on a GPU node is memory corruption in the querying process's address space, so the risk is concentrated where a privileged agent does the querying: cgroup-BPF-based CNI, runtime security tooling and container runtimes that enumerate attached programs. tcx and netkit query paths are not exposed, since revision existed in their bpf_attr from the first release.

Who can reach it

Local user able to call BPF_PROG_QUERY on cgroup attach types, which requires CAP_BPF or root in the relevant namespace. Not network reachable and not reachable from an unprivileged tenant pod without delegated BPF capabilities.

What to do

Pick up the stable-kernel update containing the referenced commits; the fix lands in the kernel, so each node needs a drain and reboot. Nothing can be toggled at runtime to disable the affected path other than withholding CAP_BPF from workloads. The record names only stable git commits, no distribution version.

References

Related entries

All Kernel, userspace & hypervisor 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.