GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF verifier: iterator stack slots compared without pointer type, pruning an unsafe path

CVSS 7.8CVE-2026-90217Kernel, userspace & hypervisorcurated

Impact

State pruning treated two iterator stack slots as equivalent without comparing their pointer type, so a MEM_RCU slot and a PTR_UNTRUSTED slot could be considered equal. The verifier then prunes a path it has not actually proven safe, allowing a program that dereferences an untrusted iterator pointer to load. That is a verifier soundness failure: the accepted program runs in kernel context with an unchecked pointer. On GPU nodes the blast radius is the whole host and every co-tenant, since the exposure is wherever BPF program loading is permitted - observability agents, CNI dataplanes, or workloads granted CAP_BPF. NVD scores it local, low complexity, high on confidentiality, integrity and availability.

Who can reach it

Local user or container able to load BPF programs that use iterators, i.e. holding CAP_BPF or root. Unprivileged BPF is disabled by default on most distributions, which closes the path for an ordinary tenant pod.

What to do

Apply the stable kernel update that compares the pointer type for STACK_ITER slots (commits 6424b9cde9ed, 83608e303b95, e4ab436d5ba6) and reboot each node; the verifier ships in the kernel image, so there is no restart-only fix in the record. Keep unprivileged BPF disabled and review CAP_BPF grants until the reboot window. The record names no fixed release versions.

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.