GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF verifier: lost R1 precision on bpf_loop lets unsafe programs pass verification

CVSS 5.5CVE-2026-98084Kernel, userspace & hypervisorcurated

Impact

When the verifier checks a bpf_loop() call a second time, precision backtracking clears the precise mark on the loop-count register. The verifier can then prune state wrongly and accept a BPF program it should reject. The fix commit says this could trick the verifier into accepting unsafe programs, which is the usual first step toward kernel memory corruption from BPF. The record does not describe a working exploit. The risk is highest on hosts where containers or tenant workloads can load BPF programs.

Who can reach it

A local user who can load BPF programs: root, a holder of CAP_BPF, or any user if unprivileged BPF is enabled (most distributions disable it by default). On a GPU node that means privileged pods, or pods granted CAP_BPF for networking or observability agents.

What to do

Update to a stable kernel that carries the referenced fix commits, then drain and reboot each node. Until then, make sure kernel.unprivileged_bpf_disabled is set and do not grant CAP_BPF or CAP_SYS_ADMIN to tenant workloads. The record lists affected releases from 6.7 but does not name fixed 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.