GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF verifier (bpf_loop nr_loops argument type): bpf_loop() declared nr_loops as ARG_ANYTHING, so a

UnscoredCVE-2026-98007Kernel, userspace & hypervisorcurated

Impact

bpf_loop() declared nr_loops as ARG_ANYTHING, so a privileged program could pass a pointer-valued R1 through check_func_arg(). The verifier marks that register precise to bound callback simulation, precision backtracking only accepts scalars, and the mismatch triggers the "backtracking misuse" warning - which on a kernel built or booted with panic_on_warn takes the whole node down. On a GPU node the cost is not the crash itself but the drain: an in-flight training job dies with it and the node has to be rebooted and re-attested before it takes work again. Reachable only by whoever can load BPF programs, so the realistic trigger is a buggy or hostile observability/CNI agent rather than a tenant pod.

Who can reach it

Local, privileged: requires the ability to load a BPF program (CAP_BPF/CAP_SYS_ADMIN in the init namespace). Not reachable by an unprivileged tenant pod without BPF access.

What to do

Update to a stable kernel carrying the fix, which introduces ARG_SCALAR and uses it for bpf_loop()'s nr_loops so the pointer is rejected during generic argument validation. Requires a node reboot, so schedule it with the next kernel maintenance pass. Interim mitigation: do not boot GPU nodes with panic_on_warn, and keep BPF loading restricted to trusted system agents.

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.