GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF: verifier lets a program dereference a nullable map-iterator pointer, crashing the kernel

CVSS 5.5CVE-2026-43333Kernel, userspace & hypervisorcurated

Impact

check_mem_access() matched PTR_TO_BUF through base_type(), which strips PTR_MAYBE_NULL, so the verifier accepted a direct dereference of a pointer that can legitimately be NULL. BPF map iterator ctx->key and ctx->value are NULL on the stop callback, so a program the verifier approved takes a NULL dereference in kernel context and panics the host. Anyone able to load a map-iterator BPF program - the observability and networking agents that run privileged on every GPU node, and whoever can reach them - can take the whole node down, killing the co-tenant jobs with it. This is a crash, not a privilege escalation: the verifier gap is a missing null check, not a bounds bypass.

Who can reach it

Local, and requires enough privilege to load a BPF map iterator program (CAP_BPF/CAP_SYS_ADMIN in the relevant namespace). The realistic path on a fleet is a privileged DaemonSet agent or an operator-supplied program, not an unprivileged tenant pod.

What to do

Apply the stable kernel update that adds the type_may_be_null() guard to the PTR_TO_BUF branch and reboot each node; the verifier lives in the kernel image, so there is no way to fix it without a reboot. Until then, keep BPF program loading restricted to trusted agents (unprivileged_bpf_disabled, no CAP_BPF for tenant workloads).

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.