GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF verifier: narrow read of a pointer context field triggers a verifier bug warning

CVSS 5.5CVE-2025-38591Kernel, userspace & hypervisorcurated

Impact

The verifier allowed a narrow load of a pointer field in a BPF context struct (for example sk at offset 168 in __sk_buff), which then failed during context access conversion and raised a kernel warning instead of a clean rejection. On a node with panic_on_warn set, or with kernel warnings treated as fatal by monitoring, this turns into a node-level availability event; otherwise it is a verifier bug surfaced as noise. It matters on GPU fleets because BPF program loading is widely delegated - CNI plugins, service meshes and observability agents all load programs - and unprivileged BPF or a CAP_BPF-holding sidecar widens who can reach it. The record rates it local availability-only (CVSS 5.5) with no memory-corruption claim.

Who can reach it

A local user or workload able to load a BPF program - which in practice means root, CAP_BPF, or a pod granted those capabilities. Not reachable from an unprivileged pod on a fleet that keeps unprivileged BPF disabled.

What to do

Pick up a stable kernel with the linked fix commits and reboot the nodes; drain GPU workloads first. Until then, keep kernel.unprivileged_bpf_disabled=1 and avoid granting CAP_BPF to tenant workloads, which removes the reachable path for most fleets.

References

https://nvd.nist.gov/vuln/detail/CVE-2025-38591https://git.kernel.org/stable/c/058a0da4f6d916a79b693384111bb80a90d73763https://git.kernel.org/stable/c/202900ceeef67458c964c2af6e1427c8e533ea7c

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.