GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF cgroup storage: bogus pointer read and key copied to userspace at end of list

CVSS 5.5CVE-2026-45838Kernel, userspace & hypervisorcurated

Impact

list_next_entry() wraps to the list head instead of returning NULL, so the dead NULL check meant to end iteration never fires. Walking a cgroup storage map to its last element makes the kernel read storage->key from a pointer that aliases internal map fields and copy that result out to userspace. In practice this exposes a few bytes of kernel map internals to whoever can iterate the map, and can corrupt the iteration itself. On a GPU fleet the realistic reacher is the node's CNI or observability stack (Cilium-style BPF cgroup maps) rather than a tenant pod, so treat it as a hardening fix unless tenants are granted BPF.

Who can reach it

Local, requires the ability to call BPF_MAP_GET_NEXT_KEY on a cgroup storage map - normally CAP_BPF/CAP_SYS_ADMIN, i.e. a privileged node agent or a container granted those capabilities. Not remotely reachable.

What to do

Take the fixed stable kernel (five stable commits linked, the fix switches to list_entry_is_head()) and reboot the node on the next kernel pass. No runtime mitigation other than not handing CAP_BPF to 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.