GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF: out-of-bounds read copying unaligned cgroup-storage values into a per-CPU map

CVSS 7.1CVE-2026-53076Kernel, userspace & hypervisorcurated

Impact

When a BPF_MAP_TYPE_CGROUP_STORAGE map whose value_size is not a multiple of 8 is used to update a per-CPU map of the same size, pcpu_init_value assumes the source is rounded up to 8 bytes and uses copy_map_value_long, reading past the end of the source. The result is kernel memory disclosure into a map the loader can read back, or a crash. On GPU nodes this matters wherever BPF program loading is available beyond the host admin - observability agents, CNI dataplanes, or a privileged pod with CAP_BPF - because a leak of adjacent kernel memory undermines host/tenant separation on a node that is expensive to drain. Unprivileged users without BPF map-creation rights cannot reach it.

Who can reach it

Local user or container able to create BPF maps and load programs (CAP_BPF / CAP_SYS_ADMIN in the relevant namespace). Not remotely reachable.

What to do

Apply the stable kernel fix and reboot each node. Until then, remove CAP_BPF from workloads that do not need it and keep unprivileged BPF disabled.

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.