GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel BPF conntrack kfuncs: racing opts update unbalances the netns reference count

CVE-2026-74715Kernel, userspace & hypervisorcurated

Impact

__bpf_nf_ct_lookup() and __bpf_nf_ct_alloc_entry() read opts->netns_id separately when taking and releasing the namespace reference, and opts can point into a shared map value that another CPU is writing. A concurrent flip either leaks a reference or performs a put_net() with no matching get; repeating the unmatched put destroys a live network namespace and crashes a later user. The kernel report shows a general protection fault with a KASAN null-pointer-deref during a BPF program test run. This lives in the conntrack kfuncs that CNI and network-policy datapaths use, so a fault takes down the node's networking and the node with it.

Who can reach it

Requires a loaded BPF program calling the conntrack kfuncs with opts in a shared map value - CAP_BPF/CAP_NET_ADMIN, i.e. the cluster network datapath or a privileged agent, not a tenant pod. Triggering the imbalance needs concurrent writes to that map value while the kfunc runs.

What to do

Update to a stable kernel carrying the READ_ONCE() snapshot fix and reboot. No user-space knob mitigates this short of not running BPF programs that pass a shared map value as opts, which is a change to the CNI datapath rather than an operator setting. Fold it into the next kernel roll.

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.