GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel sockmap: 32-bit wrap in bpf_msg_pop_data bounds check runs the pop loop off the scatterlist

CVSS 8.4CVE-2026-74256Kernel, userspace & hypervisorcurated

Impact

bpf_msg_pop_data() computed start + len in 32-bit before storing it in a u64, so a len near U32_MAX wrapped and passed the bounds check. The pop loop then ran past the end of the scatterlist and sk_msg_shift_left() called put_page() on the empty msg->sg.end slot, which the reporter's trace shows as a general protection fault with a KASAN null-pointer-deref range. The reachable outcome is a host kernel crash, with refcount manipulation on an unowned slot as the worse case. Same subsystem as the two bpf_msg_push_data bugs but a different helper, a different check and a different fix, so it must be patched on its own. For a GPU cluster whose networking datapath is BPF-based, this is an availability risk on nodes that cannot be drained cheaply.

Who can reach it

Local: requires the ability to attach a BPF sk_msg program to a socket (CAP_BPF/CAP_NET_ADMIN, or permissive unprivileged BPF) and then send through it. No network-only path.

What to do

Apply the stable kernel update that widens the addition to 64-bit and makes the pop counter unsigned (five stable branches carry the fix; see the git.kernel.org commits), then reboot each affected node. Interim exposure control is the same as for the other sockmap bugs: no unprivileged BPF, no CAP_BPF in 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.