GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux KVM x86: kvm_zap_gfn_range races memslot deletion, use-after-free via KVM-VFIO device

UnscoredCVE-2026-99340Kernel, userspace & hypervisorcurated

Impact

kvm_zap_gfn_range() walks memslots and rmaps without holding kvm->srcu and can drop mmu_lock, so a concurrent memslot deletion frees the slot and the zap continues on freed memory. The reported trigger is the KVM-VFIO device path for noncoherent DMA, which is how device passthrough (including GPU passthrough) is set up, and KASAN caught an out-of-bounds read there. Host kernel memory corruption on a passthrough GPU host can crash the node and take down every VM on it. The record does not show whether a guest can drive the race on its own.

Who can reach it

A process holding the VM's KVM file descriptor (the VMM, or a pod given /dev/kvm in KubeVirt/Kata setups) issuing KVM-VFIO device ioctls while memslots change. Local, needs KVM access.

What to do

Install a kernel with the stable fix (commits linked) on hypervisor hosts that use VFIO passthrough, then migrate or drain guests and reboot. The record does not give fixed versions.

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.