GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel KVM guest_memfd: memslot bound to guest_memfd before it is ready, racing hole-punch into a NULL rmap

UnscoredCVE-2026-99335Kernel, userspace & hypervisorcurated

Impact

KVM bound a memslot to its guest_memfd before the memslot was fully set up, which made the half-built memslot visible to other code paths. The commit gives one example: on x86 a PUNCH_HOLE on the guest_memfd that lands between the binding and kvm_prepare_memory_region() can see a NULL rmap pointer. A kernel NULL dereference takes down the host and every VM on it. guest_memfd backs confidential VMs (SEV-SNP, TDX), so hosts running confidential GPU tenants are the ones exposed. The record has no score and says nothing about exploitability beyond the race.

Who can reach it

A local process that can create VMs through /dev/kvm and back their memory with guest_memfd (normally the VMM, e.g. QEMU), racing memslot creation against a hole-punch on the same file. Needs local access to /dev/kvm. The record does not describe a trigger from inside a guest.

What to do

Apply the stable kernel fix that creates the guest_memfd binding only after the memslot is prepared. This is a host kernel update, so each node has to be drained of VMs and rebooted. Hosts that do not use guest_memfd-backed (confidential) VMs are not affected by this path. The record has no CVSS score or vendor advisory.

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.