[PATCH 15/34] KVM: Add KVM_CREATE_GUEST_MEMFD ioctl() for guest-specific backing memory

[Sean Christopherson]

On Fri, Oct 03, 2025, Nikita Kalyazin wrote:
> On 05/11/2023 16:30, Paolo Bonzini wrote:
> > From: Sean Christopherson <seanjc at google.com>
> > 
> > Introduce an ioctl(), KVM_CREATE_GUEST_MEMFD, to allow creating file-based
> > memory that is tied to a specific KVM virtual machine and whose primary
> > purpose is to serve guest memory.
> 
> ...
> > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > index f1a575d39b3b..8f46d757a2c5 100644
> > --- a/virt/kvm/kvm_main.c
> > +++ b/virt/kvm/kvm_main.c
> 
> ...
> 
> > -static int check_memory_region_flags(const struct kvm_userspace_memory_region2 *mem)
> > +static int check_memory_region_flags(struct kvm *kvm,
> > +				     const struct kvm_userspace_memory_region2 *mem)
> >   {
> >   	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
> > +	if (kvm_arch_has_private_mem(kvm))
> > +		valid_flags |= KVM_MEM_GUEST_MEMFD;
> > +
> > +	/* Dirty logging private memory is not currently supported. */
> > +	if (mem->flags & KVM_MEM_GUEST_MEMFD)
> > +		valid_flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
> 
> I was wondering whether this restriction is still required at this stage or
> can be lifted in cases where the guest memory is accessible by the host.

Off the top of my head, I can't think of any reason why dirty logging wouldn't
work with guest_memfd for non-CoCo VMs.  We'd likely need to explicitly enumerate
support to userspace, and there might be some assumptions lurking in KVM, but
fundamentally it should Just Work (TM).

> Specifically, it would be useful to support differential memory snapshots
> based on dirty page tracking in Firecracker [1] or in live migration.  As an
> experiment, I removed the check and was able to produce a diff snapshot and
> restore a Firecracker VM from it.
> 
> [1] https://github.com/firecracker-microvm/firecracker/blob/main/docs/snapshotting/snapshot-support.md#creating-diff-snapshots
> 
> > +
> >   #ifdef __KVM_HAVE_READONLY_MEM
> >   	valid_flags |= KVM_MEM_READONLY;
> >   #endif
> > @@ -2018,7 +2029,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
> >   	int as_id, id;
> >   	int r;
> > -	r = check_memory_region_flags(mem);
> > +	r = check_memory_region_flags(kvm, mem);
> >   	if (r)
> >   		return r;