Re: [PATCH v4 08/11] KVM: guest_memfd: Add support for preservation via LUO

Tarun Sahu <[email protected]> writes:

>
> [...snip...]
>
> +static bool kvm_gmem_luo_can_preserve(struct liveupdate_file_handler *handler, struct file *file)
> +{
> +	struct inode *inode = file_inode(file);
> +	struct gmem_file *gmem_file;
> +	struct kvm *kvm;
> +
> +	if (inode->i_sb->s_magic != GUEST_MEMFD_MAGIC)
> +		return false;
> +
> +	gmem_file = file->private_data;
> +	if (!gmem_file)
> +		return false;
> +
> +	/*
> +	 * Only Fully-shared guest_memfd preservation is supported
> +	 */
> +	if (!(GMEM_I(inode)->flags & GUEST_MEMFD_FLAG_INIT_SHARED))
> +		return false;
> +

The above check can probably be removed. I think the core thing about
this series is "we can preserve any memory that doesn't have any
encryption state that goes along with it", and so the check for
kvm_arch_has_private_mem() below would be sufficient to gate that.

> +	/*
> +	 * It makes sure that no memory can be converted to private
> +	 * even if it was initially fully shared (in-place conversions are
> +	 * prevented).
> +	 */
> +	kvm = gmem_file->kvm;
> +	if (kvm_arch_has_private_mem(kvm))
> +		return false;
> +

For now, has_private_mem == has some encryption state that we can't
preserve yet.

This also gates off KVM_X86_SW_PROTECTED_VM, but I guess that is
intended, since when work is done for preserving encryption state, we
can then test using KVM_X86_SW_PROTECTED_VM, and only then, remove this
check?

> +	if (mapping_large_folio_support(inode->i_mapping))
> +		return false;
> +
> +	return true;
> +}
> +
>
> [...snip...]
>