On Thu, Oct 26, 2017 at 3:45 PM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
> On x86, ARM and s390, struct kvm_vcpu_arch has a usercopy region
> taht is read and written by the KVM_GET/SET_CPUID2 ioctls (x86)
> or KVM_GET/SET_ONE_REG (ARM/s390). Without whitelisting the area,
> KVM is completely broken on those architectures with usercopy hardening
> enabled.
>
> For now, allow writing to the entire struct on all architectures.
> The KVM tree will not refine this to an architecture-specific
> subset of struct kvm_vcpu_arch.
>
> Cc: kernel-hardening@xxxxxxxxxxxxxxxxxx
> Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> Cc: Christian Borntraeger <borntraeger@xxxxxxxxxx>
> Cc: Christoffer Dall <cdall@xxxxxxxxxx>
> Cc: Radim Krčmář <rkrcmar@xxxxxxxxxx>
> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> ---
> virt/kvm/kvm_main.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 4d81f6ded88e..b4809ccfdfa1 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -4005,8 +4005,12 @@ int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
> /* A kmem cache lets us meet the alignment requirements of fx_save. */
> if (!vcpu_align)
> vcpu_align = __alignof__(struct kvm_vcpu);
> - kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
> - 0, NULL);
> + kvm_vcpu_cache =
> + kmem_cache_create_usercopy("kvm_vcpu",
> + sizeof(struct kvm_vcpu), vcpu_align,
> + 0, offsetof(struct kvm_vcpu, arch),
> + sizeof_field(struct kvm_vcpu, arch),
> + NULL);
> if (!kvm_vcpu_cache) {
> r = -ENOMEM;
> goto out_free_3;
> --
> 2.14.2
Acked-by: Christoffer Dall <christoffer.dall@xxxxxxxxxx>
