[v2] KVM: x86: never write to memory from kvm_vcpu_check_block

[PATCH v2 1/9] KVM: x86: check validity of argument to KVM_SET_MP_STATE

Posted by Paolo Bonzini 3 years, 8 months ago

An invalid argument to KVM_SET_MP_STATE has no effect other than making the
vCPU fail to run at the next KVM_RUN.  Since it is extremely unlikely that
any userspace is relying on it, fail with -EINVAL just like for other
architectures.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch/x86/kvm/x86.c | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 132d662d9713..c44348bb6ef2 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -10653,7 +10653,8 @@ static inline int vcpu_block(struct kvm_vcpu *vcpu)
 	case KVM_MP_STATE_INIT_RECEIVED:
 		break;
 	default:
-		return -EINTR;
+		WARN_ON(1);
+		break;
 	}
 	return 1;
 }
@@ -11094,9 +11095,22 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 
 	vcpu_load(vcpu);
 
-	if (!lapic_in_kernel(vcpu) &&
-	    mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
+	switch (mp_state->mp_state) {
+	case KVM_MP_STATE_UNINITIALIZED:
+	case KVM_MP_STATE_HALTED:
+	case KVM_MP_STATE_AP_RESET_HOLD:
+	case KVM_MP_STATE_INIT_RECEIVED:
+	case KVM_MP_STATE_SIPI_RECEIVED:
+		if (!lapic_in_kernel(vcpu))
+			goto out;
+		break;
+
+	case KVM_MP_STATE_RUNNABLE:
+		break;
+
+	default:
 		goto out;
+	}
 
 	/*
 	 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
-- 
2.31.1

Re: [PATCH v2 1/9] KVM: x86: check validity of argument to KVM_SET_MP_STATE

Posted by Maxim Levitsky 3 years, 8 months ago

On Thu, 2022-08-11 at 17:05 -0400, Paolo Bonzini wrote:
> An invalid argument to KVM_SET_MP_STATE has no effect other than making the
> vCPU fail to run at the next KVM_RUN.  Since it is extremely unlikely that
> any userspace is relying on it, fail with -EINVAL just like for other
> architectures.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  arch/x86/kvm/x86.c | 20 +++++++++++++++++---
>  1 file changed, 17 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 132d662d9713..c44348bb6ef2 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -10653,7 +10653,8 @@ static inline int vcpu_block(struct kvm_vcpu *vcpu)
>         case KVM_MP_STATE_INIT_RECEIVED:
>                 break;
>         default:
> -               return -EINTR;
> +               WARN_ON(1);

Very small nitpick: Maybe WARN_ON_ONCE after all? 
(but I don't see any way after this patch to have invalid mp_state)

> +               break;



>         }
>         return 1;
>  }
> @@ -11094,9 +11095,22 @@ int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
>  
>         vcpu_load(vcpu);
>  
> -       if (!lapic_in_kernel(vcpu) &&
> -           mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
> +       switch (mp_state->mp_state) {
> +       case KVM_MP_STATE_UNINITIALIZED:
> +       case KVM_MP_STATE_HALTED:
> +       case KVM_MP_STATE_AP_RESET_HOLD:
> +       case KVM_MP_STATE_INIT_RECEIVED:
> +       case KVM_MP_STATE_SIPI_RECEIVED:
> +               if (!lapic_in_kernel(vcpu))
> +                       goto out;
> +               break;
> +
> +       case KVM_MP_STATE_RUNNABLE:
> +               break;
> +
> +       default:
>                 goto out;
> +       }
>  
>         /*
>          * KVM_MP_STATE_INIT_RECEIVED means the processor is in



Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>

Best regards,
	Maxim Levitsky

Re: [PATCH v2 1/9] KVM: x86: check validity of argument to KVM_SET_MP_STATE

Posted by Sean Christopherson 3 years, 7 months ago

On Mon, Aug 15, 2022, Maxim Levitsky wrote:
> On Thu, 2022-08-11 at 17:05 -0400, Paolo Bonzini wrote:
> > An invalid argument to KVM_SET_MP_STATE has no effect other than making the
> > vCPU fail to run at the next KVM_RUN.  Since it is extremely unlikely that
> > any userspace is relying on it, fail with -EINVAL just like for other
> > architectures.
> > 
> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > ---
> >  arch/x86/kvm/x86.c | 20 +++++++++++++++++---
> >  1 file changed, 17 insertions(+), 3 deletions(-)
> > 
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index 132d662d9713..c44348bb6ef2 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -10653,7 +10653,8 @@ static inline int vcpu_block(struct kvm_vcpu *vcpu)
> >         case KVM_MP_STATE_INIT_RECEIVED:
> >                 break;
> >         default:
> > -               return -EINTR;
> > +               WARN_ON(1);
> 
> Very small nitpick: Maybe WARN_ON_ONCE after all? 

+1, I don't think warning multiple times would help triage/debug, and this seems
like something that will fire a lot if it fires at all.