Re: [PATCH 3.2 41/52] KVM: svm: unconditionally intercept #DB

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 4870 bytes --]

On Wed, 2015-11-25 at 12:31 +0100, Paolo Bonzini wrote:
> 
> On 24/11/2015 23:33, Ben Hutchings wrote:
> > 3.2.74-rc1 review patch.  If anyone has any objections, please let me know.
> > 
> > ------------------
> > 
> > From: Paolo Bonzini <pbonzini@redhat.com>
> > 
> > commit cbdb967af3d54993f5814f1cee0ed311a055377d upstream.
> > 
> > This is needed to avoid the possibility that the guest triggers
> > an infinite stream of #DB exceptions (CVE-2015-8104).
> > 
> > VMX is not affected: because it does not save DR6 in the VMCS,
> > it already intercepts #DB unconditionally.
> > 
> > Reported-by: Jan Beulich <jbeulich@suse.com>
> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > [bwh: Backported to 3.2: #DB and #BP did not share a function, and there is
> >  no operation pointer referring to it, so remove update_db_intercept()
> >  entirely]
> 
> This is wrong, you still need to check the BP intercept in the
> (incorrectly named as of 3.2) update_db_intercept function.
> 
> Something like:
> 
> -static void update_db_intercept(struct kvm_vcpu *vcpu)
> +static void update_bp_intercept(struct kvm_vcpu *vcpu)
>  {
>  > 	> struct vcpu_svm *svm = to_svm(vcpu);
> 
> -> 	> clr_exception_intercept(svm, DB_VECTOR);
>  > 	> clr_exception_intercept(svm, BP_VECTOR);
> -
> -> 	> if (svm->nmi_singlestep)
> -> 	> 	> set_exception_intercept(svm, DB_VECTOR);
> -
>  > 	> if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
> -> 	> 	> if (vcpu->guest_debug &
> -> 	> 	>     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
> -> 	> 	> 	> set_exception_intercept(svm, DB_VECTOR);
>  > 	> 	> if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
>  > 	> 	> 	> set_exception_intercept(svm, BP_VECTOR);
>  > 	> } else
> 	> 	> vcpu->guest_debug = 0;
>  }
> 
> 
> Then the calls in db_interception and enable_nmi_window can be removed,
> but the one in svm_guest_debug is important.

Sorry about that.  I now have with this version:

From: Paolo Bonzini <pbonzini@redhat.com>
Date: Tue, 10 Nov 2015 09:14:39 +0100
Subject: KVM: svm: unconditionally intercept #DB

commit cbdb967af3d54993f5814f1cee0ed311a055377d upstream.

This is needed to avoid the possibility that the guest triggers
an infinite stream of #DB exceptions (CVE-2015-8104).

VMX is not affected: because it does not save DR6 in the VMCS,
it already intercepts #DB unconditionally.

Reported-by: Jan Beulich <jbeulich@suse.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
[bwh: Backported to 3.2, with thanks to Paolo:
 - update_db_bp_intercept() was called update_db_intercept()
 - The remaining call is in svm_guest_debug() rather than through svm_x86_ops]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
---
 arch/x86/kvm/svm.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1015,6 +1015,7 @@ static void init_vmcb(struct vcpu_svm *s
 	set_exception_intercept(svm, UD_VECTOR);
 	set_exception_intercept(svm, MC_VECTOR);
 	set_exception_intercept(svm, AC_VECTOR);
+	set_exception_intercept(svm, DB_VECTOR);
 
 	set_intercept(svm, INTERCEPT_INTR);
 	set_intercept(svm, INTERCEPT_NMI);
@@ -1550,20 +1551,13 @@ static void svm_set_segment(struct kvm_v
 	mark_dirty(svm->vmcb, VMCB_SEG);
 }
 
-static void update_db_intercept(struct kvm_vcpu *vcpu)
+static void update_bp_intercept(struct kvm_vcpu *vcpu)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
 
-	clr_exception_intercept(svm, DB_VECTOR);
 	clr_exception_intercept(svm, BP_VECTOR);
 
-	if (svm->nmi_singlestep)
-		set_exception_intercept(svm, DB_VECTOR);
-
 	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
-		if (vcpu->guest_debug &
-		    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
-			set_exception_intercept(svm, DB_VECTOR);
 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
 			set_exception_intercept(svm, BP_VECTOR);
 	} else
@@ -1581,7 +1575,7 @@ static void svm_guest_debug(struct kvm_v
 
 	mark_dirty(svm->vmcb, VMCB_DR);
 
-	update_db_intercept(vcpu);
+	update_bp_intercept(vcpu);
 }
 
 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
@@ -1655,7 +1649,6 @@ static int db_interception(struct vcpu_s
 		if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
 			svm->vmcb->save.rflags &=
 				~(X86_EFLAGS_TF | X86_EFLAGS_RF);
-		update_db_intercept(&svm->vcpu);
 	}
 
 	if (svm->vcpu.guest_debug &
@@ -3557,7 +3550,6 @@ static void enable_nmi_window(struct kvm
 	 */
 	svm->nmi_singlestep = true;
 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
-	update_db_intercept(vcpu);
 }
 
 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)

-- 
Ben Hutchings
This sentence contradicts itself - no actually it doesn't.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 811 bytes --]