KVM: nVMX: Fix kernel info-leak when enabling KVM_CAP_HYPERV_ENLIGHTENED_VMCS more...
authorLiran Alon <liran.alon@oracle.com>
Sat, 17 Nov 2018 12:05:06 +0000 (14:05 +0200)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 27 Nov 2018 11:49:46 +0000 (12:49 +0100)
Consider the case that userspace enables KVM_CAP_HYPERV_ENLIGHTENED_VMCS twice:
1) kvm_vcpu_ioctl_enable_cap() is called to enable
KVM_CAP_HYPERV_ENLIGHTENED_VMCS which calls nested_enable_evmcs().
2) nested_enable_evmcs() sets enlightened_vmcs_enabled to true and fills
vmcs_version which is then copied to userspace.
3) kvm_vcpu_ioctl_enable_cap() is called again to enable
KVM_CAP_HYPERV_ENLIGHTENED_VMCS which calls nested_enable_evmcs().
4) This time nested_enable_evmcs() just returns 0 as
enlightened_vmcs_enabled is already true. *Without filling
vmcs_version*.
5) kvm_vcpu_ioctl_enable_cap() continues as usual and copies
*uninitialized* vmcs_version to userspace which leads to kernel info-leak.

Fix this issue by simply changing nested_enable_evmcs() to always fill
vmcs_version output argument. Even when enlightened_vmcs_enabled is
already set to true.

Note that SVM's nested_enable_evmcs() should not be modified because it
always returns a non-zero value (-ENODEV) which results in
kvm_vcpu_ioctl_enable_cap() skipping the copy of vmcs_version to
userspace (as it should).

Fixes: 57b119da3594 ("KVM: nVMX: add KVM_CAP_HYPERV_ENLIGHTENED_VMCS capability")
Reported-by: syzbot+cfbc368e283d381f8cef@syzkaller.appspotmail.com
Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index be6f13f..065f1df 100644 (file)
@@ -1611,12 +1611,6 @@ static int nested_enable_evmcs(struct kvm_vcpu *vcpu,
 {
        struct vcpu_vmx *vmx = to_vmx(vcpu);
 
-       /* We don't support disabling the feature for simplicity. */
-       if (vmx->nested.enlightened_vmcs_enabled)
-               return 0;
-
-       vmx->nested.enlightened_vmcs_enabled = true;
-
        /*
         * vmcs_version represents the range of supported Enlightened VMCS
         * versions: lower 8 bits is the minimal version, higher 8 bits is the
@@ -1626,6 +1620,12 @@ static int nested_enable_evmcs(struct kvm_vcpu *vcpu,
        if (vmcs_version)
                *vmcs_version = (KVM_EVMCS_VERSION << 8) | 1;
 
+       /* We don't support disabling the feature for simplicity. */
+       if (vmx->nested.enlightened_vmcs_enabled)
+               return 0;
+
+       vmx->nested.enlightened_vmcs_enabled = true;
+
        vmx->nested.msrs.pinbased_ctls_high &= ~EVMCS1_UNSUPPORTED_PINCTRL;
        vmx->nested.msrs.entry_ctls_high &= ~EVMCS1_UNSUPPORTED_VMENTRY_CTRL;
        vmx->nested.msrs.exit_ctls_high &= ~EVMCS1_UNSUPPORTED_VMEXIT_CTRL;