Re: [PATCH] kvm: nVMX: VMWRITE checks VMCS-link pointer before VMCS field

[Nadav Amit <nadav.amit@gmail.com>]

> On Dec 5, 2019, at 1:54 PM, Liran Alon <liran.alon@oracle.com> wrote:
> 
> 
> 
>> On 5 Dec 2019, at 23:30, Jim Mattson <jmattson@google.com> wrote:
>> 
>> On Thu, Dec 5, 2019 at 5:11 AM Jim Mattson <jmattson@google.com> wrote:
>>> On Thu, Dec 5, 2019 at 3:46 AM Paolo Bonzini <pbonzini@redhat.com> wrote:
>>>> On 04/12/19 22:40, Jim Mattson wrote:
>>>>> According to the SDM, a VMWRITE in VMX non-root operation with an
>>>>> invalid VMCS-link pointer results in VMfailInvalid before the validity
>>>>> of the VMCS field in the secondary source operand is checked.
>>>>> 
>>>>> Fixes: 6d894f498f5d1 ("KVM: nVMX: vmread/vmwrite: Use shadow vmcs12 if running L2")
>>>>> Signed-off-by: Jim Mattson <jmattson@google.com>
>>>>> Cc: Liran Alon <liran.alon@oracle.com>
>>>>> ---
>>>>> arch/x86/kvm/vmx/nested.c | 38 +++++++++++++++++++-------------------
>>>>> 1 file changed, 19 insertions(+), 19 deletions(-)
>>>> 
>>>> As Vitaly pointed out, the test must be split in two, like this:
>>> 
>>> Right. Odd that no kvm-unit-tests noticed.
>>> 
>>>> ---------------- 8< -----------------------
>>>> From 3b9d87060e800ffae2bd19da94ede05018066c87 Mon Sep 17 00:00:00 2001
>>>> From: Paolo Bonzini <pbonzini@redhat.com>
>>>> Date: Thu, 5 Dec 2019 12:39:07 +0100
>>>> Subject: [PATCH] kvm: nVMX: VMWRITE checks VMCS-link pointer before VMCS field
>>>> 
>>>> According to the SDM, a VMWRITE in VMX non-root operation with an
>>>> invalid VMCS-link pointer results in VMfailInvalid before the validity
>>>> of the VMCS field in the secondary source operand is checked.
>>>> 
>>>> While cleaning up handle_vmwrite, make the code of handle_vmread look
>>>> the same, too.
>>> 
>>> Okay.
>>> 
>>>> Fixes: 6d894f498f5d1 ("KVM: nVMX: vmread/vmwrite: Use shadow vmcs12 if running L2")
>>>> Signed-off-by: Jim Mattson <jmattson@google.com>
>>>> Cc: Liran Alon <liran.alon@oracle.com>
>>>> 
>>>> diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
>>>> index 4aea7d304beb..c080a879b95d 100644
>>>> --- a/arch/x86/kvm/vmx/nested.c
>>>> +++ b/arch/x86/kvm/vmx/nested.c
>>>> @@ -4767,14 +4767,13 @@ static int handle_vmread(struct kvm_vcpu *vcpu)
>>>>       if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
>>>>               return nested_vmx_failInvalid(vcpu);
>>>> 
>>>> -       if (!is_guest_mode(vcpu))
>>>> -               vmcs12 = get_vmcs12(vcpu);
>>>> -       else {
>>>> +       vmcs12 = get_vmcs12(vcpu);
>>>> +       if (is_guest_mode(vcpu)) {
>>>>               /*
>>>>                * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
>>>>                * to shadowed-field sets the ALU flags for VMfailInvalid.
>>>>                */
>>>> -               if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
>>>> +               if (vmcs12->vmcs_link_pointer == -1ull)
>>>>                       return nested_vmx_failInvalid(vcpu);
>>>>               vmcs12 = get_shadow_vmcs12(vcpu);
>>>>       }
>>>> @@ -4878,8 +4877,19 @@ static int handle_vmwrite(struct kvm_vcpu *vcpu)
>>>>               }
>>>>       }
>>>> 
>>>> +       vmcs12 = get_vmcs12(vcpu);
>>>> +       if (is_guest_mode(vcpu)) {
>>>> +               /*
>>>> +                * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
>>>> +                * to shadowed-field sets the ALU flags for VMfailInvalid.
>>>> +                */
>>>> +               if (vmcs12->vmcs_link_pointer == -1ull)
>>>> +                       return nested_vmx_failInvalid(vcpu);
>>>> +               vmcs12 = get_shadow_vmcs12(vcpu);
>>>> +       }
>>>> 
>>>>       field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
>>>> +
>>>>       /*
>>>>        * If the vCPU supports "VMWRITE to any supported field in the
>>>>        * VMCS," then the "read-only" fields are actually read/write.
>>>> @@ -4889,24 +4899,12 @@ static int handle_vmwrite(struct kvm_vcpu *vcpu)
>>>>               return nested_vmx_failValid(vcpu,
>>>>                       VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
>>>> 
>>>> -       if (!is_guest_mode(vcpu)) {
>>>> -               vmcs12 = get_vmcs12(vcpu);
>>>> -
>>>> -               /*
>>>> -                * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
>>>> -                * vmcs12, else we may crush a field or consume a stale value.
>>>> -                */
>>>> -               if (!is_shadow_field_rw(field))
>>>> -                       copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
>>>> -       } else {
>>>> -               /*
>>>> -                * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
>>>> -                * to shadowed-field sets the ALU flags for VMfailInvalid.
>>>> -                */
>>>> -               if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
>>>> -                       return nested_vmx_failInvalid(vcpu);
>>>> -               vmcs12 = get_shadow_vmcs12(vcpu);
>>>> -       }
>>>> +       /*
>>>> +        * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
>>>> +        * vmcs12, else we may crush a field or consume a stale value.
>>>> +        */
>>>> +       if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
>>>> +               copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
>>>> 
>>>>       offset = vmcs_field_to_offset(field);
>>>>       if (offset < 0)
>>>> 
>>>> 
>>>> ... and also, do you have a matching kvm-unit-tests patch?
>>> 
>>> I'll put one together, along with a test that shows the current
>>> priority inversion between read-only and unsupported VMCS fields.
>> 
>> I can't figure out how to clear IA32_VMX_MISC[bit 29] in qemu, so I'm
>> going to add the test to tools/testing/selftests/kvm instead.
> 
> Please don’t.
> 
> I wish that we keep clear separation between kvm-unit-tests and self-tests.
> In the sense that kvm-unit-tests tests for correct CPU behaviour semantics
> and self-tests tests for correctness of KVM userspace API.
> 
> In the future, I wish to change kvm-unit-tests to cpu-unit-tests. As there is no
> real connection to KVM. It’s a bunch of tests that can be run on top of any CPU
> Implementation (weather vCPU by some hypervisor or bare-metal CPU) and
> test for it’s semantics.
> I have already used this to find semantic issues on Hyper-V vCPU implementation for example.

Did you use for the matter the “infrastructure” that I added?