Re: [PATCH] kvm: fix page struct leak in handle_vmon

[David Hildenbrand <david@xxxxxxxxxx>]

Am 25.01.2017 um 10:57 schrieb David Hildenbrand:
> Am 25.01.2017 um 10:52 schrieb Paolo Bonzini:
>>
>>
>> ----- Original Message -----
>>> From: "David Hildenbrand" <david@xxxxxxxxxx>
>>> To: "Paolo Bonzini" <pbonzini@xxxxxxxxxx>, linux-kernel@xxxxxxxxxxxxxxx, kvm@xxxxxxxxxxxxxxx
>>> Cc: dvyukov@xxxxxxxxxx
>>> Sent: Wednesday, January 25, 2017 10:31:13 AM
>>> Subject: Re: [PATCH] kvm: fix page struct leak in handle_vmon
>>>
>>> Am 24.01.2017 um 11:56 schrieb Paolo Bonzini:
>>>> handle_vmon gets a reference on VMXON region page,
>>>> but does not release it. Release the reference.
>>>>
>>>> Found by syzkaller; based on a patch by Dmitry.
>>>>
>>>> Reported-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
>>>> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
>>>> ---
>>>>  arch/x86/kvm/vmx.c | 9 +++++++--
>>>>  1 file changed, 7 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>>>> index 42cc3d6f4d20..0f7345035210 100644
>>>> --- a/arch/x86/kvm/vmx.c
>>>> +++ b/arch/x86/kvm/vmx.c
>>>> @@ -7085,13 +7085,18 @@ static int nested_vmx_check_vmptr(struct kvm_vcpu
>>>> *vcpu, int exit_reason,
>>>>  		}
>>>>  
>>>>  		page = nested_get_page(vcpu, vmptr);
>>>> -		if (page == NULL ||
>>>> -		    *(u32 *)kmap(page) != VMCS12_REVISION) {
>>>> +		if (page == NULL) {
>>>>  			nested_vmx_failInvalid(vcpu);
>>>> +			return kvm_skip_emulated_instruction(vcpu);
>>>> +		}
>>>> +		if (*(u32 *)kmap(page) != VMCS12_REVISION) {
>>>
>>> shouldn't we also check if kmap even returned a valid pointer before
>>> dereferencing it?
>>
>> I don't think kmap can fail (page_address can)?
> 
> Then I wonder why there are some checks:
> 
> e.g. nested_vmx_merge_msr_bitmap()
> 
> msr_bitmap_l1 = (unsigned long *)kmap(page);
> if (!msr_bitmap_l1) {
> 	// no unmap
> 	...
> 	return false;
> 
> or vmx_complete_nested_posted_interrupt()
> 
> vapic_page = kmap(vmx->nested.virtual_apic_page);
> if (!vapic_page) {
> 	// no unmap
> 	...
> 	return -ENOMEM;
> 
> 
> But there is also no check in handle_vmptrld() for example.
> 
> 
>>
>> Paolo
> 
> 

Think you're right it can't fail.

So something like that could most likely be done

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index a236dec..a9be221 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -4973,10 +4973,6 @@ static int
vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
                        return 0;

                vapic_page = kmap(vmx->nested.virtual_apic_page);
-               if (!vapic_page) {
-                       WARN_ON(1);
-                       return -ENOMEM;
-               }
                __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
                kunmap(vmx->nested.virtual_apic_page);

@@ -9730,12 +9726,6 @@ static inline bool
nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
                return false;
        }
        msr_bitmap_l1 = (unsigned long *)kmap(page);
-       if (!msr_bitmap_l1) {
-               nested_release_page_clean(page);
-               WARN_ON(1);
-               return false;
-       }
-
        memset(msr_bitmap_l0, 0xff, PAGE_SIZE);

        if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {

-- 

David