On 2014-01-03 20:36, Marcelo Tosatti wrote:
> On Fri, Jan 03, 2014 at 08:27:07PM +0100, Jan Kiszka wrote:
>> On 2014-01-03 20:00, Marcelo Tosatti wrote:
>>>
>>> After free_loaded_vmcs executes, the "loaded_vmcs" structure
>>> is kfreed, and now vmx->loaded_vmcs points to a kfreed area.
>>> Subsequent free_loaded_vmcs then attempts to manipulate
>>> vmx->loaded_vmcs.
>>
>> Cannot follow yet. How precisely do we call free_loaded_vmcs twice on
>> the same loaded_vmcs?
>
> You don't:
>
> nested_free_all_saved_vmcss calls kfree(item). item is struct
> vmcs02_list *, which is:
>
> /* Used to remember the last vmcs02 used for some recently used vmcs12s
> * */
> struct vmcs02_list {
> struct list_head list;
> gpa_t vmptr;
> struct loaded_vmcs vmcs02;
> };
>
> And vmx->loaded_vmcs = &item->vmcs02.
Yeah, now I see. You may add my
Reviewed-by: Jan Kiszka <jan.kiszka@xxxxxxxxxxx>
>
>> I thought the frees triggered by free_nested ->
>> nested_free_all_saved_vmcss stay away from vmx->loaded_vmcs, no?
>
> Stays away as far as free_loaded_vmcs, yes.
> Except it frees the structure pointed to by vmx->loaded_vmcs.
>
>
>
> The separate question is about when is vmcs01 ever allocated again
> if freed by nested_free_all_saved_vmcss (the other email).
>
>> Jan
>>
>>>
>>> Switch the order to avoid the problem.
>>>
>>> https://bugzilla.redhat.com/show_bug.cgi?id=1047892
>>>
>>> Signed-off-by: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
>>>
>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>>> index da7837e..2efa33f0 100644
>>> --- a/arch/x86/kvm/vmx.c
>>> +++ b/arch/x86/kvm/vmx.c
>>> @@ -7332,8 +7332,8 @@ static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
>>> struct vcpu_vmx *vmx = to_vmx(vcpu);
>>>
>>> free_vpid(vmx);
>>> - free_nested(vmx);
>>> free_loaded_vmcs(vmx->loaded_vmcs);
>>> + free_nested(vmx);
>>> kfree(vmx->guest_msrs);
>>> kvm_vcpu_uninit(vcpu);
>>> kmem_cache_free(kvm_vcpu_cache, vmx);
>>>
>>
>>
>
>
Attachment:
signature.asc
Description: OpenPGP digital signature
