Re: KASAN: use-after-free Read in do_general_protection

[Dmitry Vyukov]

On Tue, Jun 5, 2018 at 10:16 AM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Mon, May 28, 2018 at 1:25 PM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>> On 26/05/2018 11:29, Dmitry Vyukov wrote:
>>> KASAN: stack-out-of-bounds Read in do_general_protection
>>> KASAN: slab-out-of-bounds Read in vmx_vcpu_run
>>> KASAN: use-after-scope Read in vmx_vcpu_run
>>> KASAN: stack-out-of-bounds Write in notify_die
>>>
>>> See full info at:
>>> https://syzkaller.appspot.com/bug?extid=a1264132fc103340628f
>>>
>>>
>>> There seems to be 2 problems:
>>>
>>> 1. msr_write_intercepted doing something notoriously bad.
>>
>> The faulting line is
>>
>>         msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
>>
>> so I suppose to_vmx(vcpu)->loaded_vmcs is bogus?  That seems like "just"
>> a corruption of the struct kvm_vcpu, because the loaded_vmcs field is
>> pointing elsewhere inside the same struct.
>
> This is reproducible and the reproducer only uses KVM syscalls. So it
> seems that KVM corrupts itself. Perhaps it's KVM_SET_MSRS that messes
> things?

This one looks like the same root cause as here:

KASAN: stack-out-of-bounds Read in unwind_next_frame
https://groups.google.com/d/msg/syzkaller-bugs/MGQ_W8DPYXA/k-xK7sBrBgAJ