re: KVM: Introduce a 'release' method for KVM devices

re: KVM: Introduce a 'release' method for KVM devices

[Colin Ian King]

Hi,

Static analysis with Coverity picked up an issue in the following commit:

commit 2bde9b3ec8bdf60788e9e2ce8c07a2f8d6003dbd
Author: Cédric Le Goater <clg@kaod.org>
Date:   Thu Apr 18 12:39:41 2019 +0200

    KVM: Introduce a 'release' method for KVM devices


        struct kvm *kvm = dev->kvm;

+       if (!dev)
+               return -ENODEV;

If dev is null then the dereference of dev->kvm when assigning pointer
kvm will cause an null pointer dereference.  This is easily fixed by
assigning kvm after the dev null check.

+
+       if (dev->kvm != kvm)
+               return -EPERM;

I don't understand the logic of the above check. kvm is the same
dev->kvm on the earlier assignment, so dev->kvm != kvm seems to be
always false, so this check seems to be redundant. Am I missing
something more fundamental here?

Colin

Re: KVM: Introduce a 'release' method for KVM devices

[Alexey Kardashevskiy]

On 02/05/2019 00:42, Colin Ian King wrote:
> Hi,
> 
> Static analysis with Coverity picked up an issue in the following commit:
> 
> commit 2bde9b3ec8bdf60788e9e2ce8c07a2f8d6003dbd
> Author: Cédric Le Goater <clg@kaod.org>
> Date:   Thu Apr 18 12:39:41 2019 +0200
> 
>     KVM: Introduce a 'release' method for KVM devices
> 
> 
>         struct kvm *kvm = dev->kvm;
> 
> +       if (!dev)
> +               return -ENODEV;
> 
> If dev is null then the dereference of dev->kvm when assigning pointer
> kvm will cause an null pointer dereference.  This is easily fixed by
> assigning kvm after the dev null check.

Yes, this is a bug.

> 
> +
> +       if (dev->kvm != kvm)
> +               return -EPERM;
> 
> I don't understand the logic of the above check. kvm is the same
> dev->kvm on the earlier assignment, so dev->kvm != kvm seems to be
> always false, so this check seems to be redundant. Am I missing
> something more fundamental here?

Nope. This looks like unfortunate cut-n-paste which slipped through out
reviewing process :-D


-- 
Alexey

Re: KVM: Introduce a 'release' method for KVM devices

[Cédric Le Goater]

On 5/2/19 4:35 AM, Alexey Kardashevskiy wrote:
> 
> 
> On 02/05/2019 00:42, Colin Ian King wrote:
>> Hi,
>>
>> Static analysis with Coverity picked up an issue in the following commit:
>>
>> commit 2bde9b3ec8bdf60788e9e2ce8c07a2f8d6003dbd
>> Author: Cédric Le Goater <clg@kaod.org>
>> Date:   Thu Apr 18 12:39:41 2019 +0200
>>
>>     KVM: Introduce a 'release' method for KVM devices
>>
>>
>>         struct kvm *kvm = dev->kvm;
>>
>> +       if (!dev)
>> +               return -ENODEV;
>>
>> If dev is null then the dereference of dev->kvm when assigning pointer
>> kvm will cause an null pointer dereference.  This is easily fixed by
>> assigning kvm after the dev null check.
> 
> Yes, this is a bug.

Clearly.

>>
>> +
>> +       if (dev->kvm != kvm)
>> +               return -EPERM;
>>
>> I don't understand the logic of the above check. kvm is the same
>> dev->kvm on the earlier assignment, so dev->kvm != kvm seems to be
>> always false, so this check seems to be redundant. Am I missing
>> something more fundamental here?
> 
> Nope. This looks like unfortunate cut-n-paste which slipped through out
> reviewing process :-D

Yes. My bad :/ I will send a cleanup patch for 5.2

Thanks,

C.