Re: [PATCH] fix memory leak on kvm_vm_ioctl_create_spapr_tce

[David Hildenbrand]

On 23.08.2017 08:06, Paul Mackerras wrote:
> On Wed, Aug 23, 2017 at 01:43:08AM +0000, Nixiaoming wrote:
>>> On 22.08.2017 17:15, David Hildenbrand wrote:
>>>> On 22.08.2017 16:28, nixiaoming wrote:
>>>>> miss kfree(stt) when anon_inode_getfd return fail so add check 
>>>>> anon_inode_getfd return val, and kfree stt
>>>>>
>>>>> Signed-off-by: nixiaoming <nixiaoming@xxxxxxxxxx>
>>>>> ---
>>>>>  arch/powerpc/kvm/book3s_64_vio.c | 5 ++++-
>>>>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/arch/powerpc/kvm/book3s_64_vio.c 
>>>>> b/arch/powerpc/kvm/book3s_64_vio.c
>>>>> index a160c14..a0b4459 100644
>>>>> --- a/arch/powerpc/kvm/book3s_64_vio.c
>>>>> +++ b/arch/powerpc/kvm/book3s_64_vio.c
>>>>> @@ -341,8 +341,11 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm 
>>>>> *kvm,
>>>>>  
>>>>>  	mutex_unlock(&kvm->lock);
>>>>>  
>>>>> -	return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
>>>>> +	ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
>>>>>  				stt, O_RDWR | O_CLOEXEC);
>>>>> +	if (ret < 0)
>>>>> +		goto fail;
>>>>> +	return ret;
>>>>>  
>>>>>  fail:
>>>>>  	if (stt) {
>>>>>
>>>>
>>>>
>>>> stt has already been added to kvm->arch.spapr_tce_tables, so freeing 
>>>> it is evil IMHO. I don't know that code, so I don't know if there is 
>>>> some other place that will make sure that everything in
>>>> kvm->arch.spapr_tce_tables will properly get freed, even when no 
>>>> kvm->release
>>>> function has been called (kvm_spapr_tce_release).
>>>>
>>>
>>> If it is really not freed, than also kvm_put_kvm(stt->kvm) is missing.
>>>
>>> -- 
>>>
>>> Thanks,
>>>
>>> David
>>>
>>
>> if (!stt) return -ENOMEM;
>> kvm_get_kvm(kvm);
>> if anon_inode_getfd return -ENOMEM
>> The user can not determine whether kvm_get_kvm has been called
>> so need add kvm_pet_kvm when anon_inode_getfd fail
>>
>> stt has already been added to kvm->arch.spapr_tce_tables,
>> but if anon_inode_getfd fail, stt is unused val, 
>> so call list_del_rcu, and  free as quickly as possible
>>
>> new patch:
>>
>> ---
>>  arch/powerpc/kvm/book3s_64_vio.c | 10 +++++++++-
>>  1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
>> index a160c14..e2228f1 100644
>> --- a/arch/powerpc/kvm/book3s_64_vio.c
>> +++ b/arch/powerpc/kvm/book3s_64_vio.c
>> @@ -341,8 +341,16 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
>>
>>         mutex_unlock(&kvm->lock);
>>
>> -       return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
>> +       ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
>>                                 stt, O_RDWR | O_CLOEXEC);
>> +       if (ret < 0) {
>> +               mutex_lock(&kvm->lock);
>> +               list_del_rcu(&stt->list);

... don't we have to take care of rcu synchronization before freeing it?

>> +               mutex_unlock(&kvm->lock);
>> +               kvm_put_kvm(kvm);
>> +               goto fail;
>> +       }
>> +       return ret;

of simply

if (!ret)
	return 0;

mutex_lock(&kvm->lock);
list_del_rcu(&stt->list);
mutex_unlock(&kvm->lock);
kvm_put_kvm(kvm);


> 
> It seems to me that it would be better to do the anon_inode_getfd()
> call before the kvm_get_kvm() call, and go to the fail label if it
> fails.

I would have suggested to not add it to the list before it has been
fully created (so nobody can have access to it). But I guess than we
need another level of protection(e.g. kvm->lock).

Am I missing something, or is kvm_vm_ioctl_create_spapr_tce() racy?

The -EBUSY check is done without any locking, so two parallel creators
could create an inconsistency, no? Shouldn't this all be protected by
kvm->lock?

> 
> Paul.
> 

Independent of the fix, I'd suggest the following cleanup.