Re: [PATCH v2 05/11] iommu/sva: Assign a PASID to mm on PASID allocation and free it on mm exit

[Thomas Gleixner]

On Wed, Jan 26 2022 at 09:36, Fenghua Yu wrote:
> On Wed, Jan 26, 2022 at 03:23:42PM +0100, Thomas Gleixner wrote:
>> On Tue, Jan 25 2022 at 07:18, Fenghua Yu wrote:
>> While looking at ioasid_put() usage I tripped over the following UAF
>> issue:
>> 
>> --- a/drivers/iommu/intel/iommu.c
>> +++ b/drivers/iommu/intel/iommu.c
>> @@ -4817,8 +4817,10 @@ static int aux_domain_add_dev(struct dma
>>  	auxiliary_unlink_device(domain, dev);
>>  link_failed:
>>  	spin_unlock_irqrestore(&device_domain_lock, flags);
>> -	if (list_empty(&domain->subdevices) && domain->default_pasid > 0)
>> +	if (list_empty(&domain->subdevices) && domain->default_pasid > 0) {
>>  		ioasid_put(domain->default_pasid);
>> +		domain->default_pasid = INVALID_IOASID;
>> +	}
>>  
>>  	return ret;
>>  }
>> @@ -4847,8 +4849,10 @@ static void aux_domain_remove_dev(struct
>>  
>>  	spin_unlock_irqrestore(&device_domain_lock, flags);
>>  
>> -	if (list_empty(&domain->subdevices) && domain->default_pasid > 0)
>> +	if (list_empty(&domain->subdevices) && domain->default_pasid > 0) {
>>  		ioasid_put(domain->default_pasid);
>> +		domain->default_pasid = INVALID_IOASID;
>> +	}
>>  }
>>  
>>  static int prepare_domain_attach_device(struct iommu_domain *domain,
>
> The above patch fixes an existing issue. I will put it in a separate patch,
> right?

Correct.

> It cannot be applied cleanly to the upstream tree. Do you want me to base
> the above patch (and the whole patch set) to the upstream tree or a specific
> tip branch?

Against Linus tree please so that the bugfix applies.

> I will fold the following patch into patch #5. The patch #11 (the doc patch)
> also needs to remove one paragraph talking about refcount.
>
> So I will send the whole patch set with the following changes:
> 1. One new bug fix patch (the above patch)
> 2. Updated patch #5 (with the following patch folded)
> 3. Updated patch #11 (removing refcount description)

Looks good.

Thanks,

        tglx