Re: [PATCH v2 2/8] iommu/dma: Fail unaligned map requests for untrusted devs

["Sven Peter" <sven@svenpeter.dev>]

and ofc shortly after submitting this series I realized this doesn't quite work yet:
swiotlb_tbl_map_single can return a 16KB buffer that's only aligned to a 4KB boundary. 
v3 will need at least another change to ensure that the result will be aligned to
a 16KB boundary as well.


Sven


On Sat, Aug 28, 2021, at 17:36, Sven Peter wrote:
> If swiotlb is enabled we should never try to create any mappings that
> would expose more memory than requested to the device.
> WARN_ON and refuse those mappings just in case.
> 
> Signed-off-by: Sven Peter <sven@svenpeter.dev>
> ---
>  drivers/iommu/dma-iommu.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index e8eae34e9e4f..d6e273ec3de6 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -534,13 +534,20 @@ static dma_addr_t __iommu_dma_map(struct device 
> *dev, phys_addr_t phys,
>  	struct iommu_dma_cookie *cookie = domain->iova_cookie;
>  	struct iova_domain *iovad = &cookie->iovad;
>  	size_t iova_off = iova_offset(iovad, phys);
> +	size_t size_aligned = iova_align(iovad, size + iova_off);
>  	dma_addr_t iova;
>  
>  	if (static_branch_unlikely(&iommu_deferred_attach_enabled) &&
>  	    iommu_deferred_attach(dev, domain))
>  		return DMA_MAPPING_ERROR;
>  
> -	size = iova_align(iovad, size + iova_off);
> +	if (IS_ENABLED(CONFIG_SWIOTLB) && dev_is_untrusted(dev)) {
> +		if (WARN_ON(iova_off))
> +			return DMA_MAPPING_ERROR;
> +		if (WARN_ON(size_aligned != size))
> +			return DMA_MAPPING_ERROR;
> +	}
> +	size = size_aligned;
>  
>  	iova = iommu_dma_alloc_iova(domain, size, dma_mask, dev);
>  	if (!iova)
> -- 
> 2.25.1
> 
> 


-- 
Sven Peter