Re: [PATCH v3 25/32] iommufd/selftest: switch to iommu_iova_to_phys_length

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Race condition and lockless access in IOMMU_TEST_OP_MD_CHECK_MAP leading to UAF
--

commit bb925df740c15b99ff28ec1c448ae8c5dd87cd5d
Author: Guanghui Feng <guanghuifeng@linux.alibaba.com>

iommufd/selftest: switch to iommu_iova_to_phys_length

This commit replaces the direct domain->ops->iova_to_phys() call with
the new iommu_iova_to_phys_length() interface in the iommufd selftest.

> diff --git a/drivers/iommu/iommufd/selftest.c b/drivers/iommu/iommufd/selftest.c
> index af07c642a5260..d4dd399302243 100644
> --- a/drivers/iommu/iommufd/selftest.c
> +++ b/drivers/iommu/iommufd/selftest.c
[ ... ]
> @@ -1214,7 +1214,7 @@ static int iommufd_test_md_check_pa(struct iommufd_ucmd *ucmd,
>  		pfn = page_to_pfn(pages[0]);
>  		put_page(pages[0]);
>  
> -		io_phys = mock->domain.ops->iova_to_phys(&mock->domain, iova);
> +		io_phys = iommu_iova_to_phys_length(&mock->domain, iova, NULL);

[Severity: High]
This is a pre-existing issue, but does this lockless call to
iommu_iova_to_phys_length() risk a use-after-free?

In iommufd_test_md_check_pa(), the page table walk is initiated on the
mock domain without holding any IOAS read range lock (like iopt->iova_rwsem).
Since the mock domain page tables use the generic_pt implementation, its API
contract requires the caller to hold a read range lock that includes the iova.

If a concurrent IOMMU_IOAS_UNMAP ioctl frees the intermediate page table
directories while this lockless walk is ongoing, could it lead to a
kernel crash?

>  		if (io_phys !=
>  		    pfn * PAGE_SIZE + ((uintptr_t)uptr % PAGE_SIZE)) {
>  			rc = -EINVAL;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260603151804.1963871-1-guanghuifeng@linux.alibaba.com?part=25