On Thu, 7 Jan 2021 17:29:00 +0800
Keqian Zhu <zhukeqi...@huawei.com> wrote:
> If we detach group during dirty page tracking, we shouldn't remove
> vfio_dma, because dirty log will lose.
>
> But we don't prevent unmap_unpin_all in vfio_iommu_release, because
> under normal procedure, dirty tracking has been stopped.
This looks like it's creating a larger problem than it's fixing, it's
not our job to maintain the dirty bitmap regardless of what the user
does. If the user detaches the last group in a container causing the
mappings within that container to be deconstructed before the user has
collected dirty pages, that sounds like a user error. A container with
no groups is de-privileged and therefore loses all state. Thanks,
Alex
> Fixes: d6a4c185660c ("vfio iommu: Implementation of ioctl for dirty pages
> tracking")
> Signed-off-by: Keqian Zhu <zhukeqi...@huawei.com>
> ---
> drivers/vfio/vfio_iommu_type1.c | 14 ++++++++++++--
> 1 file changed, 12 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
> index 26b7eb2a5cfc..9776a059904d 100644
> --- a/drivers/vfio/vfio_iommu_type1.c
> +++ b/drivers/vfio/vfio_iommu_type1.c
> @@ -2373,7 +2373,12 @@ static void vfio_iommu_type1_detach_group(void
> *iommu_data,
> if (list_empty(&iommu->external_domain->group_list)) {
> vfio_sanity_check_pfn_list(iommu);
>
> - if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
> + /*
> + * During dirty page tracking, we can't remove
> + * vfio_dma because dirty log will lose.
> + */
> + if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu) &&
> + !iommu->dirty_page_tracking)
> vfio_iommu_unmap_unpin_all(iommu);
>
> kfree(iommu->external_domain);
> @@ -2406,10 +2411,15 @@ static void vfio_iommu_type1_detach_group(void
> *iommu_data,
> * iommu and external domain doesn't exist, then all the
> * mappings go away too. If it's the last domain with iommu and
> * external domain exist, update accounting
> + *
> + * Note: During dirty page tracking, we can't remove vfio_dma
> + * because dirty log will lose. Just update accounting is a good
> + * choice.
> */
> if (list_empty(&domain->group_list)) {
> if (list_is_singular(&iommu->domain_list)) {
> - if (!iommu->external_domain)
> + if (!iommu->external_domain &&
> + !iommu->dirty_page_tracking)
> vfio_iommu_unmap_unpin_all(iommu);
> else
> vfio_iommu_unmap_unpin_reaccount(iommu);
