Re: [PATCH rc v2] iommufd: Explicitize struct iommu_hwpt_pgfault padding
From: Nicolin Chen
Date: Mon Jan 20 2025 - 14:47:17 EST
On Mon, Jan 20, 2025 at 10:37:19AM -0400, Jason Gunthorpe wrote:
> On Sun, Jan 19, 2025 at 10:24:11PM -0800, Nicolin Chen wrote:
> > --- a/drivers/iommu/iommufd/fault.c
> > +++ b/drivers/iommu/iommufd/fault.c
> > @@ -253,6 +253,7 @@ static void iommufd_compose_fault_message(struct iommu_fault *fault,
> > hwpt_fault->pasid = fault->prm.pasid;
> > hwpt_fault->grpid = fault->prm.grpid;
> > hwpt_fault->perm = fault->prm.perm;
> > + hwpt_fault->__reserved = 0;
> > hwpt_fault->addr = fault->prm.addr;
> > hwpt_fault->length = 0;
> > hwpt_fault->cookie = cookie;
>
> Yikes, so it was leaking kernel stack memory through the padding too.
>
> We should zero init the stack struct to be safe:
>
> @@ -247,7 +247,7 @@ static ssize_t iommufd_fault_fops_read(struct file *filep, char __user *buf,
> {
> size_t fault_size = sizeof(struct iommu_hwpt_pgfault);
> struct iommufd_fault *fault = filep->private_data;
> - struct iommu_hwpt_pgfault data;
> + struct iommu_hwpt_pgfault data = {};
> struct iommufd_device *idev;
> struct iopf_group *group;
> struct iopf_fault *iopf;
>
> I can fix it up if that is the only change
Sending a v3.
I think having the zeroing in iommufd_fault_fops_read can drop
the "__reserved = 0" iommufd_compose_fault_message.
Thanks
Nicolin