On Tue, Aug 13, 2024 at 03:28:03PM -0400, Matthew Rosato wrote:
> This fixes a crash when surprise hot-unplugging a PCI device. This crash
> happens because during hot-unplug __iommu_group_set_domain_nofail()
> attaching the default domain fails when the platform no longer
> recognizes the device as it has already been removed and we end up with
> a NULL domain pointer and UAF. This is exactly the case referred to in
> the second comment in __iommu_device_set_domain() and just as stated
> there if we can instead attach the blocking domain the UAF is prevented
> as this can handle the already removed device. Implement the blocking
> domain to use this handling. With this change, the crash is fixed but
> we still hit a warning attempting to change DMA ownership on a blocked
> device.
>
> Fixes: c76c067e488c ("s390/pci: Use dma-iommu layer")
> Co-developed-by: Niklas Schnelle <schnelle@xxxxxxxxxxxxx>
> Signed-off-by: Niklas Schnelle <schnelle@xxxxxxxxxxxxx>
> Signed-off-by: Matthew Rosato <mjrosato@xxxxxxxxxxxxx>
> ---
> Changes for v2:
> - Added co-author tag from Niklas + my SoB
> - Removed changes to drivers/iommu/iommu.c
> - Revert back to -EIO for failed attach in s390-iommu
> - Set blocking domain during probe_device / remove s390_domain check during
> blocking attach
> - Remove s390_iommu_release_device
> - Update commit message to reflect changes
> ---
> drivers/iommu/s390-iommu.c | 64 ++++++++++++++++++++++++++------------
> 1 file changed, 44 insertions(+), 20 deletions(-)
This is probably OK as is, but there are a few things that don't quite
match the pattern still here
The blocking domain should be the iommu_domain type since it is
global and shared. None of the additional driver members should ever
be touched as that would maybe become dangerous.
> +static struct s390_domain s390_blocking_domain = {
Ideally would be struct iommu_domain s390_blocking_domain
That in turn means going around and looking carefully at
zdev->s390_domain. All uses can probably be removed except for this:
struct zpci_iommu_ctrs *zpci_get_iommu_ctrs(struct zpci_dev *zdev)
{
if (!zdev || !zdev->s390_domain)
return NULL;
return &zdev->s390_domain->ctrs;
}
Which doesn't look good for a blocking domain anyhow. Also the above
looks racy, nothing prevents s390_domain from being freed if it is
read outside an iommu op.
The checks for null zdev also don't make sense:
static int blocking_domain_attach_device(struct iommu_domain *domain,
struct device *dev)
{
struct s390_domain *s390_domain = to_s390_domain(domain);
struct zpci_dev *zdev = to_zpci_dev(dev);
unsigned long flags;
if (!zdev)
return 0;
The core guarentees these functions are never called unless probe
succeeds and probe won't succeed if zdev is NULL.
And it would be good to rename s390_iommu_detach_device() to
blocking_domain_attach_device(), then the obsolete "detach" naming is
gone and this driver just uses a blocked before attach pattern.
> @@ -403,16 +414,14 @@ static int s390_iommu_attach_device(struct iommu_domain *domain,
>
> if (zdev->s390_domain)
> s390_iommu_detach_device(&zdev->s390_domain->domain, dev);
s390_domain is never NULL now, the test can go away
> +static int blocking_domain_attach_device(struct iommu_domain *domain,
> + struct device *dev)
> +{
> + struct s390_domain *s390_domain = to_s390_domain(domain);
> + struct zpci_dev *zdev = to_zpci_dev(dev);
> + unsigned long flags;
flags is never used? Compiler didn't warn?
Jason
