[PATCH v3 1/9] vfio: Fix group release deadlock

[Alex Williamson <alex.williamson@redhat.com>]

If vfio_iommu_group_notifier() acquires a group reference and that
reference becomes the last reference to the group, then vfio_group_put
introduces a deadlock code path where we're trying to unregister from
the iommu notifier chain from within a callout of that chain.  Use a
work_struct to release this reference asynchronously.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Cc: stable@vger.kernel.org
---
 drivers/vfio/vfio.c |   37 ++++++++++++++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
index 6a49485eb49d..54dd2fbf83d9 100644
--- a/drivers/vfio/vfio.c
+++ b/drivers/vfio/vfio.c
@@ -423,6 +423,34 @@ static void vfio_group_put(struct vfio_group *group)
 	kref_put_mutex(&group->kref, vfio_group_release, &vfio.group_lock);
 }
 
+struct vfio_group_put_work {
+	struct work_struct work;
+	struct vfio_group *group;
+};
+
+static void vfio_group_put_bg(struct work_struct *work)
+{
+	struct vfio_group_put_work *do_work;
+
+	do_work = container_of(work, struct vfio_group_put_work, work);
+
+	vfio_group_put(do_work->group);
+	kfree(do_work);
+}
+
+static void vfio_group_schedule_put(struct vfio_group *group)
+{
+	struct vfio_group_put_work *do_work;
+
+	do_work = kmalloc(sizeof(*do_work), GFP_KERNEL);
+	if (WARN_ON(!do_work))
+		return;
+
+	INIT_WORK(&do_work->work, vfio_group_put_bg);
+	do_work->group = group;
+	schedule_work(&do_work->work);
+}
+
 /* Assume group_lock or group reference is held */
 static void vfio_group_get(struct vfio_group *group)
 {
@@ -762,7 +790,14 @@ static int vfio_iommu_group_notifier(struct notifier_block *nb,
 		break;
 	}
 
-	vfio_group_put(group);
+	/*
+	 * If we're the last reference to the group, the group will be
+	 * released, which includes unregistering the iommu group notifier.
+	 * We hold a read-lock on that notifier list, unregistering needs
+	 * a write-lock... deadlock.  Release our reference asynchronously
+	 * to avoid that situation.
+	 */
+	vfio_group_schedule_put(group);
 	return NOTIFY_OK;
 }