vfio: Include No-IOMMU mode

There is really no way to safely give a user full access to a DMA
capable device without an IOMMU to protect the host system. There is
also no way to provide DMA translation, for use cases such as device
assignment to virtual machines. However, there are still those users
that want userspace drivers even under those conditions. The UIO
driver exists for this use case, but does not provide the degree of
device access and programming that VFIO has. In an effort to avoid
code duplication, this introduces a No-IOMMU mode for VFIO.

This mode requires building VFIO with CONFIG_VFIO_NOIOMMU and enabling
the "enable_unsafe_noiommu_mode" option on the vfio driver. This
should make it very clear that this mode is not safe. Additionally,
CAP_SYS_RAWIO privileges are necessary to work with groups and
containers using this mode. Groups making use of this support are
named /dev/vfio/noiommu-$GROUP and can only make use of the special
VFIO_NOIOMMU_IOMMU for the container. Use of this mode, specifically
binding a device without a native IOMMU group to a VFIO bus driver
will taint the kernel and should therefore not be considered
supported. This patch includes no-iommu support for the vfio-pci bus
driver only.

Signed-off-by: Alex Williamson <[e-mail]>
Acked-by: Michael S. Tsirkin <[e-mail]>

drivers/vfio/Kconfig

···
 
 	  If you don't know what to do here, say N.
 
+menuconfig VFIO_NOIOMMU
+	bool "VFIO No-IOMMU support"
+	depends on VFIO
+	help
+	  VFIO is built on the ability to isolate devices using the IOMMU.
+	  Only with an IOMMU can userspace access to DMA capable devices be
+	  considered secure.  VFIO No-IOMMU mode enables IOMMU groups for
+	  devices without IOMMU backing for the purpose of re-using the VFIO
+	  infrastructure in a non-secure mode.  Use of this mode will result
+	  in an unsupportable kernel and will therefore taint the kernel.
+	  Device assignment to virtual machines is also not possible with
+	  this mode since there is no IOMMU to provide DMA translation.
+
+	  If you don't know what to do here, say N.
+
 source "drivers/vfio/pci/Kconfig"
 source "drivers/vfio/platform/Kconfig"
 source "virt/lib/Kconfig"

drivers/vfio/pci/vfio_pci.c

···
 	if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL)
 		return -EINVAL;
 
-	group = iommu_group_get(&pdev->dev);
+	group = vfio_iommu_group_get(&pdev->dev);
 	if (!group)
 		return -EINVAL;
 
 	vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
 	if (!vdev) {
-		iommu_group_put(group);
+		vfio_iommu_group_put(group, &pdev->dev);
 		return -ENOMEM;
 	}
 
···
 
 	ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
 	if (ret) {
-		iommu_group_put(group);
+		vfio_iommu_group_put(group, &pdev->dev);
 		kfree(vdev);
 		return ret;
 	}
···
 	if (!vdev)
 		return;
 
-	iommu_group_put(pdev->dev.iommu_group);
+	vfio_iommu_group_put(pdev->dev.iommu_group, &pdev->dev);
 	kfree(vdev);
 
 	if (vfio_pci_is_vga(pdev)) {

drivers/vfio/vfio.c

···
 	struct rw_semaphore		group_lock;
 	struct vfio_iommu_driver	*iommu_driver;
 	void				*iommu_data;
+	bool				noiommu;
 };
 
 struct vfio_unbound_dev {
···
 	struct list_head		unbound_list;
 	struct mutex			unbound_lock;
 	atomic_t			opened;
+	bool				noiommu;
 };
 
 struct vfio_device {
···
 	struct list_head		group_next;
 	void				*device_data;
 };
+
+#ifdef CONFIG_VFIO_NOIOMMU
+static bool noiommu __read_mostly;
+module_param_named(enable_unsafe_noiommu_mode,
+		   noiommu, bool, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(enable_unsafe_noiommu_mode, "Enable UNSAFE, no-IOMMU mode.  This mode provides no device isolation, no DMA translation, no host kernel protection, cannot be used for device assignment to virtual machines, requires RAWIO permissions, and will taint the kernel.  If you do not know what this is for, step away. (default: false)");
+#endif
+
+/*
+ * vfio_iommu_group_{get,put} are only intended for VFIO bus driver probe
+ * and remove functions, any use cases other than acquiring the first
+ * reference for the purpose of calling vfio_add_group_dev() or removing
+ * that symmetric reference after vfio_del_group_dev() should use the raw
+ * iommu_group_{get,put} functions.  In particular, vfio_iommu_group_put()
+ * removes the device from the dummy group and cannot be nested.
+ */
+struct iommu_group *vfio_iommu_group_get(struct device *dev)
+{
+	struct iommu_group *group;
+	int __maybe_unused ret;
+
+	group = iommu_group_get(dev);
+
+#ifdef CONFIG_VFIO_NOIOMMU
+	/*
+	 * With noiommu enabled, an IOMMU group will be created for a device
+	 * that doesn't already have one and doesn't have an iommu_ops on their
+	 * bus.  We use iommu_present() again in the main code to detect these
+	 * fake groups.
+	 */
+	if (group || !noiommu || iommu_present(dev->bus))
+		return group;
+
+	group = iommu_group_alloc();
+	if (IS_ERR(group))
+		return NULL;
+
+	iommu_group_set_name(group, "vfio-noiommu");
+	ret = iommu_group_add_device(group, dev);
+	iommu_group_put(group);
+	if (ret)
+		return NULL;
+
+	/*
+	 * Where to taint?  At this point we've added an IOMMU group for a
+	 * device that is not backed by iommu_ops, therefore any iommu_
+	 * callback using iommu_ops can legitimately Oops.  So, while we may
+	 * be about to give a DMA capable device to a user without IOMMU
+	 * protection, which is clearly taint-worthy, let's go ahead and do
+	 * it here.
+	 */
+	add_taint(TAINT_USER, LOCKDEP_STILL_OK);
+	dev_warn(dev, "Adding kernel taint for vfio-noiommu group on device\n");
+#endif
+
+	return group;
+}
+EXPORT_SYMBOL_GPL(vfio_iommu_group_get);
+
+void vfio_iommu_group_put(struct iommu_group *group, struct device *dev)
+{
+#ifdef CONFIG_VFIO_NOIOMMU
+	if (!iommu_present(dev->bus))
+		iommu_group_remove_device(dev);
+#endif
+
+	iommu_group_put(group);
+}
+EXPORT_SYMBOL_GPL(vfio_iommu_group_put);
+
+#ifdef CONFIG_VFIO_NOIOMMU
+static void *vfio_noiommu_open(unsigned long arg)
+{
+	if (arg != VFIO_NOIOMMU_IOMMU)
+		return ERR_PTR(-EINVAL);
+	if (!capable(CAP_SYS_RAWIO))
+		return ERR_PTR(-EPERM);
+
+	return NULL;
+}
+
+static void vfio_noiommu_release(void *iommu_data)
+{
+}
+
+static long vfio_noiommu_ioctl(void *iommu_data,
+			       unsigned int cmd, unsigned long arg)
+{
+	if (cmd == VFIO_CHECK_EXTENSION)
+		return noiommu && (arg == VFIO_NOIOMMU_IOMMU) ? 1 : 0;
+
+	return -ENOTTY;
+}
+
+static int vfio_iommu_present(struct device *dev, void *unused)
+{
+	return iommu_present(dev->bus) ? 1 : 0;
+}
+
+static int vfio_noiommu_attach_group(void *iommu_data,
+				     struct iommu_group *iommu_group)
+{
+	return iommu_group_for_each_dev(iommu_group, NULL,
+					vfio_iommu_present) ? -EINVAL : 0;
+}
+
+static void vfio_noiommu_detach_group(void *iommu_data,
+				      struct iommu_group *iommu_group)
+{
+}
+
+static const struct vfio_iommu_driver_ops vfio_noiommu_ops = {
+	.name = "vfio-noiommu",
+	.owner = THIS_MODULE,
+	.open = vfio_noiommu_open,
+	.release = vfio_noiommu_release,
+	.ioctl = vfio_noiommu_ioctl,
+	.attach_group = vfio_noiommu_attach_group,
+	.detach_group = vfio_noiommu_detach_group,
+};
+#endif
+
 
 /**
  * IOMMU driver registration
···
 /**
  * Group objects - create, release, get, put, search
  */
-static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group)
+static struct vfio_group *vfio_create_group(struct iommu_group *iommu_group,
+					    bool iommu_present)
 {
 	struct vfio_group *group, *tmp;
 	struct device *dev;
···
 	atomic_set(&group->container_users, 0);
 	atomic_set(&group->opened, 0);
 	group->iommu_group = iommu_group;
+	group->noiommu = !iommu_present;
 
 	group->nb.notifier_call = vfio_iommu_group_notifier;
 
···
 
 	dev = device_create(vfio.class, NULL,
 			    MKDEV(MAJOR(vfio.group_devt), minor),
-			    group, "%d", iommu_group_id(iommu_group));
+			    group, "%s%d", group->noiommu ? "noiommu-" : "",
+			    iommu_group_id(iommu_group));
 	if (IS_ERR(dev)) {
 		vfio_free_group_minor(minor);
 		vfio_group_unlock_and_free(group);
···
 
 	group = vfio_group_get_from_iommu(iommu_group);
 	if (!group) {
-		group = vfio_create_group(iommu_group);
+		group = vfio_create_group(iommu_group, iommu_present(dev->bus));
 		if (IS_ERR(group)) {
 			iommu_group_put(iommu_group);
 			return PTR_ERR(group);
···
 			mutex_lock(&vfio.iommu_drivers_lock);
 			list_for_each_entry(driver, &vfio.iommu_drivers_list,
 					    vfio_next) {
+
+#ifdef CONFIG_VFIO_NOIOMMU
+				if (!list_empty(&container->group_list) &&
+				    (container->noiommu !=
+				     (driver->ops == &vfio_noiommu_ops)))
+					continue;
+#endif
+
 				if (!try_module_get(driver->ops->owner))
 					continue;
 
···
 	mutex_lock(&vfio.iommu_drivers_lock);
 	list_for_each_entry(driver, &vfio.iommu_drivers_list, vfio_next) {
 		void *data;
+
+#ifdef CONFIG_VFIO_NOIOMMU
+		/*
+		 * Only noiommu containers can use vfio-noiommu and noiommu
+		 * containers can only use vfio-noiommu.
+		 */
+		if (container->noiommu != (driver->ops == &vfio_noiommu_ops))
+			continue;
+#endif
 
 		if (!try_module_get(driver->ops->owner))
 			continue;
···
 	if (atomic_read(&group->container_users))
 		return -EINVAL;
 
+	if (group->noiommu && !capable(CAP_SYS_RAWIO))
+		return -EPERM;
+
 	f = fdget(container_fd);
 	if (!f.file)
 		return -EBADF;
···
 
 	down_write(&container->group_lock);
 
+	/* Real groups and fake groups cannot mix */
+	if (!list_empty(&container->group_list) &&
+	    container->noiommu != group->noiommu) {
+		ret = -EPERM;
+		goto unlock_out;
+	}
+
 	driver = container->iommu_driver;
 	if (driver) {
 		ret = driver->ops->attach_group(container->iommu_data,
···
 	}
 
 	group->container = container;
+	container->noiommu = group->noiommu;
 	list_add(&group->container_next, &container->group_list);
 
 	/* Get a reference on the container and mark a user within the group */
···
 	if (0 == atomic_read(&group->container_users) ||
 	    !group->container->iommu_driver || !vfio_group_viable(group))
 		return -EINVAL;
+
+	if (group->noiommu && !capable(CAP_SYS_RAWIO))
+		return -EPERM;
 
 	device = vfio_device_get_from_name(group, buf);
 	if (!device)
···
 	atomic_inc(&group->container_users);
 
 	fd_install(ret, filep);
+
+	if (group->noiommu)
+		dev_warn(device->dev, "vfio-noiommu device opened by user "
+			 "(%s:%d)\n", current->comm, task_pid_nr(current));
 
 	return ret;
 }
···
 	group = vfio_group_get_from_minor(iminor(inode));
 	if (!group)
 		return -ENODEV;
+
+	if (group->noiommu && !capable(CAP_SYS_RAWIO)) {
+		vfio_group_put(group);
+		return -EPERM;
+	}
 
 	/* Do we need multiple instances of the group open?  Seems not. */
 	opened = atomic_cmpxchg(&group->opened, 0, 1);
···
 	if (!atomic_inc_not_zero(&group->container_users))
 		return ERR_PTR(-EINVAL);
 
+	if (group->noiommu) {
+		atomic_dec(&group->container_users);
+		return ERR_PTR(-EPERM);
+	}
+
 	if (!group->container->iommu_driver ||
 			!vfio_group_viable(group)) {
 		atomic_dec(&group->container_users);
···
 	request_module_nowait("vfio_iommu_type1");
 	request_module_nowait("vfio_iommu_spapr_tce");
 
+#ifdef CONFIG_VFIO_NOIOMMU
+	vfio_register_iommu_driver(&vfio_noiommu_ops);
+#endif
 	return 0;
 
 err_cdev_add:
···
 {
 	WARN_ON(!list_empty(&vfio.group_list));
 
+#ifdef CONFIG_VFIO_NOIOMMU
+	vfio_unregister_iommu_driver(&vfio_noiommu_ops);
+#endif
 	idr_destroy(&vfio.group_idr);
 	cdev_del(&vfio.group_cdev);
 	unregister_chrdev_region(vfio.group_devt, MINORMASK);

include/linux/vfio.h

···
 	void	(*request)(void *device_data, unsigned int count);
 };
 
+extern struct iommu_group *vfio_iommu_group_get(struct device *dev);
+extern void vfio_iommu_group_put(struct iommu_group *group, struct device *dev);
+
 extern int vfio_add_group_dev(struct device *dev,
 			      const struct vfio_device_ops *ops,
 			      void *device_data);

include/uapi/linux/vfio.h

···
 #define VFIO_SPAPR_TCE_v2_IOMMU		7
 
 /*
+ * The No-IOMMU IOMMU offers no translation or isolation for devices and
+ * supports no ioctls outside of VFIO_CHECK_EXTENSION.  Use of VFIO's No-IOMMU
+ * code will taint the host kernel and should be used with extreme caution.
+ */
+#define VFIO_NOIOMMU_IOMMU		8
+
+/*
  * The IOCTL interface is designed for extensibility by embedding the
  * structure length (argsz) and flags into structures passed between
  * kernel and userspace.  We therefore use the _IO() macro for these