When device_add(&udev->dev) failed, calling put_device() to explicitly
release udev->dev. And the routine which calls usb_new_device() does
not call put_device() when an error occurs. As comment of device_add()
says, 'if device_add() succeeds, you should call device_del() when you
want to get rid of it. If device_add() has not succeeded, use only
put_device() to drop the reference count'.

Found by code review.

Cc: stable@vger.kernel.org
Fixes: 9f8b17e643fe ("USB: make usbdevices export their device nodes instead of using a separate class")
Signed-off-by: Ma Ke <make_ruc2021@163.com>
---
Changes in v2:
- modified the bug description to make it more clear;
- added the missed part of the patch.
---
 drivers/usb/core/hub.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
index 4b93c0bd1d4b..ddd572312296 100644
--- a/drivers/usb/core/hub.c
+++ b/drivers/usb/core/hub.c
@@ -2651,6 +2651,7 @@ int usb_new_device(struct usb_device *udev)
 	err = device_add(&udev->dev);
 	if (err) {
 		dev_err(&udev->dev, "can't device_add, error %d\n", err);
+		put_device(&udev->dev);
 		goto fail;
 	}
 
@@ -2663,13 +2664,13 @@ int usb_new_device(struct usb_device *udev)
 		err = sysfs_create_link(&udev->dev.kobj,
 				&port_dev->dev.kobj, "port");
 		if (err)
-			goto fail;
+			goto out_del_dev;
 
 		err = sysfs_create_link(&port_dev->dev.kobj,
 				&udev->dev.kobj, "device");
 		if (err) {
 			sysfs_remove_link(&udev->dev.kobj, "port");
-			goto fail;
+			goto out_del_dev;
 		}
 
 		if (!test_and_set_bit(port1, hub->child_usage_bits))
@@ -2683,6 +2684,9 @@ int usb_new_device(struct usb_device *udev)
 	pm_runtime_put_sync_autosuspend(&udev->dev);
 	return err;
 
+out_del_dev:
+	device_del(&udev->dev);
+	put_device(&udev->dev);
 fail:
 	usb_set_device_state(udev, USB_STATE_NOTATTACHED);
 	pm_runtime_disable(&udev->dev);
-- 
2.25.1

On Wed, Dec 18, 2024 at 10:19:40AM +0800, Ma Ke wrote:
> When device_add(&udev->dev) failed, calling put_device() to explicitly
> release udev->dev. And the routine which calls usb_new_device() does
> not call put_device() when an error occurs.

That is wrong.

usb_new_device() is called by hub_port_connect().  The code does:

			status = usb_new_device(udev);
			...

		if (status)
			goto loop_disable;
		...

loop_disable:
		hub_port_disable(hub, port1, 1);
loop:
		usb_ep0_reinit(udev);
		release_devnum(udev);
		hub_free_dev(udev);
		if (retry_locked) {
			mutex_unlock(hcd->address0_mutex);
			usb_unlock_port(port_dev);
		}
		usb_put_dev(udev);

And usb_put_dev() is defined in usb.c as:

void usb_put_dev(struct usb_device *dev)
{
	if (dev)
		put_device(&dev->dev);
}

So you see, if usb_new_device() returns a nonzero value then 
put_device() _is_ called.

>  As comment of device_add()
> says, 'if device_add() succeeds, you should call device_del() when you
> want to get rid of it. If device_add() has not succeeded, use only
> put_device() to drop the reference count'.

You are correct that if device_add() succeeds and a later call fails, 
then usb_new_device() does not properly call device_del().  Please 
rewrite your patch to fix only that problem.

Alan Stern

Alan Stern<stern@rowland.harvard.edu> wrote:
> Ma Ke <make_ruc2021@163.com> writes:
> > When device_add(&udev->dev) failed, calling put_device() to explicitly
> > release udev->dev. And the routine which calls usb_new_device() does
> > not call put_device() when an error occurs.
> 
> That is wrong.
> 
> usb_new_device() is called by hub_port_connect().  The code does:
> 
> 			status = usb_new_device(udev);
> 			...
> 
> 		if (status)
> 			goto loop_disable;
> 		...
> 
> loop_disable:
> 		hub_port_disable(hub, port1, 1);
> loop:
> 		usb_ep0_reinit(udev);
> 		release_devnum(udev);
> 		hub_free_dev(udev);
> 		if (retry_locked) {
> 			mutex_unlock(hcd->address0_mutex);
> 			usb_unlock_port(port_dev);
> 		}
> 		usb_put_dev(udev);
> 
> And usb_put_dev() is defined in usb.c as:
> 
> void usb_put_dev(struct usb_device *dev)
> {
> 	if (dev)
> 		put_device(&dev->dev);
> }
> 
> So you see, if usb_new_device() returns a nonzero value then 
> put_device() _is_ called.
> 
> >  As comment of device_add()
> > says, 'if device_add() succeeds, you should call device_del() when you
> > want to get rid of it. If device_add() has not succeeded, use only
> > put_device() to drop the reference count'.
> 
> You are correct that if device_add() succeeds and a later call fails, 
> then usb_new_device() does not properly call device_del().  Please 
> rewrite your patch to fix only that problem.
> 
> Alan Stern
Thank you for guiding me on the vulnerability I submitted. I will 
resubmit the patch based on your guidance and suggestions.
--
Regards,

Ma Ke