Re: [PATCH 2/2] USB: dwc3: fix use-after-free on core driver unbind

[Johan Hovold <johan@kernel.org>]

On Thu, Jun 08, 2023 at 06:53:13PM +0530, Manivannan Sadhasivam wrote:
> On Thu, Jun 08, 2023 at 03:09:27PM +0200, Johan Hovold wrote:
> > On Thu, Jun 08, 2023 at 06:32:46PM +0530, Manivannan Sadhasivam wrote:
> > > On Wed, Jun 07, 2023 at 12:05:40PM +0200, Johan Hovold wrote:
> > > > Some dwc3 glue drivers are currently accessing the driver data of the
> > > > child core device directly, which is clearly a bad idea as the child may
> > > > not have probed yet or may have been unbound from its driver.
> > > > 
> > > > As a workaround until the glue drivers have been fixed, clear the driver
> > > > data pointer before allowing the glue parent device to runtime suspend
> > > > to prevent its driver from accessing data that has been freed during
> > > > unbind.
> > 
> > > > @@ -1929,6 +1929,11 @@ static int dwc3_remove(struct platform_device *pdev)
> > > >  	pm_runtime_disable(&pdev->dev);
> > > >  	pm_runtime_dont_use_autosuspend(&pdev->dev);
> > > >  	pm_runtime_put_noidle(&pdev->dev);
> > > > +	/*
> > > > +	 * HACK: Clear the driver data, which is currently accessed by parent
> > > > +	 * glue drivers, before allowing the parent to suspend.
> > > > +	 */
> > > > +	platform_set_drvdata(pdev, NULL);
> > > 
> > > This is required because you have seen the glue driver going to runtime suspend
> > > once the below pm_runtime_set_suspended() is completed?
> > 
> > This is based on analysis of the code. The parent (glue) can not suspend
> > while the child (core) is in the active state, but once we set the
> > suspended state that could happen.
> 
> I could see that the driver core is setting drvdata to NULL during
> device_unbind_cleanup(), so not sure if this scenario could be met otherwise it
> will be redundant.

If this was redundant I wouldn't have added it. ;)

The parent driver has no business accessing the driver data of the child
in the first place, but it must absolutely not do so after the child has
been unbound from its driver and the driver data is getting freed.

Relying on the clean up in driver core that resets this pointer does not
work as that would still leave a window where the parent could access
this stale data.

Johan