Re: [RFC PATCH] spi: fix use-after-free of the add_lock mutex

[Uwe Kleine-König]

Hello,

On Wed, Nov 10, 2021 at 05:08:36PM +0100, Michael Walle wrote:
> Commit 6098475d4cb4 ("spi: Fix deadlock when adding SPI controllers on
> SPI buses") introduced a per-controller mutex. But mutex_unlock() of
> said lock is called after the controller is already freed:
> 
>   spi_unregister_controller(ctlr)
>    -> put_device(&ctlr->dev)
>     -> spi_controller_release(dev)
>   mutex_unlock(&ctrl->add_lock)

This is indented in a misleading way. mutex_unlock() has to be on the
same level as put_device().

> Move the put_device() after the mutex_unlock().
> 
> Fixes: 6098475d4cb4 ("spi: Fix deadlock when adding SPI controllers on SPI buses")
> Signed-off-by: Michael Walle <michael@xxxxxxxx>

I first thought this was wrong, and the put_device must be dropped
altogether, but after some code reading I agree this is the right fix.

Reviewed-by: Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxxxx>

> ---
> I'm not sure if this is the correct fix. I don't know if the put_device()
> will have to be protected by the add_lock (remember before, the add_lock
> was a global lock).

No, put_device doesn't need to be protected by this lock.

Best regards and thanks for the report and diagnosis,
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

Attachment: signature.asc
Description: PGP signature