Re: [PATCH] drivers/media/usb/gspca/stv06xx: fix memory leak

[Pavel Skripkin <paskripkin@gmail.com>]

Hi! Thanks for the review.

On Tue, 2021-03-23 at 17:13 +0100, Mauro Carvalho Chehab wrote:
> Em Sat, 27 Feb 2021 02:37:31 +0300
> Pavel Skripkin <paskripkin@gmail.com> escreveu:
> 
> > Syzbot reported memory leak in hdcs_probe_1x00()[1].
> > hdcs_probe_1x00() allocates memory for struct hdcs, but if
> > hdcs_init() fails in gspca_dev_probe2()
> > this memory becomes leaked.
> > 
> > int gspca_dev_probe2(struct usb_interface *intf,
> >                 const struct usb_device_id *id,
> >                 const struct sd_desc *sd_desc,
> >                 int dev_size,
> >                 struct module *module)
> > {
> > ...
> > 
> >         ret = sd_desc->config(gspca_dev, id);
> >         if (ret < 0)
> >                 goto out;
> >         ret = sd_desc->init(gspca_dev);
> >         if (ret < 0)
> >                 goto out;
> > ...
> > out:
> >         if (gspca_dev->input_dev)
> >                 input_unregister_device(gspca_dev->input_dev);
> >         v4l2_ctrl_handler_free(gspca_dev->vdev.ctrl_handler);
> >         v4l2_device_unregister(&gspca_dev->v4l2_dev);
> >         kfree(gspca_dev->usb_buf);
> >         kfree(gspca_dev);
> >         return ret;
> > }
> > 
> > Reported-by: syzbot+e7f4c64a4248a0340c37@syzkaller.appspotmail.com
> > Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
> > Change-Id: Ia198671177ee346de61780813025110c7c491d7a
> > ---
> >  drivers/media/usb/gspca/stv06xx/stv06xx_hdcs.c | 12 ++++++++----
> >  1 file changed, 8 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/media/usb/gspca/stv06xx/stv06xx_hdcs.c
> > b/drivers/media/usb/gspca/stv06xx/stv06xx_hdcs.c
> > index 5a47dcbf1c8e..24df13b33a02 100644
> > --- a/drivers/media/usb/gspca/stv06xx/stv06xx_hdcs.c
> > +++ b/drivers/media/usb/gspca/stv06xx/stv06xx_hdcs.c
> > @@ -485,7 +485,7 @@ static int hdcs_init(struct sd *sd)
> >                                            stv_bridge_init[i][1]);
> >         }
> >         if (err < 0)
> > -               return err;
> > +               goto error;
> >  
> >         /* sensor soft reset */
> >         hdcs_reset(sd);
> > @@ -496,12 +496,12 @@ static int hdcs_init(struct sd *sd)
> >                                              stv_sensor_init[i][1]);
> >         }
> >         if (err < 0)
> > -               return err;
> > +               goto error;
> >  
> >         /* Enable continuous frame capture, bit 2: stop when frame
> > complete */
> >         err = stv06xx_write_sensor(sd, HDCS_REG_CONFIG(sd), BIT(3));
> >         if (err < 0)
> > -               return err;
> > +               goto error;
> >  
> >         /* Set PGA sample duration
> >         (was 0x7E for the STV602, but caused slow framerate with
> > HDCS-1020) */
> > @@ -512,9 +512,13 @@ static int hdcs_init(struct sd *sd)
> >                 err = stv06xx_write_sensor(sd, HDCS_TCTRL,
> >                                 (HDCS_ADC_START_SIG_DUR << 5) | hdcs-
> > >psmp);
> >         if (err < 0)
> > -               return err;
> > +               goto error;
> >  
> >         return hdcs_set_size(sd, hdcs->array.width, hdcs-
> > >array.height);
> > +
> > +error:
> > +       kfree(hdcs);
> > +       return err;
> >  }
> 
> This doesn't seem the right fix here, as it is not the _init function
> that allocates it. Also, when the device is disconnected, a memory
> leak
> will happen.

It won't.

static int hdcs_probe_1x00(struct sd *sd)
{
	....
	sd->sensor_priv = hdcs;
	....
}


static void sd_disconnect(struct usb_interface *intf)
{
	void *priv = sd->sensor_priv;
	....
	kfree(priv);
}

Is it correct?

> 
> I suspect that the right fix would be to move this:
> 
>         hdcs = kmalloc(sizeof(struct hdcs), GFP_KERNEL);
>         if (!hdcs)
>                 return -ENOMEM;
> 
> To the main driver (stv06xx.c) - probably replacing it by kzalloc(),
> 

I don't really understand why, because this allocation refers only to
stv06xx_hdcs, and other stv06xx sensors don't use it.

If hdcs_init() fails, we won't be able to access this pointer, because 
in gspca_dev_probe2() only this code will be executed on error
condition:

	v4l2_ctrl_handler_free(gspca_dev->vdev.ctrl_handler);
	v4l2_device_unregister(&gspca_dev->v4l2_dev);
	kfree(gspca_dev->usb_buf);
	kfree(gspca_dev);

Maybe, I don't properly understand it, can You explain, please?

> and then handle the free code both both sd_probe() and sd_disconnect().
> 
> 
> Thanks,

-- 
With regards,
Pavel Skripkin