Re: [PATCH] staging: media: tegra-video: fix device_node use after free

[Luca Ceresoli <luca.ceresoli@bootlin.com>]

Hello Dan,

On Fri, 28 Oct 2022 20:06:34 +0300
Dan Carpenter <dan.carpenter@oracle.com> wrote:

> On Fri, Oct 28, 2022 at 06:58:47PM +0200, Luca Ceresoli wrote:
> > Hello Dan,
> > 
> > On Fri, 28 Oct 2022 15:34:39 +0300
> > Dan Carpenter <dan.carpenter@oracle.com> wrote:
> >   
> > > On Fri, Oct 28, 2022 at 10:19:26AM +0200, luca.ceresoli@bootlin.com wrote:  
> > > > From: Luca Ceresoli <luca.ceresoli@bootlin.com>
> > > > 
> > > > At probe time this code path is followed:
> > > > 
> > > >  * tegra_csi_init
> > > >    * tegra_csi_channels_alloc
> > > >      * for_each_child_of_node(node, channel) -- iterates over channels
> > > >        * automatically gets 'channel'
> > > >          * tegra_csi_channel_alloc()
> > > >            * saves into chan->of_node a pointer to the channel OF node
> > > >        * automatically gets and puts 'channel'
> > > >        * now the node saved in chan->of_node has refcount 0, can disappear
> > > >    * tegra_csi_channels_init
> > > >      * iterates over channels
> > > >        * tegra_csi_channel_init -- uses chan->of_node
> > > > 
> > > > After that, chan->of_node keeps storing the node until the device is
> > > > removed.
> > > > 
> > > > of_node_get() the node and of_node_put() it during teardown to avoid any
> > > > risk.
> > > > 
> > > > Fixes: 1ebaeb09830f ("media: tegra-video: Add support for external sensor capture")
> > > > Cc: stable@vger.kernel.org
> > > > Cc: Sowjanya Komatineni <skomatineni@nvidia.com>
> > > > Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
> > > > ---
> > > >  drivers/staging/media/tegra-video/csi.c | 3 ++-
> > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/staging/media/tegra-video/csi.c b/drivers/staging/media/tegra-video/csi.c
> > > > index b26e44adb2be..1b05f620b476 100644
> > > > --- a/drivers/staging/media/tegra-video/csi.c
> > > > +++ b/drivers/staging/media/tegra-video/csi.c
> > > > @@ -433,7 +433,7 @@ static int tegra_csi_channel_alloc(struct tegra_csi *csi,
> > > >  	for (i = 0; i < chan->numgangports; i++)
> > > >  		chan->csi_port_nums[i] = port_num + i * CSI_PORTS_PER_BRICK;
> > > >  
> > > > -	chan->of_node = node;
> > > > +	chan->of_node = of_node_get(node);
> > > >  	chan->numpads = num_pads;
> > > >  	if (num_pads & 0x2) {
> > > >  		chan->pads[0].flags = MEDIA_PAD_FL_SINK;
> > > > @@ -640,6 +640,7 @@ static void tegra_csi_channels_cleanup(struct tegra_csi *csi)
> > > >  			media_entity_cleanup(&subdev->entity);
> > > >  		}
> > > >  
> > > > +		of_node_put(chan->of_node);
> > > >  		list_del(&chan->list);
> > > >  		kfree(chan);    
> > > 
> > > Not related to your patch, but this kind of "one function cleans up
> > > everything" style is always buggy.  For example, here it should be:
> > > 
> > > -		if (chan->mipi)
> > > +		if (!IS_ERR_OR_NULL(chan->mipi))
> > > 			tegra_mipi_free(chan->mipi);  
> > 
> > I sort of agree the code could be clearer here, but looking at the code
> > in detail, this cannot happen. chan->mipi is set in one place only, and
> > if it is an error the whole probe fails. So it can be either NULL or a
> > valid pointer here.  
> 
> I assumed that tegra_csi_channels_cleanup() would clean up if
> tegra_csi_channel_alloc() fails.  Otherwise then that's several
> different even worse bugs.
> 
> HINT: Let's all hope my initial analysis was correct.

Indeed you have a point. Apologies for having replied in a hurry and for
the resulting noise.

I'm going to send a patch to fix the chan->mipi mess. However I think
the best way to do so would be a oneliner:

@@ -491,6 +491,7 @@ static int tegra_csi_channel_alloc(struct tegra_csi *csi,
        chan->mipi = tegra_mipi_request(csi->dev, node);
        if (IS_ERR(chan->mipi)) {
                ret = PTR_ERR(chan->mipi);
+               chan->mipi = NULL;
                dev_err(csi->dev, "failed to get mipi device: %d\n", ret);
        }

This would be a correct implementation of the initial intent as it can
be inferred from the code, i.e. chan->mipi being either NULL or a valid
pointer. This makes sense as chan->mipi is assigned in a single place.

-- 
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com