Re: [PATCH] media: v4l: async: Fix NULL pointer when v4l2 flash subdev binding

[ChiYuan Huang]

Hi, Sakari:

	Thanks for your reply.
If any misunderstanding, please correct me.

On Thu, May 16, 2024 at 10:42:05AM +0000, Sakari Ailus wrote:
> Hi Chi Yuan,
> 
> On Wed, May 08, 2024 at 10:51:49AM +0800, cy_huang@xxxxxxxxxxx wrote:
> > From: ChiYuan Huang <cy_huang@xxxxxxxxxxx>
> > 
> > In v4l2_async_create_ancillary_links(), if v4l2 async notifier is
> > created from v4l2 device, the v4l2 flash subdev async binding will enter
> > the logic to create media link. Due to the subdev of notifier is NULL,
> > this will cause NULL pointer to access the subdev entity. Therefore, add
> > the check to bypass it.
> > 
> > Fixes: aa4faf6eb271 ("media: v4l2-async: Create links during v4l2_async_match_notify()")
> > Signed-off-by: ChiYuan Huang <cy_huang@xxxxxxxxxxx>
> > ---
> > Hi,
> > 
> >   I'm trying to bind the v4l2 subdev for flashlight testing. It seems
> > some logic in v4l2 asynd binding is incorrect.
> > 
> > From the change, I modified vim2m as the test driver to bind mt6370 flashlight.
> > 
> > Here's the backtrace log.
> > 
> >  vim2m soc:vim2m: bound [white:flash-2]
> >  Unable to handle kernel NULL pointer dereference at virtual address 0000000000000058
> >  ......skipping
> >  Call trace:
> >   media_create_ancillary_link+0x48/0xd8 [mc]
> >   v4l2_async_match_notify+0x17c/0x208 [v4l2_async]
> >   v4l2_async_register_subdev+0xb8/0x1d0 [v4l2_async]
> 
> There's something wrong obviously somewhere but wherea?
> 
In vim2m driver, I added v4l2_async_nf_init -> v4l2_async_nf_add_fwnode_remote ->
v4l2_async_nf_register.

>From the async flow, in notifier complete ops to create v4l-subdevX node for the 
specified subdev.
> A sub-notifier does have a sub-device after the notifier initialisation.

Why? Are you saying to the notifier can only be used for subdev and subdev binding, 
not v4l2 and subdev binding?

But to create v4l-subdevX, the key is only v4l2 device and its needed subdev.

> Maybe the initialisation does not happen in the right order?
AFAIK, Async flow can solve the probe order and makes the user no need to care
the probe order.

>From the stacktrace, I'm pretty sure it's not the probe order issue.
> 
> >   __v4l2_flash_init.part.0+0x3b4/0x4b0 [v4l2_flash_led_class]
> >   v4l2_flash_init+0x28/0x48 [v4l2_flash_led_class]
> >   mt6370_led_probe+0x348/0x690 [leds_mt6370_flash]
> > 
> > After tracing the code, it will let the subdev labeled as F_LENS or
> > F_FLASH function to create media link. To prevent the NULL pointer
> > issue, the simplest way is add a check when 'n->sd' is NULL and bypass
> > the later media link creataion.
> > ---
> >  drivers/media/v4l2-core/v4l2-async.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/drivers/media/v4l2-core/v4l2-async.c b/drivers/media/v4l2-core/v4l2-async.c
> > index 3ec323bd528b..9d3161c51954 100644
> > --- a/drivers/media/v4l2-core/v4l2-async.c
> > +++ b/drivers/media/v4l2-core/v4l2-async.c
> > @@ -324,6 +324,9 @@ static int v4l2_async_create_ancillary_links(struct v4l2_async_notifier *n,
> >  	    sd->entity.function != MEDIA_ENT_F_FLASH)
> >  		return 0;
> >  
> > +	if (!n->sd)
> > +		return 0;
> 
> This isn't the right fix: the ancillary link won't be created as a result.
> 
Due to the notifier is created by v4l2 device not subdev, this 'n->sd' is NULL.
The NULL 'n->sd' will be referenced by the next flow 'media_create_ancillary_link'.

Or is it caused by the wrong usage? 

> > +
> >  	link = media_create_ancillary_link(&n->sd->entity, &sd->entity);
> >  
> >  #endif
> 
> -- 
> Regards,
> 
> Sakari Ailus