Re: NULL pointer dereference in iio_buffer_register()

[Fengguang Wu <fengguang.wu@intel.com>]

On Tue, Jul 31, 2012 at 03:06:11PM +0200, Lars-Peter Clausen wrote:
> On 07/31/2012 02:55 PM, Fengguang Wu wrote:
> >>> The panic happens while trying to dereference the NULL indio_dev->buffer:
> >>>
> >>>         266 int iio_buffer_register(struct iio_dev *indio_dev,
> >>>         267                         const struct iio_chan_spec *channels,
> >>>         268                         int num_channels)
> >>>         269 {  
> >>>         270         struct iio_dev_attr *p;
> >>>         271         struct attribute **attr;
> >>>         272         struct iio_buffer *buffer = indio_dev->buffer;
> >>>         273         int ret, i, attrn, attrcount, attrcount_orig = 0;
> >>>         274 
> >>> ==>     275         if (buffer->attrs)
> >>>         276                 indio_dev->groups[indio_dev->groupcounter++] = buffer->attrs;
> >>>
> >>> iio_dummy_probe() has the code to configure that buffer, however
> >>> iio_simple_dummy_configure_buffer() is defined to do nothing on
> >>> !CONFIG_IIO_SIMPLE_DUMMY_BUFFER..
> >>>
> >>>         448         /* Configure buffered capture support. */
> >>> ==>     449         ret = iio_simple_dummy_configure_buffer(indio_dev);
> >>>         450         if (ret < 0)
> >>>         451                 goto error_unregister_events;
> >>>         452 
> >>>         453         /*
> >>>         454          * Register the channels with the buffer, but avoid the output
> >>>         455          * channel being registered by reducing the number of channels by 1.
> >>>         456          */
> >>>         457         ret = iio_buffer_register(indio_dev, iio_dummy_channels, 5);
> >>>         458         if (ret < 0)
> >>>         459                 goto error_unconfigure_buffer;
> >>>
> >>> Any ideas to fix it?
> >>>
> >>
> >> Hi,
> >>
> >> I think the best would be to move the iio_buffer_register to
> >> iio_simple_dummy_configure_buffer.
> > 
> > Lars, thanks for the quick reply! Hmm, that looks more like a code
> > refactor recommendation than fix ;)  In the simplest form, can the
> > bug fixed like this?
> > 
> >   static inline int iio_simple_dummy_configure_buffer(struct iio_dev *indio_dev)
> >   {
> > -         return 0;
> > +         return -1;
> >   };
> > 
> 
> No, we want iio_simple_dummy_configure_buffer to be a noop if buffer support
> is disabled, since the driver works fine without it. Except for the issue
> you discovered. This issue only appears if CONFIG_IIO_BUFFER=y and
> CONFIG_IIO_SIMPLE_DUMMY_BUFFER=n. E.g. if both are not set the driver works
> fine without buffers.

OK.

> I can prepare a patch which moves the iio_buffer_register to
> iio_simple_dummy_configure_buffer if you want to.

That would be good, thank you very much!

Thanks,
Fengguang