Re: [PATCH 06/10] iio: adc: pac1934: Fix custom attribute allocation and channel bounds
From: Andy Shevchenko
Date: Thu Nov 19 2026 - 12:32:12 EST
On Thu, Nov 19, 2026 at 03:30:39PM +0000, Lee Jones wrote:
> pac1934_prep_custom_attributes() sizes the pac1934_custom_attr pointer
> array using sizeof(*pac1934_group) instead of
> sizeof(*pac1934_custom_attr) and adds 1 byte rather than one pointer
> element for the terminating NULL entry. When active_channels_count is
> zero, only a single byte is allocated and iio_device_register_sysfs()
> reads past the allocation when scanning the attribute list.
>
> In addition, active_channels_count is counted over info->phys_channels,
> whereas the population steps unconditionally inspect all four entries of
> info->active_channels[], which can write past the end of the array on
> 1-, 2-, and 3-channel chip variants.
>
> Allocate active_channels_count + 1 attribute pointers with
> devm_kzalloc_objs() and bound the population loop by
> info->phys_channels.
...
> - int i, active_channels_count = 0;
> + int i, j, active_channels_count = 0;
While at it, can we move to make it more like a reversed xmas tree order?
But see below first.
...
> + j = 0;
> + for (i = 0; i < info->phys_channels; i++)
> + if (info->active_channels[i])
> + pac1934_custom_attr[j++] = pac1934_shunt_attrs[i];
But i is not used outside of the loop, can we rather have
i = 0;
for (int j = 0; j < info->phys_channels; j++)
if (info->active_channels[j])
pac1934_custom_attr[i++] = pac1934_shunt_attrs[j];
? In such a case the above hunk won't be needed.
--
With Best Regards,
Andy Shevchenko