Re: [PATCH] nvmem: core: fix error path in nvmem_add_cells()

[Srinivas Kandagatla]

Hi Rasmus,

Thanks for the patch,

On 08/02/16 21:04, Rasmus Villemoes wrote:
> The current code fails to nvmem_cell_drop(cells[0]) - even worse, if
> the loop above fails already at i==0, we'll enter an essentially
> infinite loop doing nvmem_cell_drop on cells[-1], cells[-2], ... which
> is unlikely to end well.
I agree, it would fail in case of zero.
> Also, we're not freeing the temporary backing array cells on the error
> path.
>
> Signed-off-by: Rasmus Villemoes <linux@xxxxxxxxxxxxxxxxxx>
> ---
>   drivers/nvmem/core.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
> index 6fd4e5a5ef4a..1e65eccfea83 100644
> --- a/drivers/nvmem/core.c
> +++ b/drivers/nvmem/core.c
> @@ -288,9 +288,11 @@ static int nvmem_add_cells(struct nvmem_device *nvmem,
>
>   	return 0;
>   err:
> -	while (--i)
> +	while (i--)
>   		nvmem_cell_drop(cells[i]);
No, this will not work.

3 issues,

1> If we enter this err path from nvmem_cell_info_to_nvmem_cell() 
failures, you would be accessing already freed cells[i].

2> accessing un-allocated cells[i].

3> you would be trying to drop cells which are not in the list.


This is what you need here to fix it correctly.

while (--i >= 0)

> +	kfree(cells);
This change looks good.
> +
>   	return rval;
>   }

--srini