Re: [PATCH] of: overlay: validate offset from property fixups

[Frank Rowand]

Hi Rob,

On 05/16/18 21:19, frowand.list@xxxxxxxxx wrote:
> From: Frank Rowand <frank.rowand@xxxxxxxx>
> 
> The smatch static checker marks the data in offset as untrusted,
> leading it to warn:
> 
>   drivers/of/resolver.c:125 update_usages_of_a_phandle_reference()
>   error: buffer underflow 'prop->value' 's32min-s32max'
> 
> Add check to verify that offset is within the property data.
> 
> Reported-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> Signed-off-by: Frank Rowand <frank.rowand@xxxxxxxx>
> ---
>  drivers/of/resolver.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/of/resolver.c b/drivers/of/resolver.c
> index 65d0b7adfcd4..7edfac6f1914 100644
> --- a/drivers/of/resolver.c
> +++ b/drivers/of/resolver.c
> @@ -122,6 +122,11 @@ static int update_usages_of_a_phandle_reference(struct device_node *overlay,
>  			goto err_fail;
>  		}
>  
> +		if (offset < 0 || offset + sizeof(__be32) > prop->length) {
> +			err = -EINVAL;
> +			goto err_fail;
> +		}
> +
>  		*(__be32 *)(prop->value + offset) = cpu_to_be32(phandle);
>  	}
>  
> 

I should have mentioned that this results in a new compile warning
for W=2 and W=3.  The new if statement results in:

drivers/of/resolver.c:125:45: warning: comparison between signed and unsigned integer expressions [-Wsign-compare]

There are other pre-existing warnings in the same file for comparing
an integer to prop->length.  The correct solution is probably to
change the type of the length field in struct property to be
unsigned.  I have added that task to my todo list.

-Frank