[PATCH 2/3] efi_loader: Prevent dereference of uninitialised variable

[Ilias Apalodimas]

Hi Andrew,


On Tue, 1 Jul 2025 at 16:38, Andrew Goodbody <andrew.goodbody at linaro.org> wrote:
>
> If phandler is returned as NULL from efi_search_protocol then
> protocol_interface is never assigned to. Add a check to prevent
> protocol_interface being dereferenced in this case. Small refactor to
> coalesce the two identical NULL checks of phandler.
>
> This issue found by Smatch.
>
> Signed-off-by: Andrew Goodbody <andrew.goodbody at linaro.org>
> ---
>  lib/efi_loader/efi_http.c | 14 +++++++-------
>  1 file changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/lib/efi_loader/efi_http.c b/lib/efi_loader/efi_http.c
> index 189317fe2d2..ce3a7a831ca 100644
> --- a/lib/efi_loader/efi_http.c
> +++ b/lib/efi_loader/efi_http.c
> @@ -463,18 +463,18 @@ static efi_status_t EFIAPI efi_http_service_binding_destroy_child(
>
>         efi_search_protocol(child_handle, &efi_http_guid, &phandler);
>
> -       if (phandler)
> -               protocol_interface = phandler->protocol_interface;
> -
>         ret = efi_delete_handle(child_handle);
>         if (ret != EFI_SUCCESS)
>                 return EFI_EXIT(ret);
>
> -       http_instance = (struct efi_http_instance *)protocol_interface;
> -       efi_free_pool(http_instance->http_load_addr);
> -       http_instance->http_load_addr = NULL;
> +       if (phandler) {
> +               protocol_interface = phandler->protocol_interface;
> +               http_instance = (struct efi_http_instance *)protocol_interface;
> +               efi_free_pool(http_instance->http_load_addr);
> +               http_instance->http_load_addr = NULL;
>
> -       free(protocol_interface);
> +               free(protocol_interface);
> +       }

Reading at the spec I think we can't return EFI_SUCCESS in that case.
11.6.3 EFI_SERVICE_BINDING_PROTOCOL.DestroyChild() is the chapter that
describes the protocol.
EFI_UNSUPPORTED is supposed to be returned if the ChildHandle does not
support the protocol that is being removed.

So I think you should return a different error if phadnle is indeed NULL

Thanks
/Ilias

>
>         num_instances--;
>
>
> --
> 2.39.5
>