[PATCH 07/10] Switch from using os_random() to using os_get_random(...) in advertisement_state_machine_start(...)

[Jouni Malinen]

On Sun, Feb 14, 2016 at 07:39:32PM +0000, Nick Lowe wrote:
>  Switch from using os_random() to using
>  os_get_random(...) in advertisement_state_machine_start(...) The intention is
>  to facilitate the future removal of os_random(). os_random() uses a low
>  quality PRNG which we should avoid using outright unless there is a
>  compelling performance justification to do so.

> diff --git a/src/wps/wps_upnp_ssdp.c b/src/wps/wps_upnp_ssdp.c
> @@ -354,7 +354,10 @@ int advertisement_state_machine_start(struct
>      /* First timeout should be random interval < 100 msec */
> -    next_timeout_msec = (100 * (os_random() & 0xFF)) >> 8;
> +    if (os_get_random((u8 *) &next_timeout_msec,
> +           sizeof(next_timeout_msec)) < 0)
> +        return -1;
> +    next_timeout_msec = (100 * (next_timeout_msec & 0xFF)) >> 8;
>      return eloop_register_timeout(0, next_timeout_msec,

> @@ -475,7 +478,10 @@ static void
> msearchreply_state_machine_start(struct upnp_wps_device_sm *sm,
> -    next_timeout_msec = (1000 * mx * (os_random() & 0xFF)) >> 8;
> +    if (os_get_random((u8 *) &next_timeout_msec,
> +            sizeof(next_timeout_msec)) < 0)
> +        goto fail;
> +    next_timeout_msec = (1000 * mx * (next_timeout_msec & 0xFF)) >> 8;

Both of these would need to be unsigned values to avoid issues with
negative timeout values. In addition, I don't really like the error path
options here.. Likely better to just hardcode a timeout in the unlikely
error case.
 
-- 
Jouni Malinen                                            PGP id EFC895FA