[PATCH] text-backend: check calloc results

[Pekka Paalanen]

On Thu, 15 May 2014 13:03:50 -0700
"U. Artie Eoff" <ullysses.a.eoff at intel.com> wrote:

> Signed-off-by: U. Artie Eoff <ullysses.a.eoff at intel.com>
> ---
>  src/text-backend.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/src/text-backend.c b/src/text-backend.c
> index d6a6f3b..362a3fa 100644
> --- a/src/text-backend.c
> +++ b/src/text-backend.c
> @@ -358,6 +358,8 @@ static void text_input_manager_create_text_input(struct wl_client *client,
>  	struct text_input *text_input;
>  
>  	text_input = calloc(1, sizeof *text_input);
> +	if (text_input == NULL)
> +		return;

Silent failure here is a no-go, it confuses the client, as server vs.
client state on what objects exist disagrees, and the client will get a
random protocol error eventually.

Better to just use wl_resource_post_no_memory() here to kill the client
immediately.

>  
>  	text_input->resource =
>  		wl_resource_create(client, &wl_text_input_interface, 1, id);
> @@ -411,6 +413,9 @@ text_input_manager_create(struct weston_compositor *ec)
>  
>  	text_input_manager = calloc(1, sizeof *text_input_manager);
>  
> +	if (text_input_manager == NULL)
> +		return;

We're already pretty screwed if this fails anyway, but is hiding the
failure ok?

> +
>  	text_input_manager->ec = ec;
>  
>  	text_input_manager->text_input_manager_global =

wl_global_create() could fail, too.

> @@ -917,6 +922,9 @@ handle_seat_created(struct wl_listener *listener,
>  
>  	input_method = calloc(1, sizeof *input_method);
>  
> +	if (input_method == NULL)
> +		return;
> +

Ditto.

>  	input_method->seat = seat;
>  	input_method->model = NULL;
>  	input_method->focus_listener_initialized = 0;
> @@ -969,6 +977,9 @@ text_backend_init(struct weston_compositor *ec)
>  
>  	text_backend = calloc(1, sizeof(*text_backend));
>  
> +	if (text_backend == NULL)
> +		return -1;
> +
>  	text_backend->compositor = ec;
>  
>  	text_backend->seat_created_listener.notify = handle_seat_created;

Looks like lots of places are just hiding failures instead of
communicating them to the caller, or at least yelling about it.

Still, wl_resource_post_no_memory() is the only thing I'd want to see
fixed in this patch before it gets in, the others are not so important
as they are not client-triggerable.


Thanks,
pq