[PATCH 07/11] coio: fix file descriptor leak on accept

[Vladimir Davydov]

On Fri, Nov 30, 2018 at 06:39:39PM +0300, Vladislav Shpilevoy wrote:
> @@ -90,44 +92,46 @@ evio_setsockopt_keepalive(int fd)
>  	 */
>  	int keepcnt = 5;
>  	if (sio_setsockopt(fd, IPPROTO_TCP, TCP_KEEPCNT, &keepcnt,
> -		       sizeof(int)))
> -		diag_raise();
> +			   sizeof(int)) != 0)
> +		return -1;
>  	int keepidle = 30;
>  
>  	if (sio_setsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &keepidle,
> -		       sizeof(int)))
> -		diag_raise();
> +			   sizeof(int)) != 0)
> +		return -1;
>  
>  	int keepintvl = 60;
>  	if (sio_setsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &keepintvl,
> -		       sizeof(int)))
> -		diag_raise();
> +			   sizeof(int)) != 0)
> +		return -1;
>  #endif
> +	return 0;
>  }
>  
>  /** Set common client socket options. */
> -void
> +int
>  evio_setsockopt_client(int fd, int family, int type)
>  {
>  	int on = 1;
> -	/* In case this throws, the socket is not leaked. */
>  	if (sio_setfl(fd, O_NONBLOCK) != 0)
> -		diag_raise();
> +		return -1;
>  	if (type == SOCK_STREAM && family != AF_UNIX) {
>  		/*
>  		 * SO_KEEPALIVE to ensure connections don't hang
>  		 * around for too long when a link goes away.
>  		 */
> -		evio_setsockopt_keepalive(fd);
> +		if (evio_setsockopt_keepalive(fd) != 0)
> +			return -1;
>  		/*
>  		 * Lower latency is more important than higher
>  		 * bandwidth, and we usually write entire
>  		 * request/response in a single syscall.
>  		 */
>  		if (sio_setsockopt(fd, IPPROTO_TCP, TCP_NODELAY,
> -				   &on, sizeof(on)))
> -			diag_raise();
> +				   &on, sizeof(on)) != 0)
> +			return -1;
>  	}
> +	return 0;
>  }
>  
>  /** Set options for server sockets. */
> @@ -150,8 +154,9 @@ evio_setsockopt_server(int fd, int family, int type)
>  	if (sio_setsockopt(fd, SOL_SOCKET, SO_LINGER,
>  		       &linger, sizeof(linger)))
>  		diag_raise();
> -	if (type == SOCK_STREAM && family != AF_UNIX)
> -		evio_setsockopt_keepalive(fd);
> +	if (type == SOCK_STREAM && family != AF_UNIX &&
> +	    evio_setsockopt_keepalive(fd) != 0)
> +		diag_raise();
>  }

I don't like this patch, because apart from fixing the leak, it also
removes exceptions from some evio functions, but not all of them, e.g.
evio_setsockopt_keepalive and evio_setsockopt_client are now exception
free while evio_setsockopt_server is not. And then you have a separate
patch that removes the rest of exception from evio.

Let's please first remove all exceptions from evio in one go, and only
then fix the leak?