[CRIU] [PATCH 1/3] service: allow to execute page-server as a child process

[Pavel Emelyanov]

On 04/25/2017 12:37 AM, Andrei Vagin wrote:
> From: Andrei Vagin <avagin at virtuozzo.com>

How is it executed now? Not as a child?

> In this case we can wait it and get an exit code.
> 
> For example, it will be useful for p.haul where one connection
> is used several times, so we need a way how to understand  that
> page-server exited unexpectedly.
> 
> Signed-off-by: Andrei Vagin <avagin at virtuozzo.com>
> ---
>  criu/cr-service.c | 85 ++++++++++++++++++++++++++++++++++++++++++-------------
>  images/rpc.proto  |  8 ++++++
>  2 files changed, 74 insertions(+), 19 deletions(-)
> 
> diff --git a/criu/cr-service.c b/criu/cr-service.c
> index d59ab8b..515d040 100644
> --- a/criu/cr-service.c
> +++ b/criu/cr-service.c
> @@ -710,7 +710,7 @@ struct ps_info {
>  	unsigned short port;
>  };
>  
> -static int start_page_server_req(int sk, CriuOpts *req)
> +static int start_page_server_req(int sk, CriuOpts *req, bool daemon_mode)
>  {
>  	int ret = -1, pid, start_pipe[2];
>  	ssize_t count;
> @@ -735,35 +735,46 @@ static int start_page_server_req(int sk, CriuOpts *req)
>  
>  		pr_debug("Starting page server\n");
>  
> -		pid = cr_page_server(true, false, start_pipe[1]);
> +		if (!daemon_mode) {
> +			count = write(start_pipe[1], &info, sizeof(info));
> +			if (count != sizeof(info))
> +				goto out_ch;
> +		}
> +
> +		pid = cr_page_server(daemon_mode, false, start_pipe[1]);
>  		if (pid < 0)
>  			goto out_ch;
>  
> -		info.pid = pid;
> -		info.port = opts.port;
> +		if (daemon_mode) {
> +			info.pid = pid;
> +			info.port = opts.port;
>  
> -		count = write(start_pipe[1], &info, sizeof(info));
> -		if (count != sizeof(info))
> -			goto out_ch;
> +			count = write(start_pipe[1], &info, sizeof(info));
> +			if (count != sizeof(info))
> +				goto out_ch;
> +		}
>  
>  		ret = 0;
>  out_ch:
> -		if (ret < 0 && pid > 0)
> +		if (daemon_mode && ret < 0 && pid > 0)
>  			kill(pid, SIGKILL);
>  		close(start_pipe[1]);
>  		exit(ret);
>  	}
>  
>  	close(start_pipe[1]);
> -	wait(&ret);
> -	if (WIFEXITED(ret)) {
> -		if (WEXITSTATUS(ret)) {
> -			pr_err("Child exited with an error\n");
> +
> +	if (daemon_mode) {
> +		wait(&ret);
> +		if (WIFEXITED(ret)) {
> +			if (WEXITSTATUS(ret)) {
> +				pr_err("Child exited with an error\n");
> +				goto out;
> +			}
> +		} else {
> +			pr_err("Child wasn't terminated normally\n");
>  			goto out;
>  		}
> -	} else {
> -		pr_err("Child wasn't terminated normally\n");
> -		goto out;
>  	}
>  
>  	count = read(start_pipe[0], &info, sizeof(info));
> @@ -771,11 +782,15 @@ out_ch:
>  	if (count != sizeof(info))
>  		goto out;
>  
> +	if (daemon_mode) {
> +		ps.pid = info.pid;
> +		ps.has_port = true;
> +		ps.port = info.port;
> +	} else
> +		ps.pid = pid;
> +
>  	success = true;
>  	ps.has_pid = true;
> -	ps.pid = info.pid;
> -	ps.has_port = true;
> -	ps.port = info.port;
>  	resp.ps = &ps;
>  
>  	pr_debug("Page server started\n");
> @@ -802,6 +817,9 @@ static int chk_keepopen_req(CriuReq *msg)
>  	if (msg->type == CRIU_REQ_TYPE__PAGE_SERVER)
>  		/* This just fork()-s so no leaks */
>  		return 0;
> +	if (msg->type == CRIU_REQ_TYPE__PAGE_SERVER_CHLD)
> +		/* This just fork()-s so no leaks */
> +		return 0;
>  	else if (msg->type == CRIU_REQ_TYPE__CPUINFO_DUMP ||
>  		 msg->type == CRIU_REQ_TYPE__CPUINFO_CHECK)
>  		return 0;
> @@ -946,6 +964,29 @@ out:
>  	return send_criu_msg(sk, &resp);
>  }
>  
> +static int handle_wait_pid(int sk, int pid)
> +{
> +	CriuResp resp = CRIU_RESP__INIT;
> +	bool success = false;
> +	int status;
> +
> +	if (waitpid(pid, &status, 0) == -1) {
> +		resp.cr_errno = errno;
> +		pr_perror("Unable to wait %d", pid);
> +		goto out;
> +	}
> +
> +	resp.status = status;
> +	resp.has_status = true;
> +
> +	success = true;
> +out:
> +	resp.type = CRIU_REQ_TYPE__WAIT_PID;
> +	resp.success = success;
> +
> +	return send_criu_msg(sk, &resp);
> +}
> +
>  static int handle_cpuinfo(int sk, CriuReq *msg)
>  {
>  	CriuResp resp = CRIU_RESP__INIT;
> @@ -1031,7 +1072,13 @@ more:
>  		ret = pre_dump_loop(sk, msg);
>  		break;
>  	case CRIU_REQ_TYPE__PAGE_SERVER:
> -		ret =  start_page_server_req(sk, msg->opts);
> +		ret = start_page_server_req(sk, msg->opts, true);
> +		break;
> +	case CRIU_REQ_TYPE__PAGE_SERVER_CHLD:
> +		ret = start_page_server_req(sk, msg->opts, false);
> +		break;
> +	case CRIU_REQ_TYPE__WAIT_PID:
> +		ret =  handle_wait_pid(sk, msg->pid);
>  		break;
>  	case CRIU_REQ_TYPE__CPUINFO_DUMP:
>  	case CRIU_REQ_TYPE__CPUINFO_CHECK:
> diff --git a/images/rpc.proto b/images/rpc.proto
> index 48e42e2..71f47d5 100644
> --- a/images/rpc.proto
> +++ b/images/rpc.proto
> @@ -142,6 +142,9 @@ enum criu_req_type {
>  	FEATURE_CHECK	= 9;
>  
>  	VERSION		= 10;
> +
> +	WAIT_PID	= 11;
> +	PAGE_SERVER_CHLD = 12;
>  }
>  
>  /*
> @@ -176,6 +179,9 @@ message criu_req {
>  	 * via RPC.
>  	 */
>  	optional criu_features		features	= 5;
> +
> +	/* 'pid' is used for WAIT_PID */
> +	optional uint32			pid		= 6;
>  }
>  
>  /*
> @@ -196,6 +202,8 @@ message criu_resp {
>  	optional criu_features		features	= 8;
>  	optional string			cr_errmsg	= 9;
>  	optional criu_version		version		= 10;
> +
> +	optional int32			status		= 11;
>  }
>  
>  /* Answer for criu_req_type.VERSION requests */
>