X-Git-Url: https://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-relayd%2Fmain.c;h=fe4a898bc6d349dfd866218e597d77ef12a746aa;hb=719af1bb6e6719c6170e059a7e473aa7923ddae9;hp=7ac5630fb9790262b7355df43f7066deadfb671b;hpb=b5218ffbf53aa46007927f1e2cda0ad4549abcf9;p=lttng-tools.git

diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c
index 7ac5630fb..fe4a898bc 100644
--- a/src/bin/lttng-relayd/main.c
+++ b/src/bin/lttng-relayd/main.c
@@ -45,6 +45,7 @@
 #include <common/compat/poll.h>
 #include <common/compat/socket.h>
 #include <common/defaults.h>
+#include <common/daemonize.h>
 #include <common/futex.h>
 #include <common/sessiond-comm/sessiond-comm.h>
 #include <common/sessiond-comm/inet.h>
@@ -60,10 +61,21 @@
 #include "lttng-relayd.h"
 #include "live.h"
 #include "health-relayd.h"
+#include "testpoint.h"
 
 /* command line options */
 char *opt_output_path;
 static int opt_daemon, opt_background;
+
+/*
+ * We need to wait for listener and live listener threads, as well as
+ * health check thread, before being ready to signal readiness.
+ */
+#define NR_LTTNG_RELAY_READY	3
+static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
+static int recv_child_signal;	/* Set to 1 when a SIGUSR1 signal is received. */
+static pid_t child_ppid;	/* Internal parent PID use with daemonize. */
+
 static struct lttng_uri *control_uri;
 static struct lttng_uri *data_uri;
 static struct lttng_uri *live_uri;
@@ -79,7 +91,7 @@ const char * const config_section_name = "relayd";
  * Quit pipe for all threads. This permits a single cancellation point
  * for all threads when receiving an event on the pipe.
  */
-static int thread_quit_pipe[2] = { -1, -1 };
+int thread_quit_pipe[2] = { -1, -1 };
 
 /*
  * This pipe is used to inform the worker thread that a command is queued and
@@ -509,6 +521,9 @@ void sighandler(int sig)
 		DBG("SIGTERM caught");
 		stop_threads();
 		break;
+	case SIGUSR1:
+		CMM_STORE_SHARED(recv_child_signal, 1);
+		break;
 	default:
 		break;
 	}
@@ -548,11 +563,26 @@ int set_signal_handler(void)
 		return ret;
 	}
 
-	DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
+	if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
+		PERROR("sigaction");
+		return ret;
+	}
+
+	DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
 
 	return ret;
 }
 
+void lttng_relay_notify_ready(void)
+{
+	/* Notify the parent of the fork() process that we are ready. */
+	if (opt_daemon || opt_background) {
+		if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
+			kill(child_ppid, SIGUSR1);
+		}
+	}
+}
+
 /*
  * Init thread quit pipe.
  *
@@ -587,7 +617,7 @@ int create_thread_poll_set(struct lttng_poll_event *events, int size)
 	}
 
 	/* Add quit pipe */
-	ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
+	ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
 	if (ret < 0) {
 		goto error;
 	}
@@ -722,6 +752,12 @@ void *relay_thread_listener(void *data)
 		goto error_poll_add;
 	}
 
+	lttng_relay_notify_ready();
+
+	if (testpoint(relayd_thread_listener)) {
+		goto error_testpoint;
+	}
+
 	while (1) {
 		health_code_update();
 
@@ -822,6 +858,7 @@ restart:
 exit:
 error:
 error_poll_add:
+error_testpoint:
 	lttng_poll_clean(&events);
 error_create_poll:
 	if (data_sock->fd >= 0) {
@@ -865,6 +902,10 @@ void *relay_thread_dispatcher(void *data)
 
 	health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
 
+	if (testpoint(relayd_thread_dispatcher)) {
+		goto error_testpoint;
+	}
+
 	health_code_update();
 
 	while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
@@ -911,6 +952,7 @@ void *relay_thread_dispatcher(void *data)
 	err = 0;
 
 error:
+error_testpoint:
 	if (err) {
 		health_error();
 		ERR("Health error occurred in %s", __func__);
@@ -1123,6 +1165,10 @@ int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
 	reply.session_id = htobe64(session->id);
 
 	switch (cmd->minor) {
+		case 1:
+		case 2:
+		case 3:
+			break;
 		case 4: /* LTTng sessiond 2.4 */
 		default:
 			ret = cmd_create_session_2_4(cmd, session);
@@ -2529,6 +2575,10 @@ void *relay_thread_worker(void *data)
 
 	health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
 
+	if (testpoint(relayd_thread_worker)) {
+		goto error_testpoint;
+	}
+
 	health_code_update();
 
 	/* table of connections indexed on socket */
@@ -2790,6 +2840,7 @@ relay_connections_ht_error:
 	}
 	DBG("Worker thread cleanup complete");
 	free(data_buffer);
+error_testpoint:
 	if (err) {
 		health_error();
 		ERR("Health error occurred in %s", __func__);
@@ -2822,11 +2873,6 @@ int main(int argc, char **argv)
 	void *status;
 	struct relay_local_data *relay_ctx;
 
-	/* Create thread quit pipe */
-	if ((ret = init_thread_quit_pipe()) < 0) {
-		goto error;
-	}
-
 	/* Parse arguments */
 	progname = argv[0];
 	if ((ret = set_options(argc, argv)) < 0) {
@@ -2853,11 +2899,27 @@ int main(int argc, char **argv)
 
 	/* Daemonize */
 	if (opt_daemon || opt_background) {
-		ret = daemon(0, opt_background);
+		int i;
+
+		ret = lttng_daemonize(&child_ppid, &recv_child_signal,
+			!opt_background);
 		if (ret < 0) {
-			PERROR("daemon");
 			goto exit;
 		}
+
+		/*
+		 * We are in the child. Make sure all other file
+		 * descriptors are closed, in case we are called with
+		 * more opened file descriptors than the standard ones.
+		 */
+		for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
+			(void) close(i);
+		}
+	}
+
+	/* Create thread quit pipe */
+	if ((ret = init_thread_quit_pipe()) < 0) {
+		goto error;
 	}
 
 	/* We need those values for the file/dir creation. */
@@ -2956,7 +3018,7 @@ int main(int argc, char **argv)
 		goto exit_listener;
 	}
 
-	ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe);
+	ret = live_start_threads(live_uri, relay_ctx);
 	if (ret != 0) {
 		ERR("Starting live viewer threads");
 		goto exit_live;