X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=550aa1b83f68d65f01e936d1e81fb5308b0b2b4e;hp=f3b007a4fac7b1fc3cf17ad1927c1264c375c1d0;hb=55d097957f5bb8138959ad2202a40d85d49f029e;hpb=67e056446003680180ed4f384217facd90b1a776

diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c
index f3b007a4f..550aa1b83 100644
--- a/src/bin/lttng-sessiond/main.c
+++ b/src/bin/lttng-sessiond/main.c
@@ -60,7 +60,7 @@
 #include "ust-consumer.h"
 #include "utils.h"
 #include "fd-limit.h"
-#include "health.h"
+#include "health-sessiond.h"
 #include "testpoint.h"
 #include "ust-thread.h"
 
@@ -233,6 +233,9 @@ static int app_socket_timeout;
 /* Set in main() with the current page size. */
 long page_size;
 
+/* Application health monitoring */
+struct health_app *health_sessiond;
+
 static
 void setup_consumerd_path(void)
 {
@@ -755,7 +758,7 @@ static void *thread_manage_kernel(void *data)
 
 	DBG("[thread] Thread manage kernel started");
 
-	health_register(HEALTH_TYPE_KERNEL);
+	health_register(health_sessiond, HEALTH_TYPE_KERNEL);
 
 	/*
 	 * This first step of the while is to clean this structure which could free
@@ -880,7 +883,7 @@ error_testpoint:
 		WARN("Kernel thread died unexpectedly. "
 				"Kernel tracing can continue but CPU hotplug is disabled.");
 	}
-	health_unregister();
+	health_unregister(health_sessiond);
 	DBG("Kernel thread dying");
 	return NULL;
 }
@@ -921,7 +924,7 @@ static void *thread_manage_consumer(void *data)
 
 	DBG("[thread] Manage consumer started");
 
-	health_register(HEALTH_TYPE_CONSUMER);
+	health_register(health_sessiond, HEALTH_TYPE_CONSUMER);
 
 	health_code_update();
 
@@ -1200,7 +1203,7 @@ error_poll:
 		health_error();
 		ERR("Health error occurred in %s", __func__);
 	}
-	health_unregister();
+	health_unregister(health_sessiond);
 	DBG("consumer thread cleanup completed");
 
 	return NULL;
@@ -1220,7 +1223,7 @@ static void *thread_manage_apps(void *data)
 	rcu_register_thread();
 	rcu_thread_online();
 
-	health_register(HEALTH_TYPE_APP_MANAGE);
+	health_register(health_sessiond, HEALTH_TYPE_APP_MANAGE);
 
 	if (testpoint(thread_manage_apps)) {
 		goto error_testpoint;
@@ -1366,7 +1369,7 @@ error_testpoint:
 		health_error();
 		ERR("Health error occurred in %s", __func__);
 	}
-	health_unregister();
+	health_unregister(health_sessiond);
 	DBG("Application communication apps thread cleanup complete");
 	rcu_thread_offline();
 	rcu_unregister_thread();
@@ -1377,6 +1380,9 @@ error_testpoint:
  * Send a socket to a thread This is called from the dispatch UST registration
  * thread once all sockets are set for the application.
  *
+ * The sock value can be invalid, we don't really care, the thread will handle
+ * it and make the necessary cleanup if so.
+ *
  * On success, return 0 else a negative value being the errno message of the
  * write().
  */
@@ -1384,9 +1390,14 @@ static int send_socket_to_thread(int fd, int sock)
 {
 	int ret;
 
-	/* Sockets MUST be set or else this should not have been called. */
-	assert(fd >= 0);
-	assert(sock >= 0);
+	/*
+	 * It's possible that the FD is set as invalid with -1 concurrently just
+	 * before calling this function being a shutdown state of the thread.
+	 */
+	if (fd < 0) {
+		ret = -EBADF;
+		goto error;
+	}
 
 	do {
 		ret = write(fd, &sock, sizeof(sock));
@@ -1504,7 +1515,7 @@ static void *thread_dispatch_ust_registration(void *data)
 		.count = 0,
 	};
 
-	health_register(HEALTH_TYPE_APP_REG_DISPATCH);
+	health_register(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH);
 
 	health_code_update();
 
@@ -1649,7 +1660,12 @@ static void *thread_dispatch_ust_registration(void *data)
 				if (ret < 0) {
 					rcu_read_unlock();
 					session_unlock_list();
-					/* No notify thread, stop the UST tracing. */
+					/*
+					 * No notify thread, stop the UST tracing. However, this is
+					 * not an internal error of the this thread thus setting
+					 * the health error code to a normal exit.
+					 */
+					err = 0;
 					goto error;
 				}
 
@@ -1674,7 +1690,12 @@ static void *thread_dispatch_ust_registration(void *data)
 				if (ret < 0) {
 					rcu_read_unlock();
 					session_unlock_list();
-					/* No apps. thread, stop the UST tracing. */
+					/*
+					 * No apps. thread, stop the UST tracing. However, this is
+					 * not an internal error of the this thread thus setting
+					 * the health error code to a normal exit.
+					 */
+					err = 0;
 					goto error;
 				}
 
@@ -1705,7 +1726,7 @@ error:
 		health_error();
 		ERR("Health error occurred in %s", __func__);
 	}
-	health_unregister();
+	health_unregister(health_sessiond);
 	return NULL;
 }
 
@@ -1725,7 +1746,7 @@ static void *thread_registration_apps(void *data)
 
 	DBG("[thread] Manage application registration started");
 
-	health_register(HEALTH_TYPE_APP_REG);
+	health_register(health_sessiond, HEALTH_TYPE_APP_REG);
 
 	if (testpoint(thread_registration_apps)) {
 		goto error_testpoint;
@@ -1905,7 +1926,7 @@ error_listen:
 error_create_poll:
 error_testpoint:
 	DBG("UST Registration thread cleanup complete");
-	health_unregister();
+	health_unregister(health_sessiond);
 
 	return NULL;
 }
@@ -2282,7 +2303,7 @@ static int check_consumer_health(void)
 {
 	int ret;
 
-	ret = health_check_state(HEALTH_TYPE_CONSUMER);
+	ret = health_check_state(health_sessiond, HEALTH_TYPE_CONSUMER);
 
 	DBG3("Health consumer check %d", ret);
 
@@ -2450,6 +2471,7 @@ static int create_ust_session(struct ltt_session *session,
 	lus->gid = session->gid;
 	lus->output_traces = session->output_traces;
 	lus->snapshot_mode = session->snapshot_mode;
+	lus->live_timer_interval = session->live_timer;
 	session->ust_session = lus;
 
 	/* Copy session output to the newly created UST session */
@@ -2564,6 +2586,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
 	switch (cmd_ctx->lsm->cmd_type) {
 	case LTTNG_CREATE_SESSION:
 	case LTTNG_CREATE_SESSION_SNAPSHOT:
+	case LTTNG_CREATE_SESSION_LIVE:
 	case LTTNG_DESTROY_SESSION:
 	case LTTNG_LIST_SESSIONS:
 	case LTTNG_LIST_DOMAINS:
@@ -2627,6 +2650,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
 	switch (cmd_ctx->lsm->cmd_type) {
 	case LTTNG_CREATE_SESSION:
 	case LTTNG_CREATE_SESSION_SNAPSHOT:
+	case LTTNG_CREATE_SESSION_LIVE:
 	case LTTNG_CALIBRATE:
 	case LTTNG_LIST_SESSIONS:
 	case LTTNG_LIST_TRACEPOINTS:
@@ -3071,7 +3095,7 @@ skip_domain:
 		}
 
 		ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri,
-			&cmd_ctx->creds);
+			&cmd_ctx->creds, 0);
 
 		free(uris);
 
@@ -3361,6 +3385,45 @@ skip_domain:
 		free(uris);
 		break;
 	}
+	case LTTNG_CREATE_SESSION_LIVE:
+	{
+		size_t nb_uri, len;
+		struct lttng_uri *uris = NULL;
+
+		nb_uri = cmd_ctx->lsm->u.uri.size;
+		len = nb_uri * sizeof(struct lttng_uri);
+
+		if (nb_uri > 0) {
+			uris = zmalloc(len);
+			if (uris == NULL) {
+				ret = LTTNG_ERR_FATAL;
+				goto error;
+			}
+
+			/* Receive variable len data */
+			DBG("Waiting for %zu URIs from client ...", nb_uri);
+			ret = lttcomm_recv_unix_sock(sock, uris, len);
+			if (ret <= 0) {
+				DBG("No URIs received from client... continuing");
+				*sock_error = 1;
+				ret = LTTNG_ERR_SESSION_FAIL;
+				free(uris);
+				goto error;
+			}
+
+			if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) {
+				DBG("Creating session with ONE network URI is a bad call");
+				ret = LTTNG_ERR_SESSION_FAIL;
+				free(uris);
+				goto error;
+			}
+		}
+
+		ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris,
+				nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval);
+		free(uris);
+		break;
+	}
 	default:
 		ret = LTTNG_ERR_UND;
 		break;
@@ -3504,39 +3567,39 @@ restart:
 
 		switch (msg.component) {
 		case LTTNG_HEALTH_CMD:
-			reply.ret_code = health_check_state(HEALTH_TYPE_CMD);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_CMD);
 			break;
 		case LTTNG_HEALTH_APP_MANAGE:
-			reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE);
 			break;
 		case LTTNG_HEALTH_APP_REG:
-			reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_REG);
 			break;
 		case LTTNG_HEALTH_KERNEL:
-			reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_KERNEL);
 			break;
 		case LTTNG_HEALTH_CONSUMER:
 			reply.ret_code = check_consumer_health();
 			break;
 		case LTTNG_HEALTH_HT_CLEANUP:
-			reply.ret_code = health_check_state(HEALTH_TYPE_HT_CLEANUP);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_HT_CLEANUP);
 			break;
 		case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
-			reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE_NOTIFY);
 			break;
 		case LTTNG_HEALTH_APP_REG_DISPATCH:
-			reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
+			reply.ret_code = health_check_state(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH);
 			break;
 		case LTTNG_HEALTH_ALL:
 			reply.ret_code =
-				health_check_state(HEALTH_TYPE_APP_MANAGE) &&
-				health_check_state(HEALTH_TYPE_APP_REG) &&
-				health_check_state(HEALTH_TYPE_CMD) &&
-				health_check_state(HEALTH_TYPE_KERNEL) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_APP_REG) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_CMD) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_KERNEL) &&
 				check_consumer_health() &&
-				health_check_state(HEALTH_TYPE_HT_CLEANUP) &&
-				health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
-				health_check_state(HEALTH_TYPE_APP_REG_DISPATCH);
+				health_check_state(health_sessiond, HEALTH_TYPE_HT_CLEANUP) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_APP_MANAGE_NOTIFY) &&
+				health_check_state(health_sessiond, HEALTH_TYPE_APP_REG_DISPATCH);
 			break;
 		default:
 			reply.ret_code = LTTNG_ERR_UND;
@@ -3604,7 +3667,7 @@ static void *thread_manage_clients(void *data)
 
 	rcu_register_thread();
 
-	health_register(HEALTH_TYPE_CMD);
+	health_register(health_sessiond, HEALTH_TYPE_CMD);
 
 	if (testpoint(thread_manage_clients)) {
 		goto error_testpoint;
@@ -3828,7 +3891,7 @@ error_testpoint:
 		ERR("Health error occurred in %s", __func__);
 	}
 
-	health_unregister();
+	health_unregister(health_sessiond);
 
 	DBG("Client thread dying");
 
@@ -4659,7 +4722,11 @@ int main(int argc, char **argv)
 	 * Initialize the health check subsystem. This call should set the
 	 * appropriate time values.
 	 */
-	health_init();
+	health_sessiond = health_app_create(HEALTH_NUM_TYPE);
+	if (!health_sessiond) {
+		PERROR("health_app_create error");
+		goto exit_health_sessiond_cleanup;
+	}
 
 	/* Create thread to manage the client socket */
 	ret = pthread_create(&ht_cleanup_thread, NULL,
@@ -4714,7 +4781,7 @@ int main(int argc, char **argv)
 			ust_thread_manage_notify, (void *) NULL);
 	if (ret != 0) {
 		PERROR("pthread_create apps");
-		goto exit_apps;
+		goto exit_apps_notify;
 	}
 
 	/* Don't start this thread if kernel tracing is not requested nor root */
@@ -4735,12 +4802,20 @@ int main(int argc, char **argv)
 	}
 
 exit_kernel:
+	ret = pthread_join(apps_notify_thread, &status);
+	if (ret != 0) {
+		PERROR("pthread_join apps notify");
+		goto error;	/* join error, exit without cleanup */
+	}
+
+exit_apps_notify:
 	ret = pthread_join(apps_thread, &status);
 	if (ret != 0) {
-		PERROR("pthread_join");
+		PERROR("pthread_join apps");
 		goto error;	/* join error, exit without cleanup */
 	}
 
+
 exit_apps:
 	ret = pthread_join(reg_apps_thread, &status);
 	if (ret != 0) {
@@ -4794,6 +4869,8 @@ exit_health:
 		goto error;	/* join error, exit without cleanup */
 	}
 exit_ht_cleanup:
+	health_app_destroy(health_sessiond);
+exit_health_sessiond_cleanup:
 exit:
 	/*
 	 * cleanup() is called when no other thread is running.