Fix: sessiond fails to launch on --without-ust configuration
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
index 1d9ca91c5fc3b751d061fbcb3f543fcd52626f6b..37c0a9325960f84940ae268039b3c4e8bea212ce 100644 (file)
@@ -74,7 +74,7 @@
 #include "notification-thread.h"
 #include "notification-thread-commands.h"
 #include "rotation-thread.h"
-#include "syscall.h"
+#include "lttng-syscall.h"
 #include "agent.h"
 #include "ht-cleanup.h"
 #include "sessiond-config.h"
@@ -304,7 +304,7 @@ struct lttng_ht *agent_apps_ht_by_sock = NULL;
  * NR_LTTNG_SESSIOND_READY must match the number of calls to
  * sessiond_notify_ready().
  */
-#define NR_LTTNG_SESSIOND_READY                5
+#define NR_LTTNG_SESSIOND_READY                6
 int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY;
 
 int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
@@ -595,7 +595,8 @@ static void sessiond_cleanup(void)
                /* Cleanup ALL session */
                cds_list_for_each_entry_safe(sess, stmp,
                                &session_list_ptr->head, list) {
-                       cmd_destroy_session(sess, kernel_poll_pipe[1]);
+                       cmd_destroy_session(sess, kernel_poll_pipe[1],
+                                       notification_thread_handle);
                }
        }
 
@@ -630,21 +631,6 @@ static void sessiond_cleanup(void)
                free(load_info);
        }
 
-       /*
-        * Cleanup lock file by deleting it and finaly closing it which will
-        * release the file system lock.
-        */
-       if (lockfile_fd >= 0) {
-               ret = remove(config.lock_file_path.value);
-               if (ret < 0) {
-                       PERROR("remove lock file");
-               }
-               ret = close(lockfile_fd);
-               if (ret < 0) {
-                       PERROR("close lock file");
-               }
-       }
-
        /*
         * We do NOT rmdir rundir because there are other processes
         * using it, for instance lttng-relayd, which can start in
@@ -874,7 +860,7 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
                        cds_lfht_for_each_entry(ksess->consumer->socks->ht,
                                        &iter.iter, socket, node.node) {
                                pthread_mutex_lock(socket->lock);
-                               ret = kernel_consumer_send_channel_stream(socket,
+                               ret = kernel_consumer_send_channel_streams(socket,
                                                channel, ksess,
                                                session->output_traces ? 1 : 0);
                                pthread_mutex_unlock(socket->lock);
@@ -2964,8 +2950,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
        case LTTNG_REGISTER_TRIGGER:
        case LTTNG_UNREGISTER_TRIGGER:
        case LTTNG_ROTATE_SESSION:
-       case LTTNG_ROTATE_PENDING:
-       case LTTNG_ROTATE_GET_CURRENT_PATH:
+       case LTTNG_ROTATION_GET_INFO:
+       case LTTNG_SESSION_GET_CURRENT_OUTPUT:
+       case LTTNG_ROTATION_SET_SCHEDULE:
+       case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD:
+       case LTTNG_ROTATION_SCHEDULE_GET_SIZE:
                need_domain = 0;
                break;
        default:
@@ -3009,7 +2998,9 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
        case LTTNG_LIST_TRACKER_PIDS:
        case LTTNG_DATA_PENDING:
        case LTTNG_ROTATE_SESSION:
-       case LTTNG_ROTATE_PENDING:
+       case LTTNG_ROTATION_GET_INFO:
+       case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD:
+       case LTTNG_ROTATION_SCHEDULE_GET_SIZE:
                break;
        default:
                /* Setup lttng message with no payload */
@@ -3701,6 +3692,20 @@ error_add_context:
        }
        case LTTNG_START_TRACE:
        {
+               /*
+                * On the first start, if we have a kernel session and we have
+                * enabled time or size-based rotations, we have to make sure
+                * the kernel tracer supports it.
+                */
+               if (!cmd_ctx->session->has_been_started && \
+                               cmd_ctx->session->kernel_session && \
+                               (cmd_ctx->session->rotate_timer_period || \
+                                       cmd_ctx->session->rotate_size) && \
+                               !check_rotate_compatible()) {
+                       DBG("Kernel tracer version is not compatible with the rotation feature");
+                       ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
+                       goto error;
+               }
                ret = cmd_start_trace(cmd_ctx->session);
                break;
        }
@@ -3752,7 +3757,8 @@ error_add_context:
        }
        case LTTNG_DESTROY_SESSION:
        {
-               ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1]);
+               ret = cmd_destroy_session(cmd_ctx->session, kernel_poll_pipe[1],
+                               notification_thread_handle);
 
                /* Set session to NULL so we do not unlock it after free. */
                cmd_ctx->session = NULL;
@@ -4104,6 +4110,7 @@ error_add_context:
 
                DBG("Client rotate session \"%s\"", cmd_ctx->session->name);
 
+               memset(&rotate_return, 0, sizeof(rotate_return));
                if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
                        DBG("Kernel tracer version is not compatible with the rotation feature");
                        ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
@@ -4126,20 +4133,20 @@ error_add_context:
                ret = LTTNG_OK;
                break;
        }
-       case LTTNG_ROTATE_PENDING:
+       case LTTNG_ROTATION_GET_INFO:
        {
-               struct lttng_rotate_pending_return *pending_return = NULL;
+               struct lttng_rotation_get_info_return get_info_return;
 
-               ret = cmd_rotate_pending(cmd_ctx->session, &pending_return,
-                               cmd_ctx->lsm->u.rotate_pending.rotate_id);
+               memset(&get_info_return, 0, sizeof(get_info_return));
+               ret = cmd_rotate_get_info(cmd_ctx->session, &get_info_return,
+                               cmd_ctx->lsm->u.get_rotation_info.rotation_id);
                if (ret < 0) {
                        ret = -ret;
                        goto error;
                }
 
-               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, pending_return,
-                               sizeof(struct lttng_rotate_session_handle));
-               free(pending_return);
+               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &get_info_return,
+                               sizeof(get_info_return));
                if (ret < 0) {
                        ret = -ret;
                        goto error;
@@ -4148,19 +4155,82 @@ error_add_context:
                ret = LTTNG_OK;
                break;
        }
-       case LTTNG_ROTATE_GET_CURRENT_PATH:
+       case LTTNG_SESSION_GET_CURRENT_OUTPUT:
        {
-               struct lttng_rotate_get_current_path *get_return = NULL;
+               struct lttng_session_get_current_output_return output_return;
 
-               ret = cmd_rotate_get_current_path(cmd_ctx->session, &get_return);
+               memset(&output_return, 0, sizeof(output_return));
+               ret = cmd_session_get_current_output(cmd_ctx->session,
+                               &output_return);
                if (ret < 0) {
                        ret = -ret;
                        goto error;
                }
 
-               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, get_return,
-                               sizeof(struct lttng_rotate_get_current_path));
-               free(get_return);
+               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, &output_return,
+                               sizeof(output_return));
+               if (ret < 0) {
+                       ret = -ret;
+                       goto error;
+               }
+
+               ret = LTTNG_OK;
+               break;
+       }
+       case LTTNG_ROTATION_SET_SCHEDULE:
+       {
+               if (cmd_ctx->session->kernel_session && !check_rotate_compatible()) {
+                       DBG("Kernel tracer version does not support session rotations");
+                       ret = LTTNG_ERR_ROTATION_WRONG_VERSION;
+                       goto error;
+               }
+
+               ret = cmd_rotation_set_schedule(cmd_ctx->session,
+                               cmd_ctx->lsm->u.rotate_setup.timer_us,
+                               cmd_ctx->lsm->u.rotate_setup.size,
+                               notification_thread_handle);
+               if (ret != LTTNG_OK) {
+                       goto error;
+               }
+
+               break;
+       }
+       case LTTNG_ROTATION_SCHEDULE_GET_TIMER_PERIOD:
+       {
+               struct lttng_rotation_schedule_get_timer_period *get_timer;
+
+               get_timer = zmalloc(sizeof(struct lttng_rotation_schedule_get_timer_period));
+               if (!get_timer) {
+                       ret = ENOMEM;
+                       goto error;
+               }
+               get_timer->rotate_timer = cmd_ctx->session->rotate_timer_period;
+
+               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, get_timer,
+                               sizeof(struct lttng_rotation_schedule_get_timer_period));
+               free(get_timer);
+               if (ret < 0) {
+                       ret = -ret;
+                       goto error;
+               }
+
+               ret = LTTNG_OK;
+               break;
+       }
+       case LTTNG_ROTATION_SCHEDULE_GET_SIZE:
+       {
+               struct lttng_rotation_schedule_get_size *get_size;
+
+               get_size = zmalloc(sizeof(struct lttng_rotation_schedule_get_size));
+               if (!get_size) {
+                       ret = ENOMEM;
+                       goto error;
+               }
+               get_size->rotate_size = cmd_ctx->session->rotate_size;
+
+               ret = setup_lttng_msg_no_cmd_header(cmd_ctx, get_size,
+                               sizeof(struct lttng_rotation_schedule_get_size));
+               free(get_size);
                if (ret < 0) {
                        ret = -ret;
                        goto error;
@@ -4426,12 +4496,41 @@ static void *thread_manage_clients(void *data)
        }
 
        sessiond_notify_ready();
+
        ret = sem_post(&load_info->message_thread_ready);
        if (ret) {
                PERROR("sem_post message_thread_ready");
                goto error;
        }
 
+       /*
+        * Wait until all support threads are initialized before accepting
+        * commands.
+        */
+       while (uatomic_read(&lttng_sessiond_ready) != 0) {
+               fd_set read_fds;
+               struct timeval timeout;
+
+               FD_ZERO(&read_fds);
+               FD_SET(thread_quit_pipe[0], &read_fds);
+               memset(&timeout, 0, sizeof(timeout));
+               timeout.tv_usec = 1000;
+
+               /*
+                * If a support thread failed to launch, it may signal that
+                * we must exit and the sessiond would never be marked as
+                * "ready".
+                *
+                * The timeout is set to 1ms, which serves as a way to
+                * pace down this check.
+                */
+               ret = select(thread_quit_pipe[0] + 1, &read_fds, NULL, NULL,
+                               &timeout);
+               if (ret > 0 || (ret < 0 && errno != EINTR)) {
+                       goto exit;
+               }
+       }
+
        /* This testpoint is after we signal readiness to the parent. */
        if (testpoint(sessiond_thread_manage_clients)) {
                goto error;
@@ -4844,7 +4943,7 @@ static int set_option(int opt, const char *arg, const char *optname)
        } else if (string_match(optname, "no-kernel")) {
                config.no_kernel = true;
        } else if (string_match(optname, "quiet") || opt == 'q') {
-               lttng_opt_quiet = true;
+               config.quiet = true;
        } else if (string_match(optname, "verbose") || opt == 'v') {
                /* Verbose level can increase using multiple -v */
                if (arg) {
@@ -4963,8 +5062,8 @@ static int set_option(int opt, const char *arg, const char *optname)
                                ERR("Port overflow in --agent-tcp-port parameter: %s", arg);
                                return -1;
                        }
-                       config.agent_tcp_port = (uint32_t) v;
-                       DBG3("Agent TCP port set to non default: %u", config.agent_tcp_port);
+                       config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v;
+                       DBG3("Agent TCP port set to non default: %i", (int) v);
                }
        } else if (string_match(optname, "load") || opt == 'l') {
                if (!arg || *arg == '\0') {
@@ -5247,18 +5346,57 @@ end:
        return ret;
 }
 
+/*
+ * Create lockfile using the rundir and return its fd.
+ */
+static int create_lockfile(void)
+{
+        return utils_create_lock_file(config.lock_file_path.value);
+}
+
 /*
  * Check if the global socket is available, and if a daemon is answering at the
  * other side. If yes, error is returned.
+ *
+ * Also attempts to create and hold the lock file.
  */
 static int check_existing_daemon(void)
 {
+       int ret = 0;
+
        /* Is there anybody out there ? */
        if (lttng_session_daemon_alive()) {
-               return -EEXIST;
+               ret = -EEXIST;
+               goto end;
        }
 
-       return 0;
+       lockfile_fd = create_lockfile();
+       if (lockfile_fd < 0) {
+               ret = -EEXIST;
+               goto end;
+       }
+end:
+       return ret;
+}
+
+static void sessiond_cleanup_lock_file(void)
+{
+       int ret;
+
+       /*
+        * Cleanup lock file by deleting it and finaly closing it which will
+        * release the file system lock.
+        */
+       if (lockfile_fd >= 0) {
+               ret = remove(config.lock_file_path.value);
+               if (ret < 0) {
+                       PERROR("remove lock file");
+               }
+               ret = close(lockfile_fd);
+               if (ret < 0) {
+                       PERROR("close lock file");
+               }
+       }
 }
 
 /*
@@ -5517,23 +5655,6 @@ static int write_pidfile(void)
         return utils_create_pid_file(getpid(), config.pid_file_path.value);
 }
 
-/*
- * Create lockfile using the rundir and return its fd.
- */
-static int create_lockfile(void)
-{
-        return utils_create_lock_file(config.lock_file_path.value);
-}
-
-/*
- * Write agent TCP port using the rundir.
- */
-static int write_agent_port(void)
-{
-       return utils_create_pid_file(config.agent_tcp_port,
-                       config.agent_port_file_path.value);
-}
-
 static int set_clock_plugin_env(void)
 {
        int ret = 0;
@@ -5616,15 +5737,16 @@ int main(int argc, char **argv)
        struct lttng_pipe *ust32_channel_monitor_pipe = NULL,
                        *ust64_channel_monitor_pipe = NULL,
                        *kernel_channel_monitor_pipe = NULL;
-       bool notification_thread_running = false;
-       bool rotation_thread_running = false;
-       bool timer_thread_running = false;
+       bool notification_thread_launched = false;
+       bool rotation_thread_launched = false;
+       bool timer_thread_launched = false;
        struct lttng_pipe *ust32_channel_rotate_pipe = NULL,
                        *ust64_channel_rotate_pipe = NULL,
                        *kernel_channel_rotate_pipe = NULL;
        struct timer_thread_parameters timer_thread_ctx;
        /* Queue of rotation jobs populated by the sessiond-timer. */
        struct rotation_thread_timer_queue *rotation_timer_queue = NULL;
+       sem_t notification_thread_ready;
 
        init_kernel_workarounds();
 
@@ -5700,6 +5822,18 @@ int main(int argc, char **argv)
 
        sessiond_config_log(&config);
 
+       if (create_lttng_rundir()) {
+               retval = -1;
+               goto exit_options;
+       }
+
+       /* Abort launch if a session daemon is already running. */
+       if (check_existing_daemon()) {
+               ERR("A session daemon is already running.");
+               retval = -1;
+               goto exit_options;
+       }
+
        /* Daemonize */
        if (config.daemonize || config.background) {
                int i;
@@ -5714,9 +5848,12 @@ int main(int argc, char **argv)
                /*
                 * 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.
+                * descriptors than the standard ones and the lock file.
                 */
                for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
+                       if (i == lockfile_fd) {
+                               continue;
+                       }
                        (void) close(i);
                }
        }
@@ -5755,12 +5892,6 @@ int main(int argc, char **argv)
 
        /* Check if daemon is UID = 0 */
        is_root = !getuid();
-
-       if (create_lttng_rundir()) {
-               retval = -1;
-               goto exit_init_data;
-       }
-
        if (is_root) {
                /* Create global run dir with root access */
 
@@ -5792,12 +5923,6 @@ int main(int argc, char **argv)
                }
        }
 
-       lockfile_fd = create_lockfile();
-       if (lockfile_fd < 0) {
-               retval = -1;
-               goto exit_init_data;
-       }
-
        /* Set consumer initial state */
        kernel_consumerd_state = CONSUMER_STOPPED;
        ust_consumerd_state = CONSUMER_STOPPED;
@@ -5864,19 +5989,6 @@ int main(int argc, char **argv)
                goto exit_init_data;
        }
 
-       /*
-        * See if daemon already exist.
-        */
-       if (check_existing_daemon()) {
-               ERR("Already running daemon.\n");
-               /*
-                * We do not goto exit because we must not cleanup()
-                * because a daemon is already running.
-                */
-               retval = -1;
-               goto exit_init_data;
-       }
-
        /*
         * Init UST app hash table. Alloc hash table before this point since
         * cleanup() can get called after that point.
@@ -6004,12 +6116,6 @@ int main(int argc, char **argv)
                retval = -1;
                goto exit_init_data;
        }
-       ret = write_agent_port();
-       if (ret) {
-               ERR("Error in write_agent_port");
-               retval = -1;
-               goto exit_init_data;
-       }
 
        /* Initialize communication library */
        lttcomm_init();
@@ -6032,11 +6138,19 @@ int main(int argc, char **argv)
                goto exit_health;
        }
 
+       /*
+        * The rotation thread needs the notification thread to be ready before
+        * creating the rotate_notification_channel, so we use this semaphore as
+        * a rendez-vous point.
+        */
+       sem_init(&notification_thread_ready, 0, 0);
+
        /* notification_thread_data acquires the pipes' read side. */
        notification_thread_handle = notification_thread_handle_create(
                        ust32_channel_monitor_pipe,
                        ust64_channel_monitor_pipe,
-                       kernel_channel_monitor_pipe);
+                       kernel_channel_monitor_pipe,
+                       &notification_thread_ready);
        if (!notification_thread_handle) {
                retval = -1;
                ERR("Failed to create notification thread shared data");
@@ -6054,7 +6168,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_notification;
        }
-       notification_thread_running = true;
+       notification_thread_launched = true;
 
        /* Create timer thread. */
        ret = pthread_create(&timer_thread, default_pthread_attr(),
@@ -6066,7 +6180,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_notification;
        }
-       timer_thread_running = true;
+       timer_thread_launched = true;
 
        /* rotation_thread_data acquires the pipes' read side. */
        rotation_thread_handle = rotation_thread_handle_create(
@@ -6074,7 +6188,9 @@ int main(int argc, char **argv)
                        ust64_channel_rotate_pipe,
                        kernel_channel_rotate_pipe,
                        thread_quit_pipe[0],
-                       rotation_timer_queue);
+                       rotation_timer_queue,
+                       notification_thread_handle,
+                       &notification_thread_ready);
        if (!rotation_thread_handle) {
                retval = -1;
                ERR("Failed to create rotation thread shared data");
@@ -6092,7 +6208,7 @@ int main(int argc, char **argv)
                stop_threads();
                goto exit_rotation;
        }
-       rotation_thread_running = true;
+       rotation_thread_launched = true;
 
        /* Create thread to manage the client socket */
        ret = pthread_create(&client_thread, default_pthread_attr(),
@@ -6262,6 +6378,7 @@ exit_dispatch:
 exit_client:
 exit_rotation:
 exit_notification:
+       sem_destroy(&notification_thread_ready);
        ret = pthread_join(health_thread, &status);
        if (ret) {
                errno = ret;
@@ -6297,7 +6414,7 @@ exit_init_data:
         * of the active session and channels at the moment of the teardown.
         */
        if (notification_thread_handle) {
-               if (notification_thread_running) {
+               if (notification_thread_launched) {
                        notification_thread_command_quit(
                                        notification_thread_handle);
                        ret = pthread_join(notification_thread, &status);
@@ -6311,7 +6428,7 @@ exit_init_data:
        }
 
        if (rotation_thread_handle) {
-               if (rotation_thread_running) {
+               if (rotation_thread_launched) {
                        ret = pthread_join(rotation_thread, &status);
                        if (ret) {
                                errno = ret;
@@ -6322,7 +6439,7 @@ exit_init_data:
                rotation_thread_handle_destroy(rotation_thread_handle);
        }
 
-       if (timer_thread_running) {
+       if (timer_thread_launched) {
                kill(getpid(), LTTNG_SESSIOND_SIG_EXIT);
                ret = pthread_join(timer_thread, &status);
                if (ret) {
@@ -6358,6 +6475,7 @@ exit_health_sessiond_cleanup:
 exit_create_run_as_worker_cleanup:
 
 exit_options:
+       sessiond_cleanup_lock_file();
        sessiond_cleanup_options();
 
 exit_set_signal_handler:
This page took 0.030979 seconds and 4 git commands to generate.