Fix: set globally visible flag to kernel stream
[lttng-tools.git] / src / bin / lttng-sessiond / main.c
index efaaf4ffa99f40c5cc0dac877e796a7eda86e5ea..c02f6d91e2ba904819c58f3058d83efef218c6fd 100644 (file)
@@ -161,6 +161,7 @@ static pthread_t client_thread;
 static pthread_t kernel_thread;
 static pthread_t dispatch_thread;
 static pthread_t health_thread;
+static pthread_t ht_cleanup_thread;
 
 /*
  * UST registration command queue. This queue is tied with a futex and uses a N
@@ -645,7 +646,8 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd)
 
                                                pthread_mutex_lock(socket->lock);
                                                ret = kernel_consumer_send_channel_stream(socket,
-                                                               channel, ksess);
+                                                               channel, ksess,
+                                                               session->output_traces ? 1 : 0);
                                                pthread_mutex_unlock(socket->lock);
                                                if (ret < 0) {
                                                        rcu_read_unlock();
@@ -705,9 +707,9 @@ static void *thread_manage_kernel(void *data)
 
        /*
         * This first step of the while is to clean this structure which could free
-        * non NULL pointers so zero it before the loop.
+        * non NULL pointers so initialize it before the loop.
         */
-       memset(&events, 0, sizeof(events));
+       lttng_poll_init(&events);
 
        if (testpoint(thread_manage_kernel)) {
                goto error_testpoint;
@@ -1333,25 +1335,116 @@ error:
        return ret;
 }
 
+/*
+ * Sanitize the wait queue of the dispatch registration thread meaning removing
+ * invalid nodes from it. This is to avoid memory leaks for the case the UST
+ * notify socket is never received.
+ */
+static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
+{
+       int ret, nb_fd = 0, i;
+       unsigned int fd_added = 0;
+       struct lttng_poll_event events;
+       struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
+
+       assert(wait_queue);
+
+       lttng_poll_init(&events);
+
+       /* Just skip everything for an empty queue. */
+       if (!wait_queue->count) {
+               goto end;
+       }
+
+       ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
+       if (ret < 0) {
+               goto error_create;
+       }
+
+       cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+                       &wait_queue->head, head) {
+               assert(wait_node->app);
+               ret = lttng_poll_add(&events, wait_node->app->sock,
+                               LPOLLHUP | LPOLLERR);
+               if (ret < 0) {
+                       goto error;
+               }
+
+               fd_added = 1;
+       }
+
+       if (!fd_added) {
+               goto end;
+       }
+
+       /*
+        * Poll but don't block so we can quickly identify the faulty events and
+        * clean them afterwards from the wait queue.
+        */
+       ret = lttng_poll_wait(&events, 0);
+       if (ret < 0) {
+               goto error;
+       }
+       nb_fd = ret;
+
+       for (i = 0; i < nb_fd; i++) {
+               /* Get faulty FD. */
+               uint32_t revents = LTTNG_POLL_GETEV(&events, i);
+               int pollfd = LTTNG_POLL_GETFD(&events, i);
+
+               cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
+                               &wait_queue->head, head) {
+                       if (pollfd == wait_node->app->sock &&
+                                       (revents & (LPOLLHUP | LPOLLERR))) {
+                               cds_list_del(&wait_node->head);
+                               wait_queue->count--;
+                               ust_app_destroy(wait_node->app);
+                               free(wait_node);
+                               break;
+                       }
+               }
+       }
+
+       if (nb_fd > 0) {
+               DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
+       }
+
+end:
+       lttng_poll_clean(&events);
+       return;
+
+error:
+       lttng_poll_clean(&events);
+error_create:
+       ERR("Unable to sanitize wait queue");
+       return;
+}
+
 /*
  * Dispatch request from the registration threads to the application
  * communication thread.
  */
 static void *thread_dispatch_ust_registration(void *data)
 {
-       int ret;
+       int ret, err = -1;
        struct cds_wfq_node *node;
        struct ust_command *ust_cmd = NULL;
-       struct {
-               struct ust_app *app;
-               struct cds_list_head head;
-       } *wait_node = NULL, *tmp_wait_node;
+       struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
+       struct ust_reg_wait_queue wait_queue = {
+               .count = 0,
+       };
+
+       health_register(HEALTH_TYPE_APP_REG_DISPATCH);
+
+       health_code_update();
 
-       CDS_LIST_HEAD(wait_queue);
+       CDS_INIT_LIST_HEAD(&wait_queue.head);
 
        DBG("[thread] Dispatch UST command started");
 
        while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
+               health_code_update();
+
                /* Atomically prepare the queue futex */
                futex_nto1_prepare(&ust_cmd_queue.futex);
 
@@ -1359,6 +1452,14 @@ static void *thread_dispatch_ust_registration(void *data)
                        struct ust_app *app = NULL;
                        ust_cmd = NULL;
 
+                       /*
+                        * Make sure we don't have node(s) that have hung up before receiving
+                        * the notify socket. This is to clean the list in order to avoid
+                        * memory leaks from notify socket that are never seen.
+                        */
+                       sanitize_wait_queue(&wait_queue);
+
+                       health_code_update();
                        /* Dequeue command for registration */
                        node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue);
                        if (node == NULL) {
@@ -1380,6 +1481,11 @@ static void *thread_dispatch_ust_registration(void *data)
                                wait_node = zmalloc(sizeof(*wait_node));
                                if (!wait_node) {
                                        PERROR("zmalloc wait_node dispatch");
+                                       ret = close(ust_cmd->sock);
+                                       if (ret < 0) {
+                                               PERROR("close ust sock dispatch %d", ust_cmd->sock);
+                                       }
+                                       lttng_fd_put(1, LTTNG_FD_APPS);
                                        free(ust_cmd);
                                        goto error;
                                }
@@ -1402,7 +1508,8 @@ static void *thread_dispatch_ust_registration(void *data)
                                 * Add application to the wait queue so we can set the notify
                                 * socket before putting this object in the global ht.
                                 */
-                               cds_list_add(&wait_node->head, &wait_queue);
+                               cds_list_add(&wait_node->head, &wait_queue.head);
+                               wait_queue.count++;
 
                                free(ust_cmd);
                                /*
@@ -1417,16 +1524,31 @@ static void *thread_dispatch_ust_registration(void *data)
                                 * notify socket if found.
                                 */
                                cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
-                                               &wait_queue, head) {
+                                               &wait_queue.head, head) {
+                                       health_code_update();
                                        if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
                                                wait_node->app->notify_sock = ust_cmd->sock;
                                                cds_list_del(&wait_node->head);
+                                               wait_queue.count--;
                                                app = wait_node->app;
                                                free(wait_node);
                                                DBG3("UST app notify socket %d is set", ust_cmd->sock);
                                                break;
                                        }
                                }
+
+                               /*
+                                * With no application at this stage the received socket is
+                                * basically useless so close it before we free the cmd data
+                                * structure for good.
+                                */
+                               if (!app) {
+                                       ret = close(ust_cmd->sock);
+                                       if (ret < 0) {
+                                               PERROR("close ust sock dispatch %d", ust_cmd->sock);
+                                       }
+                                       lttng_fd_put(1, LTTNG_FD_APPS);
+                               }
                                free(ust_cmd);
                        }
 
@@ -1488,29 +1610,32 @@ static void *thread_dispatch_ust_registration(void *data)
 
                                rcu_read_unlock();
                                session_unlock_list();
-                       } else {
-                               /* Application manager threads are not available. */
-                               ret = close(ust_cmd->sock);
-                               if (ret < 0) {
-                                       PERROR("close ust_cmd sock");
-                               }
-                               lttng_fd_put(1, LTTNG_FD_APPS);
                        }
                } while (node != NULL);
 
+               health_poll_entry();
                /* Futex wait on queue. Blocking call on futex() */
                futex_nto1_wait(&ust_cmd_queue.futex);
+               health_poll_exit();
        }
+       /* Normal exit, no error */
+       err = 0;
 
 error:
        /* Clean up wait queue. */
        cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
-                       &wait_queue, head) {
+                       &wait_queue.head, head) {
                cds_list_del(&wait_node->head);
+               wait_queue.count--;
                free(wait_node);
        }
 
        DBG("Dispatch thread dying");
+       if (err) {
+               health_error();
+               ERR("Health error occurred in %s", __func__);
+       }
+       health_unregister();
        return NULL;
 }
 
@@ -2253,6 +2378,7 @@ static int create_ust_session(struct ltt_session *session,
 
        lus->uid = session->uid;
        lus->gid = session->gid;
+       lus->output_traces = session->output_traces;
        session->ust_session = lus;
 
        /* Copy session output to the newly created UST session */
@@ -2309,6 +2435,7 @@ static int create_kernel_session(struct ltt_session *session)
 
        session->kernel_session->uid = session->uid;
        session->kernel_session->gid = session->gid;
+       session->kernel_session->output_traces = session->output_traces;
 
        return LTTNG_OK;
 
@@ -2370,6 +2497,10 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
        case LTTNG_START_TRACE:
        case LTTNG_STOP_TRACE:
        case LTTNG_DATA_PENDING:
+       case LTTNG_SNAPSHOT_ADD_OUTPUT:
+       case LTTNG_SNAPSHOT_DEL_OUTPUT:
+       case LTTNG_SNAPSHOT_LIST_OUTPUT:
+       case LTTNG_SNAPSHOT_RECORD:
                need_domain = 0;
                break;
        default:
@@ -2438,12 +2569,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
                session_lock_list();
                cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
                if (cmd_ctx->session == NULL) {
-                       if (cmd_ctx->lsm->session.name != NULL) {
-                               ret = LTTNG_ERR_SESS_NOT_FOUND;
-                       } else {
-                               /* If no session name specified */
-                               ret = LTTNG_ERR_SELECT_SESS;
-                       }
+                       ret = LTTNG_ERR_SESS_NOT_FOUND;
                        goto error;
                } else {
                        /* Acquire lock for the session */
@@ -3057,6 +3183,67 @@ skip_domain:
                ret = cmd_data_pending(cmd_ctx->session);
                break;
        }
+       case LTTNG_SNAPSHOT_ADD_OUTPUT:
+       {
+               struct lttcomm_lttng_output_id reply;
+
+               ret = cmd_snapshot_add_output(cmd_ctx->session,
+                               &cmd_ctx->lsm->u.snapshot_output.output, &reply.id);
+               if (ret != LTTNG_OK) {
+                       goto error;
+               }
+
+               ret = setup_lttng_msg(cmd_ctx, sizeof(reply));
+               if (ret < 0) {
+                       goto setup_error;
+               }
+
+               /* Copy output list into message payload */
+               memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply));
+               ret = LTTNG_OK;
+               break;
+       }
+       case LTTNG_SNAPSHOT_DEL_OUTPUT:
+       {
+               ret = cmd_snapshot_del_output(cmd_ctx->session,
+                               &cmd_ctx->lsm->u.snapshot_output.output);
+               break;
+       }
+       case LTTNG_SNAPSHOT_LIST_OUTPUT:
+       {
+               ssize_t nb_output;
+               struct lttng_snapshot_output *outputs = NULL;
+
+               nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs);
+               if (nb_output < 0) {
+                       ret = -nb_output;
+                       goto error;
+               }
+
+               ret = setup_lttng_msg(cmd_ctx,
+                               nb_output * sizeof(struct lttng_snapshot_output));
+               if (ret < 0) {
+                       free(outputs);
+                       goto setup_error;
+               }
+
+               if (outputs) {
+                       /* Copy output list into message payload */
+                       memcpy(cmd_ctx->llm->payload, outputs,
+                                       nb_output * sizeof(struct lttng_snapshot_output));
+                       free(outputs);
+               }
+
+               ret = LTTNG_OK;
+               break;
+       }
+       case LTTNG_SNAPSHOT_RECORD:
+       {
+               ret = cmd_snapshot_record(cmd_ctx->session,
+                               &cmd_ctx->lsm->u.snapshot_record.output,
+                               cmd_ctx->lsm->u.snapshot_record.wait);
+               break;
+       }
        default:
                ret = LTTNG_ERR_UND;
                break;
@@ -3097,9 +3284,8 @@ static void *thread_manage_health(void *data)
 
        rcu_register_thread();
 
-       /* We might hit an error path before this is set once. */
-       memset(&events, 0, sizeof(events));
-       events.epfd = -1;
+       /* We might hit an error path before this is created. */
+       lttng_poll_init(&events);
 
        /* Create unix socket */
        sock = lttcomm_create_unix_sock(health_unix_sock_path);
@@ -3215,13 +3401,25 @@ restart:
                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);
+                       break;
+               case LTTNG_HEALTH_APP_MANAGE_NOTIFY:
+                       reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY);
+                       break;
+               case LTTNG_HEALTH_APP_REG_DISPATCH:
+                       reply.ret_code = health_check_state(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) &&
-                               check_consumer_health();
+                               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);
                        break;
                default:
                        reply.ret_code = LTTNG_ERR_UND;
@@ -3266,12 +3464,6 @@ error:
                        PERROR("close");
                }
        }
-       if (new_sock >= 0) {
-               ret = close(new_sock);
-               if (ret) {
-                       PERROR("close");
-               }
-       }
 
        lttng_poll_clean(&events);
 
@@ -4132,7 +4324,7 @@ int main(int argc, char **argv)
                DBG2("Kernel consumer cmd path: %s",
                                kconsumer_data.cmd_unix_sock_path);
        } else {
-               home_path = get_home_dir();
+               home_path = utils_get_home_dir();
                if (home_path == NULL) {
                        /* TODO: Add --socket PATH option */
                        ERR("Can't get HOME directory for sockets creation.");
@@ -4289,6 +4481,11 @@ int main(int argc, char **argv)
                }
        }
 
+       /* Setup the thread ht_cleanup communication pipe. */
+       if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) {
+               goto exit;
+       }
+
        /* Setup the thread apps communication pipe. */
        if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) {
                goto exit;
@@ -4327,6 +4524,14 @@ int main(int argc, char **argv)
 
        write_pidfile();
 
+       /* Create thread to manage the client socket */
+       ret = pthread_create(&ht_cleanup_thread, NULL,
+                       thread_ht_cleanup, (void *) NULL);
+       if (ret != 0) {
+               PERROR("pthread_create ht_cleanup");
+               goto exit_ht_cleanup;
+       }
+
        /* Create thread to manage the client socket */
        ret = pthread_create(&health_thread, NULL,
                        thread_manage_health, (void *) NULL);
@@ -4446,6 +4651,12 @@ exit_client:
        }
 
 exit_health:
+       ret = pthread_join(ht_cleanup_thread, &status);
+       if (ret != 0) {
+               PERROR("pthread_join ht cleanup thread");
+               goto error;     /* join error, exit without cleanup */
+       }
+exit_ht_cleanup:
 exit:
        /*
         * cleanup() is called when no other thread is running.
This page took 0.027753 seconds and 4 git commands to generate.