X-Git-Url: https://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fclient.c;h=4565a18c31efe55f70e94293cd03d6962d574af4;hb=d295668767ac8234e83984e1812d342d03293d88;hp=119062c15c648f2dfcea8e395751e69d396080e3;hpb=917a718d4ec336ca98820f3cf56a2db57fc9b1dd;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/client.c b/src/bin/lttng-sessiond/client.c index 119062c15..4565a18c3 100644 --- a/src/bin/lttng-sessiond/client.c +++ b/src/bin/lttng-sessiond/client.c @@ -26,6 +26,8 @@ #include #include #include +#include +#include #include "client.h" #include "lttng-sessiond.h" @@ -35,35 +37,33 @@ #include "health-sessiond.h" #include "testpoint.h" #include "utils.h" +#include "manage-consumer.h" static bool is_root; static struct thread_state { - pthread_cond_t cond; - pthread_mutex_t lock; - bool is_running; -} thread_state = { - .cond = PTHREAD_COND_INITIALIZER, - .lock = PTHREAD_MUTEX_INITIALIZER, - .is_running = false -}; - -void set_thread_state_running(void) + sem_t ready; + bool running; +} thread_state; + +static void set_thread_status(bool running) { - pthread_mutex_lock(&thread_state.lock); - thread_state.is_running = true; - pthread_cond_broadcast(&thread_state.cond); - pthread_mutex_unlock(&thread_state.lock); + DBG("Marking client thread's state as %s", running ? "running" : "error"); + thread_state.running = running; + sem_post(&thread_state.ready); } -static void wait_thread_state_running(void) +static bool wait_thread_status(void) { - pthread_mutex_lock(&thread_state.lock); - while (!thread_state.is_running) { - pthread_cond_wait(&thread_state.cond, - &thread_state.lock); + DBG("Waiting for client thread to be ready"); + sem_wait(&thread_state.ready); + if (thread_state.running) { + DBG("Client thread is ready"); + } else { + ERR("Initialization of client thread failed"); } - pthread_mutex_unlock(&thread_state.lock); + + return thread_state.running; } /* @@ -83,6 +83,7 @@ static int setup_lttng_msg(struct command_ctx *cmd_ctx, const size_t payload_offset = cmd_header_offset + cmd_header_len; const size_t total_msg_size = header_len + cmd_header_len + payload_len; + free(cmd_ctx->llm); cmd_ctx->llm = zmalloc(total_msg_size); if (cmd_ctx->llm == NULL) { @@ -116,133 +117,11 @@ end: /* * Start the thread_manage_consumer. This must be done after a lttng-consumerd - * exec or it will fails. + * exec or it will fail. */ static int spawn_consumer_thread(struct consumer_data *consumer_data) { - int ret, clock_ret; - struct timespec timeout; - - /* - * Make sure we set the readiness flag to 0 because we are NOT ready. - * This access to consumer_thread_is_ready does not need to be - * protected by consumer_data.cond_mutex (yet) since the consumer - * management thread has not been started at this point. - */ - consumer_data->consumer_thread_is_ready = 0; - - /* Setup pthread condition */ - ret = pthread_condattr_init(&consumer_data->condattr); - if (ret) { - errno = ret; - PERROR("pthread_condattr_init consumer data"); - goto error; - } - - /* - * Set the monotonic clock in order to make sure we DO NOT jump in time - * between the clock_gettime() call and the timedwait call. See bug #324 - * for a more details and how we noticed it. - */ - ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC); - if (ret) { - errno = ret; - PERROR("pthread_condattr_setclock consumer data"); - goto error; - } - - ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr); - if (ret) { - errno = ret; - PERROR("pthread_cond_init consumer data"); - goto error; - } - - ret = pthread_create(&consumer_data->thread, default_pthread_attr(), - thread_manage_consumer, consumer_data); - if (ret) { - errno = ret; - PERROR("pthread_create consumer"); - ret = -1; - goto error; - } - - /* We are about to wait on a pthread condition */ - pthread_mutex_lock(&consumer_data->cond_mutex); - - /* Get time for sem_timedwait absolute timeout */ - clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout); - /* - * Set the timeout for the condition timed wait even if the clock gettime - * call fails since we might loop on that call and we want to avoid to - * increment the timeout too many times. - */ - timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT; - - /* - * The following loop COULD be skipped in some conditions so this is why we - * set ret to 0 in order to make sure at least one round of the loop is - * done. - */ - ret = 0; - - /* - * Loop until the condition is reached or when a timeout is reached. Note - * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT - * be returned but the pthread_cond(3), from the glibc-doc, says that it is - * possible. This loop does not take any chances and works with both of - * them. - */ - while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) { - if (clock_ret < 0) { - PERROR("clock_gettime spawn consumer"); - /* Infinite wait for the consumerd thread to be ready */ - ret = pthread_cond_wait(&consumer_data->cond, - &consumer_data->cond_mutex); - } else { - ret = pthread_cond_timedwait(&consumer_data->cond, - &consumer_data->cond_mutex, &timeout); - } - } - - /* Release the pthread condition */ - pthread_mutex_unlock(&consumer_data->cond_mutex); - - if (ret != 0) { - errno = ret; - if (ret == ETIMEDOUT) { - int pth_ret; - - /* - * Call has timed out so we kill the kconsumerd_thread and return - * an error. - */ - ERR("Condition timed out. The consumer thread was never ready." - " Killing it"); - pth_ret = pthread_cancel(consumer_data->thread); - if (pth_ret < 0) { - PERROR("pthread_cancel consumer thread"); - } - } else { - PERROR("pthread_cond_wait failed consumer thread"); - } - /* Caller is expecting a negative value on failure. */ - ret = -1; - goto error; - } - - pthread_mutex_lock(&consumer_data->pid_mutex); - if (consumer_data->pid == 0) { - ERR("Consumerd did not start"); - pthread_mutex_unlock(&consumer_data->pid_mutex); - goto error; - } - pthread_mutex_unlock(&consumer_data->pid_mutex); - - return 0; - -error: - return ret; + return launch_consumer_management_thread(consumer_data) ? 0 : -1; } /* @@ -512,10 +391,13 @@ static int copy_session_consumer(int domain, struct ltt_session *session) } /* Append correct directory to subdir */ - strncat(consumer->subdir, dir_name, - sizeof(consumer->subdir) - strlen(consumer->subdir) - 1); - DBG3("Copy session consumer subdir %s", consumer->subdir); - + ret = lttng_strncpy(consumer->domain_subdir, dir_name, + sizeof(consumer->domain_subdir)); + if (ret) { + ret = LTTNG_ERR_UNK; + goto error; + } + DBG3("Copy session consumer subdir %s", consumer->domain_subdir); ret = LTTNG_OK; error: @@ -599,7 +481,7 @@ static int create_kernel_session(struct ltt_session *session) ret = kernel_create_session(session, kernel_tracer_fd); if (ret < 0) { ret = LTTNG_ERR_KERN_SESS_FAIL; - goto error; + goto error_create; } /* Code flow safety */ @@ -621,6 +503,7 @@ static int create_kernel_session(struct ltt_session *session) error: trace_kernel_destroy_session(session->kernel_session); session->kernel_session = NULL; +error_create: return ret; } @@ -757,27 +640,6 @@ error: return ret; } -/* - * Join consumer thread - */ -static int join_consumer_thread(struct consumer_data *consumer_data) -{ - void *status; - - /* Consumer pid must be a real one. */ - if (consumer_data->pid > 0) { - int ret; - ret = kill(consumer_data->pid, SIGTERM); - if (ret) { - PERROR("Error killing consumer daemon"); - return ret; - } - return pthread_join(consumer_data->thread, &status); - } else { - return 0; - } -} - /* * Version of setup_lttng_msg() without command header. */ @@ -861,9 +723,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, *sock_error = 0; switch (cmd_ctx->lsm->cmd_type) { - case LTTNG_CREATE_SESSION: - case LTTNG_CREATE_SESSION_SNAPSHOT: - case LTTNG_CREATE_SESSION_LIVE: + case LTTNG_CREATE_SESSION_EXT: case LTTNG_DESTROY_SESSION: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_DOMAINS: @@ -941,9 +801,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Commands that DO NOT need a session. */ switch (cmd_ctx->lsm->cmd_type) { - case LTTNG_CREATE_SESSION: - case LTTNG_CREATE_SESSION_SNAPSHOT: - case LTTNG_CREATE_SESSION_LIVE: + case LTTNG_CREATE_SESSION_EXT: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: case LTTNG_LIST_SYSCALLS: @@ -1029,7 +887,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, if (need_tracing_session) { if (cmd_ctx->session->kernel_session == NULL) { ret = create_kernel_session(cmd_ctx->session); - if (ret < 0) { + if (ret != LTTNG_OK) { ret = LTTNG_ERR_KERN_SESS_FAIL; goto error; } @@ -1660,47 +1518,6 @@ error_add_context: ret = cmd_stop_trace(cmd_ctx->session); break; } - case LTTNG_CREATE_SESSION: - { - 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, 0); - - free(uris); - - break; - } case LTTNG_DESTROY_SESSION: { ret = cmd_destroy_session(cmd_ctx->session, @@ -1795,7 +1612,9 @@ error_add_context: nr_sessions = lttng_sessions_count( LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); - payload_len = sizeof(struct lttng_session) * nr_sessions; + + payload_len = (sizeof(struct lttng_session) * nr_sessions) + + (sizeof(struct lttng_session_extended) * nr_sessions); sessions_payload = zmalloc(payload_len); if (!sessions_payload) { @@ -1804,7 +1623,7 @@ error_add_context: goto setup_error; } - cmd_list_lttng_sessions(sessions_payload, + cmd_list_lttng_sessions(sessions_payload, nr_sessions, LTTNG_SOCK_GET_UID_CRED(&cmd_ctx->creds), LTTNG_SOCK_GET_GID_CRED(&cmd_ctx->creds)); session_unlock_list(); @@ -1935,82 +1754,35 @@ error_add_context: cmd_ctx->lsm->u.snapshot_record.wait); break; } - case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_EXT: { - 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; - } + struct lttng_dynamic_buffer payload; + struct lttng_session_descriptor *return_descriptor = NULL; - /* 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; - } + lttng_dynamic_buffer_init(&payload); + ret = cmd_create_session(cmd_ctx, sock, &return_descriptor); + if (ret != LTTNG_OK) { + goto error; } - ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris, - nb_uri, &cmd_ctx->creds); - 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 = lttng_session_descriptor_serialize(return_descriptor, + &payload); + if (ret) { + ERR("Failed to serialize session descriptor in reply to \"create session\" command"); + lttng_session_descriptor_destroy(return_descriptor); + ret = LTTNG_ERR_NOMEM; + 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); + ret = setup_lttng_msg_no_cmd_header(cmd_ctx, payload.data, + payload.size); + if (ret) { + lttng_session_descriptor_destroy(return_descriptor); + ret = LTTNG_ERR_NOMEM; + goto error; + } + lttng_dynamic_buffer_reset(&payload); + lttng_session_descriptor_destroy(return_descriptor); + ret = LTTNG_OK; break; } case LTTNG_SAVE_SESSION: @@ -2213,6 +1985,11 @@ static void cleanup_client_thread(void *data) lttng_pipe_destroy(quit_pipe); } +static void thread_init_cleanup(void *data) +{ + set_thread_status(false); +} + /* * This thread manage all clients request using the unix client socket for * communication. @@ -2232,6 +2009,7 @@ static void *thread_manage_clients(void *data) is_root = (getuid() == 0); + pthread_cleanup_push(thread_init_cleanup, NULL); client_sock = create_client_sock(); if (client_sock < 0) { goto error_listen; @@ -2269,6 +2047,10 @@ static void *thread_manage_clients(void *data) goto error; } + /* Set state as running. */ + set_thread_status(true); + pthread_cleanup_pop(0); + /* This testpoint is after we signal readiness to the parent. */ if (testpoint(sessiond_thread_manage_clients)) { goto error; @@ -2280,9 +2062,6 @@ static void *thread_manage_clients(void *data) health_code_update(); - /* Set state as running. */ - set_thread_state_running(); - while (1) { const struct cmd_completion_handler *cmd_completion_handler; @@ -2311,11 +2090,6 @@ static void *thread_manage_clients(void *data) health_code_update(); - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - if (pollfd == thread_quit_pipe_fd) { err = 0; goto exit; @@ -2490,28 +2264,6 @@ error_create_poll: DBG("Client thread dying"); rcu_unregister_thread(); - - /* - * Since we are creating the consumer threads, we own them, so we need - * to join them before our thread exits. - */ - ret = join_consumer_thread(&kconsumer_data); - if (ret) { - errno = ret; - PERROR("join_consumer"); - } - - ret = join_consumer_thread(&ustconsumer32_data); - if (ret) { - errno = ret; - PERROR("join_consumer ust32"); - } - - ret = join_consumer_thread(&ustconsumer64_data); - if (ret) { - errno = ret; - PERROR("join_consumer ust64"); - } return NULL; } @@ -2526,9 +2278,11 @@ bool shutdown_client_thread(void *thread_data) struct lttng_thread *launch_client_thread(void) { + bool thread_running; struct lttng_pipe *client_quit_pipe; struct lttng_thread *thread; + sem_init(&thread_state.ready, 0, 0); client_quit_pipe = lttng_pipe_open(FD_CLOEXEC); if (!client_quit_pipe) { goto error; @@ -2547,8 +2301,11 @@ struct lttng_thread *launch_client_thread(void) * This thread is part of the threads that need to be fully * initialized before the session daemon is marked as "ready". */ - wait_thread_state_running(); - + thread_running = wait_thread_status(); + if (!thread_running) { + lttng_thread_put(thread); + thread = NULL; + } return thread; error: cleanup_client_thread(client_quit_pipe);