X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=034621e9da466ca1b4c49d0a7e4875f515e7fdaf;hb=227e824a28deb5b5d31955908827426a03f97802;hp=6e548f3c10bf8185c13ac8156f1b5c54fe0ff4ce;hpb=d5d63bf12b65741e64735103b0a36a0ebb52b09b;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 6e548f3c1..034621e9d 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -221,12 +221,6 @@ enum consumerd_state { static enum consumerd_state ust_consumerd_state; static enum consumerd_state kernel_consumerd_state; -/* Used for the health monitoring of the session daemon. See health.h */ -struct health_state health_thread_cmd; -struct health_state health_thread_app_manage; -struct health_state health_thread_app_reg; -struct health_state health_thread_kernel; - /* * Socket timeout for receiving and sending in seconds. */ @@ -626,7 +620,7 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) struct lttng_ht_iter iter; struct consumer_socket *socket; - + rcu_read_lock(); cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { /* Code flow error */ @@ -637,9 +631,11 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) channel, ksess); pthread_mutex_unlock(socket->lock); if (ret < 0) { + rcu_read_unlock(); goto error; } } + rcu_read_unlock(); } goto error; } @@ -691,6 +687,8 @@ static void *thread_manage_kernel(void *data) DBG("[thread] Thread manage kernel started"); + health_register(HEALTH_TYPE_KERNEL); + /* * This first step of the while is to clean this structure which could free * non NULL pointers so zero it before the loop. @@ -701,14 +699,14 @@ static void *thread_manage_kernel(void *data) goto error_testpoint; } - health_code_update(&health_thread_kernel); + health_code_update(); if (testpoint(thread_manage_kernel_before_loop)) { goto error_testpoint; } while (1) { - health_code_update(&health_thread_kernel); + health_code_update(); if (update_poll_flag == 1) { /* Clean events object. We are about to populate it again. */ @@ -736,9 +734,9 @@ static void *thread_manage_kernel(void *data) /* Poll infinite value of time */ restart: - health_poll_update(&health_thread_kernel); + health_poll_update(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_kernel); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -761,7 +759,7 @@ static void *thread_manage_kernel(void *data) revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_kernel); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -772,7 +770,13 @@ static void *thread_manage_kernel(void *data) /* Check for data on kernel pipe */ if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) { - ret = read(kernel_poll_pipe[0], &tmp, 1); + do { + ret = read(kernel_poll_pipe[0], &tmp, 1); + } while (ret < 0 && errno == EINTR); + /* + * Ret value is useless here, if this pipe gets any actions an + * update is required anyway. + */ update_poll_flag = 1; continue; } else { @@ -803,12 +807,12 @@ error_testpoint: utils_close_pipe(kernel_poll_pipe); kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1; if (err) { - health_error(&health_thread_kernel); + health_error(); ERR("Health error occurred in %s", __func__); WARN("Kernel thread died unexpectedly. " "Kernel tracing can continue but CPU hotplug is disabled."); } - health_exit(&health_thread_kernel); + health_unregister(); DBG("Kernel thread dying"); return NULL; } @@ -849,6 +853,8 @@ static void *thread_manage_consumer(void *data) DBG("[thread] Manage consumer started"); + health_register(HEALTH_TYPE_CONSUMER); + /* * Since the consumer thread can be spawned at any moment in time, we init * the health to a poll status (1, which is a valid health over time). @@ -866,7 +872,7 @@ static void *thread_manage_consumer(void *data) * In a nutshell, the following poll update to the health state brings back * the state to an even value meaning a code path. */ - health_poll_update(&consumer_data->health); + health_poll_update(); /* * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock. @@ -887,18 +893,18 @@ static void *thread_manage_consumer(void *data) goto error; } - health_code_update(&consumer_data->health); + health_code_update(); /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&consumer_data->health); + health_poll_update(); if (testpoint(thread_manage_consumer)) { goto error; } ret = lttng_poll_wait(&events, -1); - health_poll_update(&consumer_data->health); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -916,7 +922,7 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&consumer_data->health); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -945,7 +951,7 @@ restart: */ (void) utils_set_fd_cloexec(sock); - health_code_update(&consumer_data->health); + health_code_update(); DBG2("Receiving code from consumer err_sock"); @@ -956,7 +962,7 @@ restart: goto error; } - health_code_update(&consumer_data->health); + health_code_update(); if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { consumer_data->cmd_sock = @@ -986,13 +992,13 @@ restart: goto error; } - health_code_update(&consumer_data->health); + health_code_update(); /* Inifinite blocking call, waiting for transmission */ restart_poll: - health_poll_update(&consumer_data->health); + health_poll_update(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&consumer_data->health); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -1010,7 +1016,7 @@ restart_poll: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&consumer_data->health); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -1028,7 +1034,7 @@ restart_poll: } } - health_code_update(&consumer_data->health); + health_code_update(); /* Wait for any kconsumerd error */ ret = lttcomm_recv_unix_sock(sock, &code, @@ -1079,10 +1085,10 @@ error: lttng_poll_clean(&events); error_poll: if (err) { - health_error(&consumer_data->health); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&consumer_data->health); + health_unregister(); DBG("consumer thread cleanup completed"); return NULL; @@ -1103,11 +1109,13 @@ static void *thread_manage_apps(void *data) rcu_register_thread(); rcu_thread_online(); + health_register(HEALTH_TYPE_APP_MANAGE); + if (testpoint(thread_manage_apps)) { goto error_testpoint; } - health_code_update(&health_thread_app_manage); + health_code_update(); ret = create_thread_poll_set(&events, 2); if (ret < 0) { @@ -1123,16 +1131,16 @@ static void *thread_manage_apps(void *data) goto error; } - health_code_update(&health_thread_app_manage); + health_code_update(); while (1) { DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events)); /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_app_manage); + health_poll_update(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_app_manage); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -1150,7 +1158,7 @@ static void *thread_manage_apps(void *data) revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_app_manage); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -1166,13 +1174,15 @@ static void *thread_manage_apps(void *data) goto error; } else if (revents & LPOLLIN) { /* Empty pipe */ - ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd)); + do { + ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd)); + } while (ret < 0 && errno == EINTR); if (ret < 0 || ret < sizeof(ust_cmd)) { PERROR("read apps cmd pipe"); goto error; } - health_code_update(&health_thread_app_manage); + health_code_update(); /* Register applicaton to the session daemon */ ret = ust_app_register(&ust_cmd.reg_msg, @@ -1183,7 +1193,7 @@ static void *thread_manage_apps(void *data) break; } - health_code_update(&health_thread_app_manage); + health_code_update(); /* * Validate UST version compatibility. @@ -1197,7 +1207,7 @@ static void *thread_manage_apps(void *data) update_ust_app(ust_cmd.sock); } - health_code_update(&health_thread_app_manage); + health_code_update(); ret = ust_app_register_done(ust_cmd.sock); if (ret < 0) { @@ -1228,7 +1238,7 @@ static void *thread_manage_apps(void *data) ust_cmd.sock); } - health_code_update(&health_thread_app_manage); + health_code_update(); break; } @@ -1250,7 +1260,7 @@ static void *thread_manage_apps(void *data) } } - health_code_update(&health_thread_app_manage); + health_code_update(); } } @@ -1269,10 +1279,10 @@ error_testpoint: */ if (err) { - health_error(&health_thread_app_manage); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&health_thread_app_manage); + health_unregister(); DBG("Application communication apps thread cleanup complete"); rcu_thread_offline(); rcu_unregister_thread(); @@ -1318,9 +1328,11 @@ static void *thread_dispatch_ust_registration(void *data) * at some point in time or wait to the end of the world :) */ if (apps_cmd_pipe[1] >= 0) { - ret = write(apps_cmd_pipe[1], ust_cmd, - sizeof(struct ust_command)); - if (ret < 0) { + do { + ret = write(apps_cmd_pipe[1], ust_cmd, + sizeof(struct ust_command)); + } while (ret < 0 && errno == EINTR); + if (ret < 0 || ret != sizeof(struct ust_command)) { PERROR("write apps cmd pipe"); if (errno == EBADF) { /* @@ -1367,6 +1379,8 @@ static void *thread_registration_apps(void *data) DBG("[thread] Manage application registration started"); + health_register(HEALTH_TYPE_APP_REG); + if (testpoint(thread_registration_apps)) { goto error_testpoint; } @@ -1404,9 +1418,9 @@ static void *thread_registration_apps(void *data) /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_app_reg); + health_poll_update(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_app_reg); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -1420,7 +1434,7 @@ static void *thread_registration_apps(void *data) nb_fd = ret; for (i = 0; i < nb_fd; i++) { - health_code_update(&health_thread_app_reg); + health_code_update(); /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -1472,7 +1486,7 @@ static void *thread_registration_apps(void *data) sock = -1; continue; } - health_code_update(&health_thread_app_reg); + health_code_update(); ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg, sizeof(struct ust_register_msg)); if (ret < 0 || ret < sizeof(struct ust_register_msg)) { @@ -1490,7 +1504,7 @@ static void *thread_registration_apps(void *data) sock = -1; continue; } - health_code_update(&health_thread_app_reg); + health_code_update(); ust_cmd->sock = sock; sock = -1; @@ -1521,7 +1535,7 @@ static void *thread_registration_apps(void *data) exit: error: if (err) { - health_error(&health_thread_app_reg); + health_error(); ERR("Health error occurred in %s", __func__); } @@ -1549,7 +1563,7 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); - health_exit(&health_thread_app_reg); + health_unregister(); return NULL; } @@ -1924,9 +1938,7 @@ static int check_consumer_health(void) { int ret; - ret = health_check_state(&kconsumer_data.health) && - health_check_state(&ustconsumer32_data.health) && - health_check_state(&ustconsumer64_data.health); + ret = health_check_state(HEALTH_TYPE_CONSUMER); DBG3("Health consumer check %d", ret); @@ -3044,26 +3056,26 @@ restart: switch (msg.component) { case LTTNG_HEALTH_CMD: - reply.ret_code = health_check_state(&health_thread_cmd); + reply.ret_code = health_check_state(HEALTH_TYPE_CMD); break; case LTTNG_HEALTH_APP_MANAGE: - reply.ret_code = health_check_state(&health_thread_app_manage); + reply.ret_code = health_check_state(HEALTH_TYPE_APP_MANAGE); break; case LTTNG_HEALTH_APP_REG: - reply.ret_code = health_check_state(&health_thread_app_reg); + reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG); break; case LTTNG_HEALTH_KERNEL: - reply.ret_code = health_check_state(&health_thread_kernel); + reply.ret_code = health_check_state(HEALTH_TYPE_KERNEL); break; case LTTNG_HEALTH_CONSUMER: reply.ret_code = check_consumer_health(); break; case LTTNG_HEALTH_ALL: reply.ret_code = - health_check_state(&health_thread_app_manage) && - health_check_state(&health_thread_app_reg) && - health_check_state(&health_thread_cmd) && - health_check_state(&health_thread_kernel) && + 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(); break; default: @@ -3138,11 +3150,13 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); + health_register(HEALTH_TYPE_CMD); + if (testpoint(thread_manage_clients)) { goto error_testpoint; } - health_code_update(&health_thread_cmd); + health_code_update(); ret = lttcomm_listen_unix_sock(client_sock); if (ret < 0) { @@ -3175,16 +3189,16 @@ static void *thread_manage_clients(void *data) goto error; } - health_code_update(&health_thread_cmd); + health_code_update(); while (1) { DBG("Accepting client command ..."); /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(&health_thread_cmd); + health_poll_update(); ret = lttng_poll_wait(&events, -1); - health_poll_update(&health_thread_cmd); + health_poll_update(); if (ret < 0) { /* * Restart interrupted system call. @@ -3202,7 +3216,7 @@ static void *thread_manage_clients(void *data) revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(&health_thread_cmd); + health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ ret = check_thread_quit_pipe(pollfd, revents); @@ -3222,7 +3236,7 @@ static void *thread_manage_clients(void *data) DBG("Wait for client response"); - health_code_update(&health_thread_cmd); + health_code_update(); sock = lttcomm_accept_unix_sock(client_sock); if (sock < 0) { @@ -3258,7 +3272,7 @@ static void *thread_manage_clients(void *data) cmd_ctx->llm = NULL; cmd_ctx->session = NULL; - health_code_update(&health_thread_cmd); + health_code_update(); /* * Data is received from the lttng client. The struct @@ -3279,7 +3293,7 @@ static void *thread_manage_clients(void *data) continue; } - health_code_update(&health_thread_cmd); + health_code_update(); // TODO: Validate cmd_ctx including sanity check for // security purpose. @@ -3312,7 +3326,7 @@ static void *thread_manage_clients(void *data) continue; } - health_code_update(&health_thread_cmd); + health_code_update(); DBG("Sending response (size: %d, retcode: %s)", cmd_ctx->lttng_msg_size, @@ -3331,7 +3345,7 @@ static void *thread_manage_clients(void *data) clean_command_ctx(&cmd_ctx); - health_code_update(&health_thread_cmd); + health_code_update(); } exit: @@ -3358,11 +3372,11 @@ error_testpoint: } if (err) { - health_error(&health_thread_cmd); + health_error(); ERR("Health error occurred in %s", __func__); } - health_exit(&health_thread_cmd); + health_unregister(); DBG("Client thread dying"); @@ -4106,26 +4120,6 @@ int main(int argc, char **argv) cmd_init(); - /* Init all health thread counters. */ - health_init(&health_thread_cmd); - health_init(&health_thread_kernel); - health_init(&health_thread_app_manage); - health_init(&health_thread_app_reg); - - /* - * Init health counters of the consumer thread. We do a quick hack here to - * the state of the consumer health is fine even if the thread is not - * started. Once the thread starts, the health state is updated with a poll - * value to set a health code path. This is simply to ease our life and has - * no cost what so ever. - */ - health_init(&kconsumer_data.health); - health_poll_update(&kconsumer_data.health); - health_init(&ustconsumer32_data.health); - health_poll_update(&ustconsumer32_data.health); - health_init(&ustconsumer64_data.health); - health_poll_update(&ustconsumer64_data.health); - /* Check for the application socket timeout env variable. */ env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV); if (env_app_timeout) {