X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=d999928feb8a0f06bfe24857fc3182f56c4f903f;hp=7e9f9dc643b3c3f97fa52f261f47f14c11bff9a7;hb=927ca06aed61ff6dd3f64ae71854f2d7f9acebe5;hpb=6993eeb37fb4caf9b92c09d4ab1730dcc2b8b097 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 7e9f9dc64..d999928fe 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -72,6 +72,7 @@ const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE; const char *progname; const char *opt_tracing_group; +static const char *opt_pidfile; static int opt_sig_parent; static int opt_verbose_consumer; static int opt_daemon; @@ -404,6 +405,17 @@ static void cleanup(void) /* First thing first, stop all threads */ utils_close_pipe(thread_quit_pipe); + /* + * If opt_pidfile is undefined, the default file will be wiped when + * removing the rundir. + */ + if (opt_pidfile) { + ret = remove(opt_pidfile); + if (ret < 0) { + PERROR("remove pidfile %s", opt_pidfile); + } + } + DBG("Removing %s directory", rundir); ret = asprintf(&cmd, "rm -rf %s", rundir); if (ret < 0) { @@ -461,7 +473,7 @@ static void cleanup(void) static int send_unix_sock(int sock, void *buf, size_t len) { /* Check valid length */ - if (len <= 0) { + if (len == 0) { return -1; } @@ -626,7 +638,7 @@ 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 +649,11 @@ 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; } @@ -691,36 +705,42 @@ 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. + */ + memset(&events, 0, sizeof(events)); + if (testpoint(thread_manage_kernel)) { goto error_testpoint; } health_code_update(&health_thread_kernel); - ret = create_thread_poll_set(&events, 2); - if (ret < 0) { - goto error_poll_create; - } - - ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); - if (ret < 0) { - goto error; - } - if (testpoint(thread_manage_kernel_before_loop)) { - goto error; + goto error_testpoint; } while (1) { health_code_update(&health_thread_kernel); if (update_poll_flag == 1) { - /* - * Reset number of fd in the poll set. Always 2 since there is the thread - * quit pipe and the kernel pipe. - */ - events.nb_fd = 2; + /* Clean events object. We are about to populate it again. */ + lttng_poll_clean(&events); + + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error_poll_create; + } + + ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); + if (ret < 0) { + goto error; + } + /* This will add the available kernel channel if any. */ ret = update_kernel_poll(&events); if (ret < 0) { goto error; @@ -728,12 +748,7 @@ static void *thread_manage_kernel(void *data) update_poll_flag = 0; } - nb_fd = LTTNG_POLL_GETNB(&events); - - DBG("Thread kernel polling on %d fds", nb_fd); - - /* Zeroed the poll events */ - lttng_poll_reset(&events); + DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events)); /* Poll infinite value of time */ restart: @@ -755,6 +770,8 @@ static void *thread_manage_kernel(void *data) continue; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -771,7 +788,13 @@ 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 { @@ -807,7 +830,7 @@ error_testpoint: 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; } @@ -848,6 +871,8 @@ 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). @@ -886,8 +911,6 @@ static void *thread_manage_consumer(void *data) goto error; } - nb_fd = LTTNG_POLL_GETNB(&events); - health_code_update(&consumer_data->health); /* Inifinite blocking call, waiting for transmission */ @@ -910,6 +933,8 @@ restart: goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -987,9 +1012,6 @@ restart: health_code_update(&consumer_data->health); - /* Update number of fd */ - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart_poll: health_poll_update(&consumer_data->health); @@ -1005,6 +1027,8 @@ restart_poll: goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -1082,7 +1106,7 @@ error_poll: health_error(&consumer_data->health); ERR("Health error occurred in %s", __func__); } - health_exit(&consumer_data->health); + health_unregister(); DBG("consumer thread cleanup completed"); return NULL; @@ -1103,6 +1127,8 @@ 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; } @@ -1126,12 +1152,7 @@ static void *thread_manage_apps(void *data) health_code_update(&health_thread_app_manage); while (1) { - /* Zeroed the events structure */ - lttng_poll_reset(&events); - - nb_fd = LTTNG_POLL_GETNB(&events); - - DBG("Apps thread polling on %d fds", nb_fd); + DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events)); /* Inifinite blocking call, waiting for transmission */ restart: @@ -1148,6 +1169,8 @@ static void *thread_manage_apps(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -1169,7 +1192,9 @@ 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; @@ -1275,7 +1300,7 @@ error_testpoint: health_error(&health_thread_app_manage); 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(); @@ -1321,9 +1346,11 @@ 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) { /* @@ -1370,6 +1397,8 @@ 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; } @@ -1405,8 +1434,6 @@ static void *thread_registration_apps(void *data) while (1) { DBG("Accepting application registration"); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: health_poll_update(&health_thread_app_reg); @@ -1422,6 +1449,8 @@ static void *thread_registration_apps(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { health_code_update(&health_thread_app_reg); @@ -1552,7 +1581,7 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); - health_exit(&health_thread_app_reg); + health_unregister(); return NULL; } @@ -1684,10 +1713,10 @@ error: static int join_consumer_thread(struct consumer_data *consumer_data) { void *status; - int ret; /* Consumer pid must be a real one. */ if (consumer_data->pid > 0) { + int ret; ret = kill(consumer_data->pid, SIGTERM); if (ret) { ERR("Error killing consumer daemon"); @@ -1866,7 +1895,7 @@ error: */ static int start_consumerd(struct consumer_data *consumer_data) { - int ret, err; + int ret; /* * Set the listen() state on the socket since there is a possible race @@ -1909,6 +1938,8 @@ end: error: /* Cleanup already created socket on error. */ if (consumer_data->err_sock >= 0) { + int err; + err = close(consumer_data->err_sock); if (err < 0) { PERROR("close consumer data error socket"); @@ -1925,9 +1956,7 @@ 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); @@ -2081,7 +2110,7 @@ static int create_ust_session(struct ltt_session *session, DBG("Creating UST session"); - lus = trace_ust_create_session(session->path, session->id, domain); + lus = trace_ust_create_session(session->path, session->id); if (lus == NULL) { ret = LTTNG_ERR_UST_SESS_FAIL; goto error; @@ -2982,8 +3011,6 @@ static void *thread_manage_health(void *data) while (1) { DBG("Health check ready"); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: ret = lttng_poll_wait(&events, -1); @@ -2997,6 +3024,8 @@ restart: goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -3045,26 +3074,26 @@ 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: @@ -3139,6 +3168,8 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); + health_register(HEALTH_TYPE_CMD); + if (testpoint(thread_manage_clients)) { goto error_testpoint; } @@ -3181,8 +3212,6 @@ static void *thread_manage_clients(void *data) while (1) { DBG("Accepting client command ..."); - nb_fd = LTTNG_POLL_GETNB(&events); - /* Inifinite blocking call, waiting for transmission */ restart: health_poll_update(&health_thread_cmd); @@ -3198,6 +3227,8 @@ static void *thread_manage_clients(void *data) goto error; } + nb_fd = ret; + for (i = 0; i < nb_fd; i++) { /* Fetch once the poll data */ revents = LTTNG_POLL_GETEV(&events, i); @@ -3363,7 +3394,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_exit(&health_thread_cmd); + health_unregister(); DBG("Client thread dying"); @@ -3397,6 +3428,7 @@ static void usage(void) fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n"); fprintf(stderr, " -q, --quiet No output at all.\n"); fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + fprintf(stderr, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n"); fprintf(stderr, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n"); fprintf(stderr, " --no-kernel Disable kernel tracer\n"); } @@ -3430,12 +3462,13 @@ static int parse_args(int argc, char **argv) { "verbose", 0, 0, 'v' }, { "verbose-consumer", 0, 0, 'Z' }, { "no-kernel", 0, 0, 'N' }, + { "pidfile", 1, 0, 'p' }, { NULL, 0, 0, 0 } }; while (1) { int option_index = 0; - c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t", + c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:", long_options, &option_index); if (c == -1) { break; @@ -3512,6 +3545,9 @@ static int parse_args(int argc, char **argv) case 'T': consumerd64_libdir = optarg; break; + case 'p': + opt_pidfile = optarg; + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -3838,6 +3874,38 @@ static void set_ulimit(void) } } +/* + * Write pidfile using the rundir and opt_pidfile. + */ +static void write_pidfile(void) +{ + int ret; + char pidfile_path[PATH_MAX]; + + assert(rundir); + + if (opt_pidfile) { + strncpy(pidfile_path, opt_pidfile, sizeof(pidfile_path)); + } else { + /* Build pidfile path from rundir and opt_pidfile. */ + ret = snprintf(pidfile_path, sizeof(pidfile_path), "%s/" + DEFAULT_LTTNG_SESSIOND_PIDFILE, rundir); + if (ret < 0) { + PERROR("snprintf pidfile path"); + goto error; + } + } + + /* + * Create pid file in rundir. Return value is of no importance. The + * execution will continue even though we are not able to write the file. + */ + (void) utils_create_pid_file(getpid(), pidfile_path); + +error: + return; +} + /* * main */ @@ -3855,7 +3923,7 @@ int main(int argc, char **argv) /* Parse arguments */ progname = argv[0]; - if ((ret = parse_args(argc, argv) < 0)) { + if ((ret = parse_args(argc, argv)) < 0) { goto error; } @@ -4107,26 +4175,6 @@ 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) { @@ -4135,6 +4183,8 @@ int main(int argc, char **argv) app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT; } + write_pidfile(); + /* Create thread to manage the client socket */ ret = pthread_create(&health_thread, NULL, thread_manage_health, (void *) NULL);