X-Git-Url: http://git.lttng.org./?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=edf52410fc699c2c8aaa38588b5b4bf8dc54a7c5;hb=ff1a2c3965e5222769364e9b1cbef06bd32066d3;hp=1d047fd85240582854f5a6c50744b23ce7ac4cb1;hpb=27babd3a0a164f71d4dc02884ef20099ddfc6755;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 1d047fd85..edf52410f 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -90,7 +90,6 @@ static struct consumer_data kconsumer_data = { .cmd_unix_sock_path = DEFAULT_KCONSUMERD_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, - .metadata_sock.fd = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -102,7 +101,6 @@ static struct consumer_data ustconsumer64_data = { .cmd_unix_sock_path = DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, - .metadata_sock.fd = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -114,7 +112,6 @@ static struct consumer_data ustconsumer32_data = { .cmd_unix_sock_path = DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, .err_sock = -1, .cmd_sock = -1, - .metadata_sock.fd = -1, .pid_mutex = PTHREAD_MUTEX_INITIALIZER, .lock = PTHREAD_MUTEX_INITIALIZER, .cond = PTHREAD_COND_INITIALIZER, @@ -446,7 +443,10 @@ static void cleanup(void) DBG("Cleaning up"); - /* First thing first, stop all threads */ + /* + * Close the thread quit pipe. It has already done its job, + * since we are now called. + */ utils_close_pipe(thread_quit_pipe); /* @@ -675,6 +675,8 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) if (ret < 0) { goto error; } + /* Update the stream global counter */ + ksess->stream_count_global += ret; /* * Have we already sent fds to the consumer? If yes, it means @@ -688,9 +690,6 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) rcu_read_lock(); cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); ret = kernel_consumer_send_channel_stream(socket, channel, ksess, @@ -725,6 +724,12 @@ static void update_ust_app(int app_sock) { struct ltt_session *sess, *stmp; + /* Consumer is in an ERROR state. Stop any application update. */ + if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { + /* Stop the update process since the consumer is dead. */ + return; + } + /* For all tracing session(s) */ cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) { session_lock(sess); @@ -1014,15 +1019,16 @@ restart: /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); - consumer_data->metadata_sock.fd = + consumer_data->metadata_fd = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); - if (consumer_data->cmd_sock < 0 || - consumer_data->metadata_sock.fd < 0) { + if (consumer_data->cmd_sock < 0 + || consumer_data->metadata_fd < 0) { PERROR("consumer connect cmd socket"); /* On error, signal condition and quit. */ signal_consumer_condition(consumer_data, -1); goto error; } + consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd; /* Create metadata socket lock. */ consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t)); if (consumer_data->metadata_sock.lock == NULL) { @@ -1035,7 +1041,7 @@ restart: signal_consumer_condition(consumer_data, 1); DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock); DBG("Consumer metadata socket ready (fd: %d)", - consumer_data->metadata_sock.fd); + consumer_data->metadata_fd); } else { ERR("consumer error when waiting for SOCK_READY : %s", lttcomm_get_readable_code(-code)); @@ -1055,7 +1061,7 @@ restart: } /* Add metadata socket that is successfully connected. */ - ret = lttng_poll_add(&events, consumer_data->metadata_sock.fd, + ret = lttng_poll_add(&events, consumer_data->metadata_fd, LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; @@ -1114,7 +1120,7 @@ restart_poll: lttcomm_get_readable_code(-code)); goto exit; - } else if (pollfd == consumer_data->metadata_sock.fd) { + } else if (pollfd == consumer_data->metadata_fd) { /* UST metadata requests */ ret = ust_consumer_metadata_request( &consumer_data->metadata_sock); @@ -1133,6 +1139,13 @@ restart_poll: exit: error: + /* + * We lock here because we are about to close the sockets and some other + * thread might be using them so get exclusive access which will abort all + * other consumer command by other threads. + */ + pthread_mutex_lock(&consumer_data->lock); + /* Immediately set the consumerd state to stopped */ if (consumer_data->type == LTTNG_CONSUMER_KERNEL) { uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR); @@ -1158,15 +1171,12 @@ error: } consumer_data->cmd_sock = -1; } - if (consumer_data->metadata_sock.fd >= 0) { - ret = close(consumer_data->metadata_sock.fd); + if (*consumer_data->metadata_sock.fd_ptr >= 0) { + ret = close(*consumer_data->metadata_sock.fd_ptr); if (ret) { PERROR("close"); } } - /* Cleanup metadata socket mutex. */ - pthread_mutex_destroy(consumer_data->metadata_sock.lock); - free(consumer_data->metadata_sock.lock); if (sock >= 0) { ret = close(sock); @@ -1178,6 +1188,11 @@ error: unlink(consumer_data->err_unix_sock_path); unlink(consumer_data->cmd_unix_sock_path); consumer_data->pid = 0; + pthread_mutex_unlock(&consumer_data->lock); + + /* Cleanup metadata socket mutex. */ + pthread_mutex_destroy(consumer_data->metadata_sock.lock); + free(consumer_data->metadata_sock.lock); lttng_poll_clean(&events); error_poll: @@ -1293,12 +1308,6 @@ static void *thread_manage_apps(void *data) goto error; } - /* Set socket timeout for both receiving and ending */ - (void) lttcomm_setsockopt_rcv_timeout(sock, - app_socket_timeout); - (void) lttcomm_setsockopt_snd_timeout(sock, - app_socket_timeout); - DBG("Apps with sock %d added to poll set", sock); health_code_update(); @@ -1356,6 +1365,9 @@ error_testpoint: * Send a socket to a thread This is called from the dispatch UST registration * thread once all sockets are set for the application. * + * The sock value can be invalid, we don't really care, the thread will handle + * it and make the necessary cleanup if so. + * * On success, return 0 else a negative value being the errno message of the * write(). */ @@ -1363,9 +1375,14 @@ static int send_socket_to_thread(int fd, int sock) { int ret; - /* Sockets MUST be set or else this should not have been called. */ - assert(fd >= 0); - assert(sock >= 0); + /* + * It's possible that the FD is set as invalid with -1 concurrently just + * before calling this function being a shutdown state of the thread. + */ + if (fd < 0) { + ret = -EBADF; + goto error; + } do { ret = write(fd, &sock, sizeof(sock)); @@ -1534,7 +1551,7 @@ static void *thread_dispatch_ust_registration(void *data) if (ret < 0) { PERROR("close ust sock dispatch %d", ust_cmd->sock); } - lttng_fd_put(1, LTTNG_FD_APPS); + lttng_fd_put(LTTNG_FD_APPS, 1); free(ust_cmd); goto error; } @@ -1548,7 +1565,7 @@ static void *thread_dispatch_ust_registration(void *data) if (ret < 0) { PERROR("close ust sock dispatch %d", ust_cmd->sock); } - lttng_fd_put(1, LTTNG_FD_APPS); + lttng_fd_put(LTTNG_FD_APPS, 1); free(wait_node); free(ust_cmd); continue; @@ -1596,7 +1613,7 @@ static void *thread_dispatch_ust_registration(void *data) if (ret < 0) { PERROR("close ust sock dispatch %d", ust_cmd->sock); } - lttng_fd_put(1, LTTNG_FD_APPS); + lttng_fd_put(LTTNG_FD_APPS, 1); } free(ust_cmd); } @@ -1628,7 +1645,12 @@ static void *thread_dispatch_ust_registration(void *data) if (ret < 0) { rcu_read_unlock(); session_unlock_list(); - /* No notify thread, stop the UST tracing. */ + /* + * No notify thread, stop the UST tracing. However, this is + * not an internal error of the this thread thus setting + * the health error code to a normal exit. + */ + err = 0; goto error; } @@ -1653,7 +1675,12 @@ static void *thread_dispatch_ust_registration(void *data) if (ret < 0) { rcu_read_unlock(); session_unlock_list(); - /* No apps. thread, stop the UST tracing. */ + /* + * No apps. thread, stop the UST tracing. However, this is + * not an internal error of the this thread thus setting + * the health error code to a normal exit. + */ + err = 0; goto error; } @@ -1783,6 +1810,18 @@ static void *thread_registration_apps(void *data) goto error; } + /* + * Set socket timeout for both receiving and ending. + * app_socket_timeout is in seconds, whereas + * lttcomm_setsockopt_rcv_timeout and + * lttcomm_setsockopt_snd_timeout expect msec as + * parameter. + */ + (void) lttcomm_setsockopt_rcv_timeout(sock, + app_socket_timeout * 1000); + (void) lttcomm_setsockopt_snd_timeout(sock, + app_socket_timeout * 1000); + /* * Set the CLOEXEC flag. Return code is useless because * either way, the show must go on. @@ -2699,6 +2738,10 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, break; case LTTNG_DOMAIN_UST: { + if (!ust_app_supported()) { + ret = LTTNG_ERR_NO_UST; + goto error; + } /* Consumer is in an ERROR state. Report back to client */ if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { ret = LTTNG_ERR_NO_USTCONSUMERD; @@ -4064,12 +4107,12 @@ static int set_permissions(char *rundir) ret = allowed_group(); if (ret < 0) { WARN("No tracing group detected"); - ret = 0; - goto end; + /* Setting gid to 0 if no tracing group is found */ + gid = 0; + } else { + gid = ret; } - gid = ret; - /* Set lttng run dir */ ret = chown(rundir, 0, gid); if (ret < 0) { @@ -4077,7 +4120,7 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure tracing group can search the run dir */ + /* Ensure all applications and tracing group can search the run dir */ ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH); if (ret < 0) { ERR("Unable to set permissions on %s", rundir); @@ -4114,7 +4157,6 @@ static int set_permissions(char *rundir) DBG("All permissions are set"); -end: return ret; } @@ -4626,6 +4668,17 @@ int main(int argc, char **argv) write_pidfile(); + /* Initialize communication library */ + lttcomm_init(); + /* This is to get the TCP timeout value. */ + lttcomm_inet_init(); + + /* + * Initialize the health check subsystem. This call should set the + * appropriate time values. + */ + health_init(); + /* Create thread to manage the client socket */ ret = pthread_create(&ht_cleanup_thread, NULL, thread_ht_cleanup, (void *) NULL);