X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=076d0b7b03ce24295d8d7aeb924c7cdbcb117e16;hp=f1cd1b6407f4ee6f34cd18d976368825c3d5d7e3;hb=db8f1377d850c1b18d33ddc2a6f2f3c91c3210bc;hpb=380e8d6f886dfe6633044abfa77739d1210979a3 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index f1cd1b640..076d0b7b0 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -20,6 +20,7 @@ #include #include #include +#include #include #include #include @@ -60,17 +61,15 @@ #include "ust-consumer.h" #include "utils.h" #include "fd-limit.h" -#include "health.h" +#include "health-sessiond.h" #include "testpoint.h" #include "ust-thread.h" +#include "jul-thread.h" #define CONSUMERD_FILE "lttng-consumerd" -/* Const values */ -const char default_tracing_group[] = DEFAULT_TRACING_GROUP; - const char *progname; -const char *opt_tracing_group; +static const char *tracing_group_name = DEFAULT_TRACING_GROUP; static const char *opt_pidfile; static int opt_sig_parent; static int opt_verbose_consumer; @@ -78,8 +77,12 @@ static int opt_daemon; static int opt_no_kernel; static int is_root; /* Set to 1 if the daemon is running as root */ static pid_t ppid; /* Parent PID for --sig-parent option */ +static pid_t child_ppid; /* Internal parent PID use with daemonize. */ static char *rundir; +/* Set to 1 when a SIGUSR1 signal is received. */ +static int recv_child_signal; + /* * Consumer daemon specific control data. Every value not initialized here is * set to 0 by the static definition. @@ -90,7 +93,6 @@ 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 +104,6 @@ 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 +115,6 @@ 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, @@ -162,6 +162,7 @@ static pthread_t kernel_thread; static pthread_t dispatch_thread; static pthread_t health_thread; static pthread_t ht_cleanup_thread; +static pthread_t jul_reg_thread; /* * UST registration command queue. This queue is tied with a futex and uses a N @@ -236,6 +237,12 @@ static int app_socket_timeout; /* Set in main() with the current page size. */ long page_size; +/* Application health monitoring */ +struct health_app *health_sessiond; + +/* JUL TCP port for registration. Used by the JUL thread. */ +unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT; + static void setup_consumerd_path(void) { @@ -325,25 +332,6 @@ int sessiond_check_thread_quit_pipe(int fd, uint32_t events) return 0; } -/* - * Return group ID of the tracing group or -1 if not found. - */ -static gid_t allowed_group(void) -{ - struct group *grp; - - if (opt_tracing_group) { - grp = getgrnam(opt_tracing_group); - } else { - grp = getgrnam(default_tracing_group); - } - if (!grp) { - return -1; - } else { - return grp->gr_gid; - } -} - /* * Init thread quit pipe. * @@ -390,18 +378,66 @@ static void stop_threads(void) futex_nto1_wake(&ust_cmd_queue.futex); } +/* + * Close every consumer sockets. + */ +static void close_consumer_sockets(void) +{ + int ret; + + if (kconsumer_data.err_sock >= 0) { + ret = close(kconsumer_data.err_sock); + if (ret < 0) { + PERROR("kernel consumer err_sock close"); + } + } + if (ustconsumer32_data.err_sock >= 0) { + ret = close(ustconsumer32_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd32 err_sock close"); + } + } + if (ustconsumer64_data.err_sock >= 0) { + ret = close(ustconsumer64_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd64 err_sock close"); + } + } + if (kconsumer_data.cmd_sock >= 0) { + ret = close(kconsumer_data.cmd_sock); + if (ret < 0) { + PERROR("kernel consumer cmd_sock close"); + } + } + if (ustconsumer32_data.cmd_sock >= 0) { + ret = close(ustconsumer32_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd32 cmd_sock close"); + } + } + if (ustconsumer64_data.cmd_sock >= 0) { + ret = close(ustconsumer64_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd64 cmd_sock close"); + } + } +} + /* * Cleanup the daemon */ static void cleanup(void) { int ret; - char *cmd = NULL; struct ltt_session *sess, *stmp; + char path[PATH_MAX]; 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); /* @@ -415,18 +451,65 @@ static void cleanup(void) } } - DBG("Removing %s directory", rundir); - ret = asprintf(&cmd, "rm -rf %s", rundir); - if (ret < 0) { - ERR("asprintf failed. Something is really wrong!"); - } + DBG("Removing sessiond and consumerd content of directory %s", rundir); + + /* sessiond */ + snprintf(path, PATH_MAX, + "%s/%s", + rundir, DEFAULT_LTTNG_SESSIOND_PIDFILE); + DBG("Removing %s", path); + (void) unlink(path); + + snprintf(path, PATH_MAX, "%s/%s", rundir, + DEFAULT_LTTNG_SESSIOND_JULPORT_FILE); + DBG("Removing %s", path); + (void) unlink(path); + + /* kconsumerd */ + snprintf(path, PATH_MAX, + DEFAULT_KCONSUMERD_ERR_SOCK_PATH, + rundir); + DBG("Removing %s", path); + (void) unlink(path); + + snprintf(path, PATH_MAX, + DEFAULT_KCONSUMERD_PATH, + rundir); + DBG("Removing directory %s", path); + (void) rmdir(path); + + /* ust consumerd 32 */ + snprintf(path, PATH_MAX, + DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, + rundir); + DBG("Removing %s", path); + (void) unlink(path); + + snprintf(path, PATH_MAX, + DEFAULT_USTCONSUMERD32_PATH, + rundir); + DBG("Removing directory %s", path); + (void) rmdir(path); + + /* ust consumerd 64 */ + snprintf(path, PATH_MAX, + DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, + rundir); + DBG("Removing %s", path); + (void) unlink(path); + + snprintf(path, PATH_MAX, + DEFAULT_USTCONSUMERD64_PATH, + rundir); + DBG("Removing directory %s", path); + (void) rmdir(path); + + /* + * We do NOT rmdir rundir because there are other processes + * using it, for instance lttng-relayd, which can start in + * parallel with this teardown. + */ - /* Remove lttng run directory */ - ret = system(cmd); - if (ret < 0) { - ERR("Unable to clean %s", rundir); - } - free(cmd); free(rundir); DBG("Cleaning up all sessions"); @@ -458,6 +541,8 @@ static void cleanup(void) modprobe_remove_lttng_all(); } + close_consumer_sockets(); + /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", @@ -628,6 +713,8 @@ 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 @@ -641,12 +728,10 @@ 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); + channel, ksess, + session->output_traces ? 1 : 0); pthread_mutex_unlock(socket->lock); if (ret < 0) { rcu_read_unlock(); @@ -677,6 +762,12 @@ 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); @@ -702,7 +793,7 @@ static void *thread_manage_kernel(void *data) DBG("[thread] Thread manage kernel started"); - health_register(HEALTH_TYPE_KERNEL); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_KERNEL); /* * This first step of the while is to clean this structure which could free @@ -785,9 +876,8 @@ static void *thread_manage_kernel(void *data) /* Check for data on kernel pipe */ if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) { - do { - ret = read(kernel_poll_pipe[0], &tmp, 1); - } while (ret < 0 && errno == EINTR); + (void) lttng_read(kernel_poll_pipe[0], + &tmp, 1); /* * Ret value is useless here, if this pipe gets any actions an * update is required anyway. @@ -827,7 +917,7 @@ error_testpoint: WARN("Kernel thread died unexpectedly. " "Kernel tracing can continue but CPU hotplug is disabled."); } - health_unregister(); + health_unregister(health_sessiond); DBG("Kernel thread dying"); return NULL; } @@ -868,7 +958,7 @@ static void *thread_manage_consumer(void *data) DBG("[thread] Manage consumer started"); - health_register(HEALTH_TYPE_CONSUMER); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); health_code_update(); @@ -966,15 +1056,16 @@ 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) { @@ -987,7 +1078,7 @@ 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)); @@ -1007,7 +1098,7 @@ 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; @@ -1066,7 +1157,7 @@ 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); @@ -1085,6 +1176,13 @@ 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); @@ -1101,22 +1199,21 @@ error: if (ret) { PERROR("close"); } + consumer_data->err_sock = -1; } if (consumer_data->cmd_sock >= 0) { ret = close(consumer_data->cmd_sock); if (ret) { PERROR("close"); } + 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); @@ -1128,6 +1225,11 @@ 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: @@ -1135,7 +1237,7 @@ error_poll: health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); return NULL; @@ -1147,6 +1249,7 @@ error_poll: static void *thread_manage_apps(void *data) { int i, ret, pollfd, err = -1; + ssize_t size_ret; uint32_t revents, nb_fd; struct lttng_poll_event events; @@ -1155,7 +1258,7 @@ static void *thread_manage_apps(void *data) rcu_register_thread(); rcu_thread_online(); - health_register(HEALTH_TYPE_APP_MANAGE); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE); if (testpoint(thread_manage_apps)) { goto error_testpoint; @@ -1222,10 +1325,8 @@ static void *thread_manage_apps(void *data) int sock; /* Empty pipe */ - do { - ret = read(apps_cmd_pipe[0], &sock, sizeof(sock)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret < sizeof(sock)) { + size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock)); + if (size_ret < sizeof(sock)) { PERROR("read apps cmd pipe"); goto error; } @@ -1243,12 +1344,6 @@ 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(); @@ -1295,7 +1390,7 @@ error_testpoint: health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Application communication apps thread cleanup complete"); rcu_thread_offline(); rcu_unregister_thread(); @@ -1306,21 +1401,27 @@ 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(). */ static int send_socket_to_thread(int fd, int sock) { - int ret; + ssize_t 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)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != sizeof(sock)) { + ret = lttng_write(fd, &sock, sizeof(sock)); + if (ret < sizeof(sock)) { PERROR("write apps pipe %d", fd); if (ret < 0) { ret = -errno; @@ -1331,7 +1432,92 @@ static int send_socket_to_thread(int fd, int sock) /* All good. Don't send back the write positive ret value. */ ret = 0; error: - return ret; + return (int) 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; } /* @@ -1340,19 +1526,25 @@ error: */ 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, + }; - CDS_LIST_HEAD(wait_queue); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); + + health_code_update(); + + 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); @@ -1360,6 +1552,14 @@ 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) { @@ -1385,7 +1585,7 @@ 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; } @@ -1399,7 +1599,7 @@ 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; @@ -1408,7 +1608,8 @@ 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); /* @@ -1423,10 +1624,12 @@ 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); @@ -1444,7 +1647,7 @@ 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); } @@ -1476,7 +1679,12 @@ 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; } @@ -1501,7 +1709,12 @@ 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; } @@ -1510,19 +1723,29 @@ static void *thread_dispatch_ust_registration(void *data) } } 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(health_sessiond); return NULL; } @@ -1542,7 +1765,7 @@ static void *thread_registration_apps(void *data) DBG("[thread] Manage application registration started"); - health_register(HEALTH_TYPE_APP_REG); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG); if (testpoint(thread_registration_apps)) { goto error_testpoint; @@ -1621,6 +1844,18 @@ 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. @@ -1722,7 +1957,7 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); - health_unregister(); + health_unregister(health_sessiond); return NULL; } @@ -1923,6 +2158,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) "lttng-consumerd", verbosity, "-k", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); break; case LTTNG_CONSUMER64_UST: @@ -1961,6 +2197,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = execl(consumerd64_bin, "lttng-consumerd", verbosity, "-u", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); if (consumerd64_libdir[0] != '\0') { free(tmpnew); @@ -2006,6 +2243,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = execl(consumerd32_bin, "lttng-consumerd", verbosity, "-u", "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path, "--consumerd-err-sock", consumer_data->err_unix_sock_path, + "--group", tracing_group_name, NULL); if (consumerd32_libdir[0] != '\0') { free(tmpnew); @@ -2091,21 +2329,6 @@ error: return ret; } -/* - * Compute health status of each consumer. If one of them is zero (bad - * state), we return 0. - */ -static int check_consumer_health(void) -{ - int ret; - - ret = health_check_state(HEALTH_TYPE_CONSUMER); - - DBG3("Health consumer check %d", ret); - - return ret; -} - /* * Setup necessary data for kernel tracer action. */ @@ -2204,6 +2427,7 @@ static int copy_session_consumer(int domain, struct ltt_session *session) consumer = session->kernel_session->consumer; dir_name = DEFAULT_KERNEL_TRACE_DIR; break; + case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_UST: DBG3("Copying tracing session consumer output in UST session"); if (session->ust_session->consumer) { @@ -2247,6 +2471,7 @@ static int create_ust_session(struct ltt_session *session, assert(session->consumer); switch (domain->type) { + case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_UST: break; default: @@ -2265,6 +2490,9 @@ static int create_ust_session(struct ltt_session *session, lus->uid = session->uid; lus->gid = session->gid; + lus->output_traces = session->output_traces; + lus->snapshot_mode = session->snapshot_mode; + lus->live_timer_interval = session->live_timer; session->ust_session = lus; /* Copy session output to the newly created UST session */ @@ -2321,6 +2549,8 @@ 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; + session->kernel_session->snapshot_mode = session->snapshot_mode; return LTTNG_OK; @@ -2376,12 +2606,18 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: + case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_LIVE: case LTTNG_DESTROY_SESSION: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_DOMAINS: 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: @@ -2434,6 +2670,8 @@ 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_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: @@ -2450,12 +2688,7 @@ 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 */ @@ -2530,8 +2763,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, } break; + case LTTNG_DOMAIN_JUL: 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; @@ -2617,6 +2855,7 @@ skip_domain: if (cmd_ctx->lsm->cmd_type == LTTNG_START_TRACE || cmd_ctx->lsm->cmd_type == LTTNG_STOP_TRACE) { switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_UST: if (uatomic_read(&ust_consumerd_state) != CONSUMER_STARTED) { ret = LTTNG_ERR_NO_USTCONSUMERD; @@ -2700,7 +2939,7 @@ skip_domain: { ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]); + &cmd_ctx->lsm->u.enable.event, NULL, NULL, kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_ALL_EVENT: @@ -2879,7 +3118,7 @@ skip_domain: } ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri, - &cmd_ctx->creds); + &cmd_ctx->creds, 0); free(uris); @@ -3061,7 +3300,7 @@ skip_domain: ret = cmd_enable_event(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]); + &cmd_ctx->lsm->u.enable.event, bytecode, NULL, kernel_poll_pipe[1]); break; } case LTTNG_DATA_PENDING: @@ -3069,6 +3308,145 @@ 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; + } + case LTTNG_CREATE_SESSION_SNAPSHOT: + { + 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_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 = 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); + break; + } default: ret = LTTNG_ERR_UND; break; @@ -3102,8 +3480,8 @@ static void *thread_manage_health(void *data) int sock = -1, new_sock = -1, ret, i, pollfd, err = -1; uint32_t revents, nb_fd; struct lttng_poll_event events; - struct lttcomm_health_msg msg; - struct lttcomm_health_data reply; + struct health_comm_msg msg; + struct health_comm_reply reply; DBG("[thread] Manage health check started"); @@ -3120,6 +3498,27 @@ static void *thread_manage_health(void *data) goto error; } + if (is_root) { + /* lttng health client socket path permissions */ + ret = chown(health_unix_sock_path, 0, + utils_get_group_id(tracing_group_name)); + if (ret < 0) { + ERR("Unable to set group on %s", health_unix_sock_path); + PERROR("chown"); + ret = -1; + goto error; + } + + ret = chmod(health_unix_sock_path, + S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); + if (ret < 0) { + ERR("Unable to set permissions on %s", health_unix_sock_path); + PERROR("chmod"); + ret = -1; + goto error; + } + } + /* * Set the CLOEXEC flag. Return code is useless because either way, the * show must go on. @@ -3210,54 +3609,18 @@ restart: rcu_thread_online(); - switch (msg.component) { - case LTTNG_HEALTH_CMD: - reply.ret_code = health_check_state(HEALTH_TYPE_CMD); - break; - case LTTNG_HEALTH_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_TYPE_APP_REG); - break; - case LTTNG_HEALTH_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_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_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() && - health_check_state(HEALTH_TYPE_HT_CLEANUP) && - health_check_state(HEALTH_TYPE_APP_MANAGE_NOTIFY); - break; - default: - reply.ret_code = LTTNG_ERR_UND; - break; - } - - /* - * Flip ret value since 0 is a success and 1 indicates a bad health for - * the client where in the sessiond it is the opposite. Again, this is - * just to make things easier for us poor developer which enjoy a lot - * lazyness. - */ - if (reply.ret_code == 0 || reply.ret_code == 1) { - reply.ret_code = !reply.ret_code; + reply.ret_code = 0; + for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) { + /* + * health_check_state returns 0 if health is + * bad. + */ + if (!health_check_state(health_sessiond, i)) { + reply.ret_code |= 1ULL << i; + } } - DBG2("Health check return value %d", reply.ret_code); + DBG2("Health check return value %" PRIx64, reply.ret_code); ret = send_unix_sock(new_sock, (void *) &reply, sizeof(reply)); if (ret < 0) { @@ -3308,7 +3671,7 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); - health_register(HEALTH_TYPE_CMD); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD); if (testpoint(thread_manage_clients)) { goto error_testpoint; @@ -3338,11 +3701,17 @@ static void *thread_manage_clients(void *data) /* * Notify parent pid that we are ready to accept command for client side. + * This ppid is the one from the external process that spawned us. */ if (opt_sig_parent) { kill(ppid, SIGUSR1); } + /* Notify the parent of the fork() process that we are ready. */ + if (opt_daemon) { + kill(child_ppid, SIGUSR1); + } + if (testpoint(thread_manage_clients_before_loop)) { goto error; } @@ -3532,7 +3901,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Client thread dying"); @@ -3563,12 +3932,13 @@ static void usage(void) fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); fprintf(stderr, " -V, --version Show version number.\n"); - fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n"); + fprintf(stderr, " -S, --sig-parent Send SIGUSR1 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"); + fprintf(stderr, " --jul-tcp-port JUL application registration TCP port\n"); } /* @@ -3601,12 +3971,13 @@ static int parse_args(int argc, char **argv) { "verbose-consumer", 0, 0, 'Z' }, { "no-kernel", 0, 0, 'N' }, { "pidfile", 1, 0, 'p' }, + { "jul-tcp-port", 1, 0, 'J' }, { 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:p:", + c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:J:", long_options, &option_index); if (c == -1) { break; @@ -3629,7 +4000,7 @@ static int parse_args(int argc, char **argv) opt_daemon = 1; break; case 'g': - opt_tracing_group = optarg; + tracing_group_name = optarg; break; case 'h': usage(); @@ -3686,6 +4057,24 @@ static int parse_args(int argc, char **argv) case 'p': opt_pidfile = optarg; break; + case 'J': /* JUL TCP port. */ + { + unsigned long v; + + errno = 0; + v = strtoul(optarg, NULL, 0); + if (errno != 0 || !isdigit(optarg[0])) { + ERR("Wrong value in --jul-tcp-port parameter: %s", optarg); + return -1; + } + if (v == 0 || v >= 65535) { + ERR("Port overflow in --jul-tcp-port parameter: %s", optarg); + return -1; + } + jul_tcp_port = (uint32_t) v; + DBG3("JUL TCP port set to non default: %u", jul_tcp_port); + break; + } default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -3790,14 +4179,7 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - ret = allowed_group(); - if (ret < 0) { - WARN("No tracing group detected"); - ret = 0; - goto end; - } - - gid = ret; + gid = utils_get_group_id(tracing_group_name); /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -3806,8 +4188,12 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure tracing group can search the run dir */ - ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH); + /* + * Ensure all applications and tracing group can search the run + * dir. Allow everyone to read the directory, since it does not + * buy us anything to hide its content. + */ + ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH); if (ret < 0) { ERR("Unable to set permissions on %s", rundir); PERROR("chmod"); @@ -3821,21 +4207,21 @@ static int set_permissions(char *rundir) } /* kconsumer error socket path */ - ret = chown(kconsumer_data.err_unix_sock_path, 0, gid); + ret = chown(kconsumer_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path); PERROR("chown"); } /* 64-bit ustconsumer error socket path */ - ret = chown(ustconsumer64_data.err_unix_sock_path, 0, gid); + ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path); PERROR("chown"); } /* 32-bit ustconsumer compat32 error socket path */ - ret = chown(ustconsumer32_data.err_unix_sock_path, 0, gid); + ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0); if (ret < 0) { ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path); PERROR("chown"); @@ -3843,7 +4229,6 @@ static int set_permissions(char *rundir) DBG("All permissions are set"); -end: return ret; } @@ -3880,7 +4265,7 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, int ret; char path[PATH_MAX]; - switch (consumer_data->type) { + switch (consumer_data->type) { case LTTNG_CONSUMER_KERNEL: snprintf(path, PATH_MAX, DEFAULT_KCONSUMERD_PATH, rundir); break; @@ -3898,7 +4283,7 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, DBG2("Creating consumer directory: %s", path); - ret = mkdir(path, S_IRWXU); + ret = mkdir(path, S_IRWXU | S_IRGRP | S_IXGRP); if (ret < 0) { if (errno != EEXIST) { PERROR("mkdir"); @@ -3907,6 +4292,14 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, } ret = -1; } + if (is_root) { + ret = chown(path, 0, utils_get_group_id(tracing_group_name)); + if (ret < 0) { + ERR("Unable to set group on %s", path); + PERROR("chown"); + goto error; + } + } /* Create the kconsumerd error unix socket */ consumer_data->err_sock = @@ -3917,6 +4310,16 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + ret = utils_set_fd_cloexec(consumer_data->err_sock); + if (ret < 0) { + PERROR("utils_set_fd_cloexec"); + /* continue anyway */ + } + /* File permission MUST be 660 */ ret = chmod(consumer_data->err_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); @@ -3950,6 +4353,9 @@ static void sighandler(int sig) DBG("SIGTERM caught"); stop_threads(); break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); + break; default: break; } @@ -3988,7 +4394,12 @@ static int set_signal_handler(void) return ret; } - DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT"); + if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { + PERROR("sigaction"); + return ret; + } + + DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); return ret; } @@ -4044,6 +4455,123 @@ error: return; } +/* + * Write JUL TCP port using the rundir. + */ +static void write_julport(void) +{ + int ret; + char path[PATH_MAX]; + + assert(rundir); + + ret = snprintf(path, sizeof(path), "%s/" + DEFAULT_LTTNG_SESSIOND_JULPORT_FILE, rundir); + if (ret < 0) { + PERROR("snprintf julport path"); + goto error; + } + + /* + * Create TCP JUL port 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(jul_tcp_port, path); + +error: + return; +} + +/* + * Daemonize this process by forking and making the parent wait for the child + * to signal it indicating readiness. Once received, the parent successfully + * quits. + * + * The child process undergoes the same action that daemon(3) does meaning + * setsid, chdir, and dup /dev/null into 0, 1 and 2. + * + * Return 0 on success else -1 on error. + */ +static int daemonize(void) +{ + int ret; + pid_t pid; + + /* Get parent pid of this process. */ + child_ppid = getppid(); + + pid = fork(); + if (pid < 0) { + PERROR("fork"); + goto error; + } else if (pid == 0) { + int fd; + pid_t sid; + + /* Child */ + + /* + * Get the newly created parent pid so we can signal that process when + * we are ready to operate. + */ + child_ppid = getppid(); + + sid = setsid(); + if (sid < 0) { + PERROR("setsid"); + goto error; + } + + /* Try to change directory to /. If we can't well at least notify. */ + ret = chdir("/"); + if (ret < 0) { + PERROR("chdir"); + } + + fd = open(_PATH_DEVNULL, O_RDWR, 0); + if (fd < 0) { + PERROR("open %s", _PATH_DEVNULL); + /* Let 0, 1 and 2 open since we can't bind them to /dev/null. */ + } else { + (void) dup2(fd, STDIN_FILENO); + (void) dup2(fd, STDOUT_FILENO); + (void) dup2(fd, STDERR_FILENO); + if (fd > 2) { + ret = close(fd); + if (ret < 0) { + PERROR("close"); + } + } + } + goto end; + } else { + /* Parent */ + + /* + * Waiting for child to notify this parent that it can exit. Note that + * sleep() is interrupted before the 1 second delay as soon as the + * signal is received, so it will not cause visible delay for the + * user. + */ + while (!CMM_LOAD_SHARED(recv_child_signal)) { + sleep(1); + } + + /* + * From this point on, the parent can exit and the child is now an + * operationnal session daemon ready to serve clients and applications. + */ + exit(EXIT_SUCCESS); + } + +end: + return 0; + +error: + return -1; +} + /* * main */ @@ -4057,6 +4585,10 @@ int main(int argc, char **argv) rcu_register_thread(); + if ((ret = set_signal_handler()) < 0) { + goto error; + } + setup_consumerd_path(); page_size = sysconf(_SC_PAGESIZE); @@ -4076,20 +4608,15 @@ int main(int argc, char **argv) if (opt_daemon) { int i; - /* - * fork - * child: setsid, close FD 0, 1, 2, chdir / - * parent: exit (if fork is successful) - */ - ret = daemon(0, 0); + ret = daemonize(); if (ret < 0) { - PERROR("daemon"); goto error; } + /* - * We are in the child. Make sure all other file - * descriptors are closed, in case we are called with - * more opened file descriptors than the standard ones. + * We are in the child. Make sure all other file descriptors are + * closed, in case we are called with more opened file descriptors than + * the standard ones. */ for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { (void) close(i); @@ -4145,7 +4672,7 @@ 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."); @@ -4241,6 +4768,12 @@ int main(int argc, char **argv) */ ust_app_ht_alloc(); + /* Initialize JUL domain subsystem. */ + if ((ret = jul_init()) < 0) { + /* ENOMEM at this point. */ + goto error; + } + /* After this point, we can safely call cleanup() with "goto exit" */ /* @@ -4276,10 +4809,6 @@ int main(int argc, char **argv) goto exit; } - if ((ret = set_signal_handler()) < 0) { - goto exit; - } - /* Setup the needed unix socket */ if ((ret = init_daemon_socket()) < 0) { goto exit; @@ -4344,8 +4873,24 @@ int main(int argc, char **argv) } write_pidfile(); + write_julport(); - /* Create thread to manage the client socket */ + /* 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_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES); + if (!health_sessiond) { + PERROR("health_app_create error"); + goto exit_health_sessiond_cleanup; + } + + /* Create thread to clean up RCU hash tables */ ret = pthread_create(&ht_cleanup_thread, NULL, thread_ht_cleanup, (void *) NULL); if (ret != 0) { @@ -4353,7 +4898,7 @@ int main(int argc, char **argv) goto exit_ht_cleanup; } - /* Create thread to manage the client socket */ + /* Create health-check thread */ ret = pthread_create(&health_thread, NULL, thread_manage_health, (void *) NULL); if (ret != 0) { @@ -4398,7 +4943,15 @@ int main(int argc, char **argv) ust_thread_manage_notify, (void *) NULL); if (ret != 0) { PERROR("pthread_create apps"); - goto exit_apps; + goto exit_apps_notify; + } + + /* Create JUL registration thread. */ + ret = pthread_create(&jul_reg_thread, NULL, + jul_thread_manage_registration, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create apps"); + goto exit_jul_reg; } /* Don't start this thread if kernel tracing is not requested nor root */ @@ -4419,12 +4972,27 @@ int main(int argc, char **argv) } exit_kernel: + ret = pthread_join(jul_reg_thread, &status); + if (ret != 0) { + PERROR("pthread_join JUL"); + goto error; /* join error, exit without cleanup */ + } + +exit_jul_reg: + ret = pthread_join(apps_notify_thread, &status); + if (ret != 0) { + PERROR("pthread_join apps notify"); + goto error; /* join error, exit without cleanup */ + } + +exit_apps_notify: ret = pthread_join(apps_thread, &status); if (ret != 0) { - PERROR("pthread_join"); + PERROR("pthread_join apps"); goto error; /* join error, exit without cleanup */ } + exit_apps: ret = pthread_join(reg_apps_thread, &status); if (ret != 0) { @@ -4478,6 +5046,8 @@ exit_health: goto error; /* join error, exit without cleanup */ } exit_ht_cleanup: + health_app_destroy(health_sessiond); +exit_health_sessiond_cleanup: exit: /* * cleanup() is called when no other thread is running.