X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=9b3828bd08feec5654c6a73cc1fdc3e01abfa672;hb=ec4c7666d3907424f62e874b78e44ce4b9456f41;hp=f3b007a4fac7b1fc3cf17ad1927c1264c375c1d0;hpb=67e056446003680180ed4f384217facd90b1a776;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index f3b007a4f..9b3828bd0 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 @@ -44,6 +45,7 @@ #include #include #include +#include #include "lttng-sessiond.h" #include "buffer-registry.h" @@ -60,25 +62,27 @@ #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; -static int opt_daemon; +static int opt_daemon, opt_background; 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; +static int lockfile_fd = -1; + +/* 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 @@ -159,6 +163,7 @@ 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 @@ -233,6 +238,47 @@ 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; + +/* Am I root or not. */ +int is_root; /* Set to 1 if the daemon is running as root */ + +/* + * Whether sessiond is ready for commands/health check requests. + * NR_LTTNG_SESSIOND_READY must match the number of calls to + * lttng_sessiond_notify_ready(). + */ +#define NR_LTTNG_SESSIOND_READY 2 +int lttng_sessiond_ready = NR_LTTNG_SESSIOND_READY; + +/* Notify parents that we are ready for cmd and health check */ +static +void lttng_sessiond_notify_ready(void) +{ + if (uatomic_sub_return(<tng_sessiond_ready, 1) == 0) { + /* + * 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 || opt_background) { + kill(child_ppid, SIGUSR1); + } + } +} + static void setup_consumerd_path(void) { @@ -322,25 +368,6 @@ 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. * @@ -432,14 +459,35 @@ static void close_consumer_sockets(void) } } +/* + * Generate the full lock file path using the rundir. + * + * Return the snprintf() return value thus a negative value is an error. + */ +static int generate_lock_file_path(char *path, size_t len) +{ + int ret; + + assert(path); + assert(rundir); + + /* Build lockfile path from rundir. */ + ret = snprintf(path, len, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE, rundir); + if (ret < 0) { + PERROR("snprintf lockfile path"); + } + + return ret; +} + /* * Cleanup the daemon */ static void cleanup(void) { int ret; - char *cmd = NULL; struct ltt_session *sess, *stmp; + char path[PATH_MAX]; DBG("Cleaning up"); @@ -460,19 +508,58 @@ 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!"); - } - - /* Remove lttng run directory */ - ret = system(cmd); - if (ret < 0) { - ERR("Unable to clean %s", rundir); - } - free(cmd); - free(rundir); + 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); DBG("Cleaning up all sessions"); @@ -505,6 +592,35 @@ static void cleanup(void) close_consumer_sockets(); + + /* + * Cleanup lock file by deleting it and finaly closing it which will + * release the file system lock. + */ + if (lockfile_fd >= 0) { + char lockfile_path[PATH_MAX]; + + ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path)); + if (ret > 0) { + ret = remove(lockfile_path); + if (ret < 0) { + PERROR("remove lock file"); + } + ret = close(lockfile_fd); + if (ret < 0) { + PERROR("close lock file"); + } + } + } + + /* + * We do NOT rmdir rundir because there are other processes + * using it, for instance lttng-relayd, which can start in + * parallel with this teardown. + */ + + free(rundir); + /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", @@ -755,7 +871,7 @@ 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 @@ -763,13 +879,13 @@ static void *thread_manage_kernel(void *data) */ lttng_poll_init(&events); - if (testpoint(thread_manage_kernel)) { + if (testpoint(sessiond_thread_manage_kernel)) { goto error_testpoint; } health_code_update(); - if (testpoint(thread_manage_kernel_before_loop)) { + if (testpoint(sessiond_thread_manage_kernel_before_loop)) { goto error_testpoint; } @@ -798,12 +914,14 @@ static void *thread_manage_kernel(void *data) update_poll_flag = 0; } - DBG("Thread kernel polling on %d fds", LTTNG_POLL_GETNB(&events)); + DBG("Thread kernel polling"); /* Poll infinite value of time */ restart: health_poll_entry(); ret = lttng_poll_wait(&events, -1); + DBG("Thread kernel return from poll on %d fds", + LTTNG_POLL_GETNB(&events)); health_poll_exit(); if (ret < 0) { /* @@ -829,6 +947,11 @@ static void *thread_manage_kernel(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -838,9 +961,8 @@ 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. @@ -880,7 +1002,7 @@ 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; } @@ -913,7 +1035,7 @@ static void signal_consumer_condition(struct consumer_data *data, int state) */ static void *thread_manage_consumer(void *data) { - int sock = -1, i, ret, pollfd, err = -1; + int sock = -1, i, ret, pollfd, err = -1, should_quit = 0; uint32_t revents, nb_fd; enum lttcomm_return_code code; struct lttng_poll_event events; @@ -921,7 +1043,7 @@ 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(); @@ -950,7 +1072,7 @@ static void *thread_manage_consumer(void *data) restart: health_poll_entry(); - if (testpoint(thread_manage_consumer)) { + if (testpoint(sessiond_thread_manage_consumer)) { goto error; } @@ -975,6 +1097,11 @@ restart: health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1014,7 +1141,6 @@ restart: } health_code_update(); - if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = @@ -1072,6 +1198,15 @@ restart: /* Infinite blocking call, waiting for transmission */ restart_poll: while (1) { + health_code_update(); + + /* Exit the thread because the thread quit pipe has been triggered. */ + if (should_quit) { + /* Not a health error. */ + err = 0; + goto exit; + } + health_poll_entry(); ret = lttng_poll_wait(&events, -1); health_poll_exit(); @@ -1094,13 +1229,18 @@ restart_poll: health_code_update(); - /* Thread quit pipe has been closed. Killing thread. */ - ret = sessiond_check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; } + /* + * Thread quit pipe has been triggered, flag that we should stop + * but continue the current loop to handle potential data from + * consumer. + */ + should_quit = sessiond_check_thread_quit_pipe(pollfd, revents); + if (pollfd == sock) { /* Event on the consumerd socket */ if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { @@ -1128,11 +1268,8 @@ restart_poll: ERR("Handling metadata request"); goto error; } - break; - } else { - ERR("Unknown pollfd"); - goto error; } + /* No need for an else branch all FDs are tested prior. */ } health_code_update(); } @@ -1171,13 +1308,13 @@ error: } consumer_data->cmd_sock = -1; } - if (*consumer_data->metadata_sock.fd_ptr >= 0) { + if (consumer_data->metadata_sock.fd_ptr && + *consumer_data->metadata_sock.fd_ptr >= 0) { ret = close(*consumer_data->metadata_sock.fd_ptr); if (ret) { PERROR("close"); } } - if (sock >= 0) { ret = close(sock); if (ret) { @@ -1191,16 +1328,17 @@ error: pthread_mutex_unlock(&consumer_data->lock); /* Cleanup metadata socket mutex. */ - pthread_mutex_destroy(consumer_data->metadata_sock.lock); - free(consumer_data->metadata_sock.lock); - + if (consumer_data->metadata_sock.lock) { + pthread_mutex_destroy(consumer_data->metadata_sock.lock); + free(consumer_data->metadata_sock.lock); + } lttng_poll_clean(&events); error_poll: if (err) { health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("consumer thread cleanup completed"); return NULL; @@ -1212,6 +1350,7 @@ 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; @@ -1220,9 +1359,9 @@ 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)) { + if (testpoint(sessiond_thread_manage_apps)) { goto error_testpoint; } @@ -1238,19 +1377,21 @@ static void *thread_manage_apps(void *data) goto error; } - if (testpoint(thread_manage_apps_before_loop)) { + if (testpoint(sessiond_thread_manage_apps_before_loop)) { goto error; } health_code_update(); while (1) { - DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events)); + DBG("Apps thread polling"); /* Inifinite blocking call, waiting for transmission */ restart: health_poll_entry(); ret = lttng_poll_wait(&events, -1); + DBG("Apps thread return from poll on %d fds", + LTTNG_POLL_GETNB(&events)); health_poll_exit(); if (ret < 0) { /* @@ -1271,6 +1412,11 @@ static void *thread_manage_apps(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1287,10 +1433,8 @@ 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; } @@ -1308,23 +1452,7 @@ static void *thread_manage_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); - DBG("Apps with sock %d added to poll set", sock); - - health_code_update(); - - break; } } else { /* @@ -1340,7 +1468,6 @@ static void *thread_manage_apps(void *data) /* Socket closed on remote end. */ ust_app_unregister(pollfd); - break; } } @@ -1366,7 +1493,7 @@ 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(); @@ -1377,21 +1504,27 @@ 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; @@ -1402,7 +1535,7 @@ 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; } /* @@ -1462,6 +1595,11 @@ static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) uint32_t revents = LTTNG_POLL_GETEV(&events, i); int pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, &wait_queue->head, head) { if (pollfd == wait_node->app->sock && @@ -1504,7 +1642,11 @@ static void *thread_dispatch_ust_registration(void *data) .count = 0, }; - health_register(HEALTH_TYPE_APP_REG_DISPATCH); + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); + + if (testpoint(sessiond_thread_app_reg_dispatch)) { + goto error_testpoint; + } health_code_update(); @@ -1555,7 +1697,7 @@ 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; } @@ -1569,7 +1711,7 @@ 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; @@ -1617,7 +1759,7 @@ 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); } @@ -1649,7 +1791,12 @@ 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; } @@ -1674,7 +1821,12 @@ 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; } @@ -1700,12 +1852,13 @@ error: free(wait_node); } +error_testpoint: DBG("Dispatch thread dying"); if (err) { health_error(); ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); return NULL; } @@ -1725,9 +1878,9 @@ 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)) { + if (testpoint(sessiond_thread_registration_apps)) { goto error_testpoint; } @@ -1786,6 +1939,11 @@ static void *thread_registration_apps(void *data) revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -1804,6 +1962,18 @@ 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. @@ -1876,11 +2046,6 @@ static void *thread_registration_apps(void *data) exit: error: - if (err) { - health_error(); - ERR("Health error occurred in %s", __func__); - } - /* Notify that the registration thread is gone */ notify_ust_apps(0); @@ -1905,7 +2070,11 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); - health_unregister(); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_sessiond); return NULL; } @@ -2001,19 +2170,23 @@ static int spawn_consumer_thread(struct consumer_data *consumer_data) 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"); - ret = pthread_cancel(consumer_data->thread); - if (ret < 0) { + 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; } @@ -2099,13 +2272,15 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) consumer_to_use = consumerd32_bin; } else { DBG("Could not find any valid consumerd executable"); + ret = -EINVAL; break; } DBG("Using kernel consumer at: %s", consumer_to_use); - execl(consumer_to_use, + ret = execl(consumer_to_use, "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: @@ -2144,13 +2319,11 @@ 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); } - if (ret) { - goto error; - } break; } case LTTNG_CONSUMER32_UST: @@ -2189,13 +2362,11 @@ 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); } - if (ret) { - goto error; - } break; } default: @@ -2203,8 +2374,9 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) exit(EXIT_FAILURE); } if (errno != 0) { - PERROR("kernel start consumer exec"); + PERROR("Consumer execl()"); } + /* Reaching this point, we got a failure on our execl(). */ exit(EXIT_FAILURE); } else if (pid > 0) { ret = pid; @@ -2274,21 +2446,6 @@ 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. */ @@ -2387,6 +2544,7 @@ 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) { @@ -2430,6 +2588,7 @@ 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: @@ -2450,6 +2609,7 @@ static int create_ust_session(struct ltt_session *session, 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 */ @@ -2564,6 +2724,7 @@ 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: @@ -2627,6 +2788,7 @@ 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_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: @@ -2718,6 +2880,7 @@ 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()) { @@ -2771,6 +2934,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, } /* 32-bit */ + pthread_mutex_lock(&ustconsumer32_data.pid_mutex); if (consumerd32_bin[0] != '\0' && ustconsumer32_data.pid == 0 && cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { @@ -2809,6 +2973,7 @@ 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; @@ -2890,9 +3055,74 @@ skip_domain: } case LTTNG_ENABLE_EVENT: { + struct lttng_event_exclusion *exclusion = NULL; + struct lttng_filter_bytecode *bytecode = NULL; + + /* Handle exclusion events and receive it from the client. */ + if (cmd_ctx->lsm->u.enable.exclusion_count > 0) { + size_t count = cmd_ctx->lsm->u.enable.exclusion_count; + + exclusion = zmalloc(sizeof(struct lttng_event_exclusion) + + (count * LTTNG_SYMBOL_NAME_LEN)); + if (!exclusion) { + ret = LTTNG_ERR_EXCLUSION_NOMEM; + goto error; + } + + DBG("Receiving var len exclusion event list from client ..."); + exclusion->count = count; + ret = lttcomm_recv_unix_sock(sock, exclusion->names, + count * LTTNG_SYMBOL_NAME_LEN); + if (ret <= 0) { + DBG("Nothing recv() from client var len data... continuing"); + *sock_error = 1; + free(exclusion); + ret = LTTNG_ERR_EXCLUSION_INVAL; + goto error; + } + } + + /* Handle filter and get bytecode from client. */ + if (cmd_ctx->lsm->u.enable.bytecode_len > 0) { + size_t bytecode_len = cmd_ctx->lsm->u.enable.bytecode_len; + + if (bytecode_len > LTTNG_FILTER_MAX_LEN) { + ret = LTTNG_ERR_FILTER_INVAL; + free(exclusion); + goto error; + } + + bytecode = zmalloc(bytecode_len); + if (!bytecode) { + free(exclusion); + ret = LTTNG_ERR_FILTER_NOMEM; + goto error; + } + + /* Receive var. len. data */ + DBG("Receiving var len filter's bytecode from client ..."); + ret = lttcomm_recv_unix_sock(sock, bytecode, bytecode_len); + if (ret <= 0) { + DBG("Nothing recv() from client car len data... continuing"); + *sock_error = 1; + free(bytecode); + free(exclusion); + ret = LTTNG_ERR_FILTER_INVAL; + goto error; + } + + if ((bytecode->len + sizeof(*bytecode)) != bytecode_len) { + free(bytecode); + free(exclusion); + ret = LTTNG_ERR_FILTER_INVAL; + goto error; + } + } + 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, bytecode, exclusion, + kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_ALL_EVENT: @@ -2909,7 +3139,9 @@ skip_domain: struct lttng_event *events; ssize_t nb_events; + session_lock_list(); nb_events = cmd_list_tracepoints(cmd_ctx->lsm->domain.type, &events); + session_unlock_list(); if (nb_events < 0) { /* Return value is a negative lttng_error_code. */ ret = -nb_events; @@ -2940,8 +3172,10 @@ skip_domain: struct lttng_event_field *fields; ssize_t nb_fields; + session_lock_list(); nb_fields = cmd_list_tracepoint_fields(cmd_ctx->lsm->domain.type, &fields); + session_unlock_list(); if (nb_fields < 0) { /* Return value is a negative lttng_error_code. */ ret = -nb_fields; @@ -3071,7 +3305,7 @@ skip_domain: } ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri, - &cmd_ctx->creds); + &cmd_ctx->creds, 0); free(uris); @@ -3115,7 +3349,7 @@ skip_domain: case LTTNG_LIST_CHANNELS: { int nb_chan; - struct lttng_channel *channels; + struct lttng_channel *channels = NULL; nb_chan = cmd_list_channels(cmd_ctx->lsm->domain.type, cmd_ctx->session, &channels); @@ -3216,46 +3450,6 @@ skip_domain: cmd_ctx->lsm->u.reg.path, cdata); break; } - case LTTNG_ENABLE_EVENT_WITH_FILTER: - { - struct lttng_filter_bytecode *bytecode; - - if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) { - ret = LTTNG_ERR_FILTER_INVAL; - goto error; - } - if (cmd_ctx->lsm->u.enable.bytecode_len == 0) { - ret = LTTNG_ERR_FILTER_INVAL; - goto error; - } - bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len); - if (!bytecode) { - ret = LTTNG_ERR_FILTER_NOMEM; - goto error; - } - /* Receive var. len. data */ - DBG("Receiving var len data from client ..."); - ret = lttcomm_recv_unix_sock(sock, bytecode, - cmd_ctx->lsm->u.enable.bytecode_len); - if (ret <= 0) { - DBG("Nothing recv() from client var len data... continuing"); - *sock_error = 1; - ret = LTTNG_ERR_FILTER_INVAL; - goto error; - } - - if (bytecode->len + sizeof(*bytecode) - != cmd_ctx->lsm->u.enable.bytecode_len) { - free(bytecode); - ret = LTTNG_ERR_FILTER_INVAL; - goto error; - } - - 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]); - break; - } case LTTNG_DATA_PENDING: { ret = cmd_data_pending(cmd_ctx->session); @@ -3361,6 +3555,45 @@ skip_domain: 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; @@ -3394,8 +3627,8 @@ 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"); @@ -3412,6 +3645,27 @@ 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. @@ -3438,6 +3692,8 @@ static void *thread_manage_health(void *data) goto error; } + lttng_sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3461,6 +3717,11 @@ restart: revents = LTTNG_POLL_GETEV(&events, i); pollfd = LTTNG_POLL_GETFD(&events, i); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -3502,58 +3763,18 @@ 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_APP_REG_DISPATCH: - reply.ret_code = health_check_state(HEALTH_TYPE_APP_REG_DISPATCH); - 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) && - health_check_state(HEALTH_TYPE_APP_REG_DISPATCH); - 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; + memset(&reply, 0, sizeof(reply)); + 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) { @@ -3604,11 +3825,7 @@ static void *thread_manage_clients(void *data) rcu_register_thread(); - health_register(HEALTH_TYPE_CMD); - - if (testpoint(thread_manage_clients)) { - goto error_testpoint; - } + health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD); health_code_update(); @@ -3632,14 +3849,14 @@ static void *thread_manage_clients(void *data) goto error; } - /* - * Notify parent pid that we are ready to accept command for client side. - */ - if (opt_sig_parent) { - kill(ppid, SIGUSR1); + lttng_sessiond_notify_ready(); + + /* This testpoint is after we signal readiness to the parent. */ + if (testpoint(sessiond_thread_manage_clients)) { + goto error; } - if (testpoint(thread_manage_clients_before_loop)) { + if (testpoint(sessiond_thread_manage_clients_before_loop)) { goto error; } @@ -3672,6 +3889,11 @@ static void *thread_manage_clients(void *data) health_code_update(); + if (!revents) { + /* No activity for this FD (poll implementation). */ + continue; + } + /* Thread quit pipe has been closed. Killing thread. */ ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { @@ -3814,7 +4036,6 @@ error: error_listen: error_create_poll: -error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3828,7 +4049,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Client thread dying"); @@ -3857,14 +4078,16 @@ static void usage(void) fprintf(stderr, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n"); fprintf(stderr, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n"); fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); + fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\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"); } /* @@ -3897,12 +4120,14 @@ 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' }, + { "background", 0, 0, 'b' }, { 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:b", long_options, &option_index); if (c == -1) { break; @@ -3924,8 +4149,11 @@ static int parse_args(int argc, char **argv) case 'd': opt_daemon = 1; break; + case 'b': + opt_background = 1; + break; case 'g': - opt_tracing_group = optarg; + tracing_group_name = optarg; break; case 'h': usage(); @@ -3982,6 +4210,24 @@ 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 */ @@ -4086,14 +4332,7 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - ret = allowed_group(); - if (ret < 0) { - WARN("No tracing group detected"); - /* Setting gid to 0 if no tracing group is found */ - gid = 0; - } else { - gid = ret; - } + gid = utils_get_group_id(tracing_group_name); /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -4102,8 +4341,12 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure all applications and 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"); @@ -4117,21 +4360,21 @@ 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"); @@ -4175,7 +4418,7 @@ 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; @@ -4193,7 +4436,7 @@ 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"); @@ -4202,6 +4445,14 @@ 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 = @@ -4255,6 +4506,9 @@ static void sighandler(int sig) DBG("SIGTERM caught"); stop_threads(); break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); + break; default: break; } @@ -4293,7 +4547,12 @@ 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; } @@ -4349,6 +4608,52 @@ error: return; } +/* + * Create lockfile using the rundir and return its fd. + */ +static int create_lockfile(void) +{ + int ret; + char lockfile_path[PATH_MAX]; + + ret = generate_lock_file_path(lockfile_path, sizeof(lockfile_path)); + if (ret < 0) { + goto error; + } + + ret = utils_create_lock_file(lockfile_path); +error: + return ret; +} + +/* + * 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; +} + /* * main */ @@ -4362,6 +4667,10 @@ 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); @@ -4378,23 +4687,19 @@ int main(int argc, char **argv) } /* Daemonize */ - if (opt_daemon) { + if (opt_daemon || opt_background) { int i; - /* - * fork - * child: setsid, close FD 0, 1, 2, chdir / - * parent: exit (if fork is successful) - */ - ret = daemon(0, 0); + ret = lttng_daemonize(&child_ppid, &recv_child_signal, + !opt_background); 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); @@ -4411,6 +4716,10 @@ int main(int argc, char **argv) if (is_root) { rundir = strdup(DEFAULT_LTTNG_RUNDIR); + if (!rundir) { + ret = -ENOMEM; + goto error; + } /* Create global run dir with root access */ ret = create_lttng_rundir(rundir); @@ -4497,6 +4806,11 @@ int main(int argc, char **argv) } } + lockfile_fd = create_lockfile(); + if (lockfile_fd < 0) { + goto error; + } + /* Set consumer initial state */ kernel_consumerd_state = CONSUMER_STOPPED; ust_consumerd_state = CONSUMER_STOPPED; @@ -4546,6 +4860,12 @@ 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" */ /* @@ -4581,10 +4901,6 @@ 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; @@ -4649,6 +4965,7 @@ int main(int argc, char **argv) } write_pidfile(); + write_julport(); /* Initialize communication library */ lttcomm_init(); @@ -4659,9 +4976,13 @@ int main(int argc, char **argv) * Initialize the health check subsystem. This call should set the * appropriate time values. */ - health_init(); + 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 manage the client socket */ + /* Create thread to clean up RCU hash tables */ ret = pthread_create(&ht_cleanup_thread, NULL, thread_ht_cleanup, (void *) NULL); if (ret != 0) { @@ -4669,7 +4990,7 @@ 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) { @@ -4713,8 +5034,16 @@ int main(int argc, char **argv) ret = pthread_create(&apps_notify_thread, NULL, ust_thread_manage_notify, (void *) NULL); if (ret != 0) { - PERROR("pthread_create apps"); - goto exit_apps; + PERROR("pthread_create notify"); + 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 JUL"); + goto exit_jul_reg; } /* Don't start this thread if kernel tracing is not requested nor root */ @@ -4735,12 +5064,27 @@ 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) { @@ -4794,6 +5138,8 @@ 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.