X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=9b3828bd08feec5654c6a73cc1fdc3e01abfa672;hb=ec4c7666d3907424f62e874b78e44ce4b9456f41;hp=8aea0e77c840e38f3a05b117064fbc21f1f900e9;hpb=67676bd8b522e83c6d45c7f6a205ab7a30001c9d;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 8aea0e77c..9b3828bd0 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -45,6 +45,7 @@ #include #include #include +#include #include "lttng-sessiond.h" #include "buffer-registry.h" @@ -73,12 +74,12 @@ 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; @@ -243,6 +244,41 @@ 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) { @@ -423,6 +459,27 @@ 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 */ @@ -504,14 +561,6 @@ static void cleanup(void) 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. - */ - - free(rundir); - DBG("Cleaning up all sessions"); /* Destroy session list mutex */ @@ -543,6 +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", @@ -801,13 +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; } @@ -836,12 +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) { /* @@ -867,6 +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) { @@ -950,7 +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; @@ -987,7 +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; } @@ -1012,6 +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) { @@ -1051,7 +1141,6 @@ restart: } health_code_update(); - if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = @@ -1109,6 +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(); @@ -1131,13 +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)) { @@ -1165,11 +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(); } @@ -1208,13 +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) { @@ -1228,9 +1328,10 @@ 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) { @@ -1260,7 +1361,7 @@ static void *thread_manage_apps(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE); - if (testpoint(thread_manage_apps)) { + if (testpoint(sessiond_thread_manage_apps)) { goto error_testpoint; } @@ -1276,19 +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) { /* @@ -1309,6 +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) { @@ -1345,10 +1453,6 @@ static void *thread_manage_apps(void *data) } DBG("Apps with sock %d added to poll set", sock); - - health_code_update(); - - break; } } else { /* @@ -1364,7 +1468,6 @@ static void *thread_manage_apps(void *data) /* Socket closed on remote end. */ ust_app_unregister(pollfd); - break; } } @@ -1492,6 +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 && @@ -1536,6 +1644,10 @@ static void *thread_dispatch_ust_registration(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH); + if (testpoint(sessiond_thread_app_reg_dispatch)) { + goto error_testpoint; + } + health_code_update(); CDS_INIT_LIST_HEAD(&wait_queue.head); @@ -1740,6 +1852,7 @@ error: free(wait_node); } +error_testpoint: DBG("Dispatch thread dying"); if (err) { health_error(); @@ -1767,7 +1880,7 @@ static void *thread_registration_apps(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG); - if (testpoint(thread_registration_apps)) { + if (testpoint(sessiond_thread_registration_apps)) { goto error_testpoint; } @@ -1826,6 +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) { @@ -1928,11 +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); @@ -1957,6 +2070,10 @@ error_listen: error_create_poll: error_testpoint: DBG("UST Registration thread cleanup complete"); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } health_unregister(health_sessiond); return NULL; @@ -2053,19 +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; } @@ -2151,10 +2272,11 @@ 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, @@ -2202,9 +2324,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) if (consumerd64_libdir[0] != '\0') { free(tmpnew); } - if (ret) { - goto error; - } break; } case LTTNG_CONSUMER32_UST: @@ -2248,9 +2367,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) if (consumerd32_libdir[0] != '\0') { free(tmpnew); } - if (ret) { - goto error; - } break; } default: @@ -2258,8 +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; @@ -2817,6 +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) { @@ -3021,7 +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; @@ -3052,8 +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; @@ -3227,7 +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); @@ -3570,6 +3692,8 @@ static void *thread_manage_health(void *data) goto error; } + lttng_sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3593,6 +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) { @@ -3634,7 +3763,7 @@ restart: rcu_thread_online(); - reply.ret_code = 0; + memset(&reply, 0, sizeof(reply)); for (i = 0; i < NR_HEALTH_SESSIOND_TYPES; i++) { /* * health_check_state returns 0 if health is @@ -3698,10 +3827,6 @@ static void *thread_manage_clients(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD); - if (testpoint(thread_manage_clients)) { - goto error_testpoint; - } - health_code_update(); ret = lttcomm_listen_unix_sock(client_sock); @@ -3724,20 +3849,14 @@ static void *thread_manage_clients(void *data) goto error; } - /* - * 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); - } + lttng_sessiond_notify_ready(); - /* Notify the parent of the fork() process that we are ready. */ - if (opt_daemon) { - kill(child_ppid, SIGUSR1); + /* 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; } @@ -3770,6 +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) { @@ -3912,7 +4036,6 @@ error: error_listen: error_create_poll: -error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3955,6 +4078,7 @@ 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 SIGUSR1 to parent pid to notify readiness.\n"); @@ -3997,12 +4121,13 @@ static int parse_args(int argc, char **argv) { "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:J:", + 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; @@ -4024,6 +4149,9 @@ static int parse_args(int argc, char **argv) case 'd': opt_daemon = 1; break; + case 'b': + opt_background = 1; + break; case 'g': tracing_group_name = optarg; break; @@ -4480,6 +4608,24 @@ 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. */ @@ -4508,95 +4654,6 @@ 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 */ @@ -4630,10 +4687,11 @@ int main(int argc, char **argv) } /* Daemonize */ - if (opt_daemon) { + if (opt_daemon || opt_background) { int i; - ret = daemonize(); + ret = lttng_daemonize(&child_ppid, &recv_child_signal, + !opt_background); if (ret < 0) { goto error; } @@ -4658,6 +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); @@ -4744,6 +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; @@ -4967,7 +5034,7 @@ 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"); + PERROR("pthread_create notify"); goto exit_apps_notify; } @@ -4975,7 +5042,7 @@ int main(int argc, char **argv) ret = pthread_create(&jul_reg_thread, NULL, jul_thread_manage_registration, (void *) NULL); if (ret != 0) { - PERROR("pthread_create apps"); + PERROR("pthread_create JUL"); goto exit_jul_reg; }