X-Git-Url: https://git.lttng.org/?a=blobdiff_plain;ds=sidebyside;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=0bb06a3efe5aedd335f5a43238300aea7a5d62ba;hb=24a54a9b81a627c91b4ca488ff5d42d00627a3ba;hp=d6bca47d59db9bdc8b116a1f2df3dd6493e4979f;hpb=71ddb4eda3ff99b05c3661b23e4b76c17444b53e;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index d6bca47d5..0bb06a3ef 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" @@ -73,7 +74,7 @@ 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 */ @@ -243,6 +244,38 @@ struct health_app *health_sessiond; /* JUL TCP port for registration. Used by the JUL thread. */ unsigned int jul_tcp_port = DEFAULT_JUL_TCP_PORT; +/* + * 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) { @@ -1051,7 +1084,6 @@ restart: } health_code_update(); - if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = @@ -1208,13 +1240,13 @@ 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 +1260,10 @@ 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) { @@ -1531,6 +1564,10 @@ 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); @@ -1735,6 +1772,7 @@ error: free(wait_node); } +error_testpoint: DBG("Dispatch thread dying"); if (err) { health_error(); @@ -1923,11 +1961,6 @@ 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); @@ -1952,6 +1985,10 @@ 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; @@ -2048,19 +2085,23 @@ 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; } @@ -2146,10 +2187,11 @@ 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, @@ -2197,9 +2239,6 @@ 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: @@ -2243,9 +2282,6 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) if (consumerd32_libdir[0] != '\0') { free(tmpnew); } - if (ret) { - goto error; - } break; } default: @@ -2253,8 +2289,9 @@ 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; @@ -3565,6 +3602,8 @@ static void *thread_manage_health(void *data) goto error; } + lttng_sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3629,7 +3668,7 @@ 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 @@ -3693,10 +3732,6 @@ static void *thread_manage_clients(void *data) health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CMD); - if (testpoint(sessiond_thread_manage_clients)) { - goto error_testpoint; - } - health_code_update(); ret = lttcomm_listen_unix_sock(client_sock); @@ -3719,17 +3754,11 @@ 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(sessiond_thread_manage_clients_before_loop)) { @@ -3907,7 +3936,6 @@ error: error_listen: error_create_poll: -error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3950,6 +3978,7 @@ 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"); @@ -3992,12 +4021,13 @@ 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; @@ -4019,6 +4049,9 @@ 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; @@ -4503,107 +4536,6 @@ 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)) { - int status; - pid_t ret; - - /* - * Check if child exists without blocking. If so, we have to stop - * this parent process and return an error. - */ - ret = waitpid(pid, &status, WNOHANG); - if (ret < 0 || (ret != 0 && WIFEXITED(status))) { - /* The child exited somehow or was not valid. */ - goto error; - } - 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 */ @@ -4637,10 +4569,11 @@ 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; }