X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=052e32e0c32ec2fb8bfe2a17e71afa0487ebf868;hp=6c6386d2b3b1107255260110175743fa04ccf9bf;hb=4421f712a81faad99bb5f20369c276803aeba381;hpb=fdadac0853f699c7dc829162955cc8f459b6d2f2 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 6c6386d2b..052e32e0c 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -1,6 +1,7 @@ /* * Copyright (C) 2011 - David Goulet * Mathieu Desnoyers + * 2013 - Jérémie Galarneau * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -20,6 +21,7 @@ #include #include #include +#include #include #include #include @@ -44,6 +46,8 @@ #include #include #include +#include +#include #include "lttng-sessiond.h" #include "buffer-registry.h" @@ -60,26 +64,29 @@ #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" +#include "save.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 *opt_pidfile; +static const char *tracing_group_name = DEFAULT_TRACING_GROUP; +static int tracing_group_name_override; +static 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; +/* 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. @@ -118,6 +125,39 @@ static struct consumer_data ustconsumer32_data = { .cond_mutex = PTHREAD_MUTEX_INITIALIZER, }; +/* Command line options */ +static const struct option long_options[] = { + { "client-sock", 1, 0, 'c' }, + { "apps-sock", 1, 0, 'a' }, + { "kconsumerd-cmd-sock", 1, 0, 'C' }, + { "kconsumerd-err-sock", 1, 0, 'E' }, + { "ustconsumerd32-cmd-sock", 1, 0, 'G' }, + { "ustconsumerd32-err-sock", 1, 0, 'H' }, + { "ustconsumerd64-cmd-sock", 1, 0, 'D' }, + { "ustconsumerd64-err-sock", 1, 0, 'F' }, + { "consumerd32-path", 1, 0, 'u' }, + { "consumerd32-libdir", 1, 0, 'U' }, + { "consumerd64-path", 1, 0, 't' }, + { "consumerd64-libdir", 1, 0, 'T' }, + { "daemonize", 0, 0, 'd' }, + { "background", 0, 0, 'b' }, + { "sig-parent", 0, 0, 'S' }, + { "help", 0, 0, 'h' }, + { "group", 1, 0, 'g' }, + { "version", 0, 0, 'V' }, + { "quiet", 0, 0, 'q' }, + { "verbose", 0, 0, 'v' }, + { "verbose-consumer", 0, 0, 'Z' }, + { "no-kernel", 0, 0, 'N' }, + { "pidfile", 1, 0, 'p' }, + { "jul-tcp-port", 1, 0, 'J' }, + { "config", 1, 0, 'f' }, + { NULL, 0, 0, 0 } +}; + +/* Command line options to ignore from configuration file */ +static const char *config_ignore_options[] = { "help", "version", "config" }; + /* Shared between threads */ static int dispatch_thread_exit; @@ -159,13 +199,16 @@ 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 * wakers / 1 waiter implemented and detailed in futex.c/.h * - * The thread_manage_apps and thread_dispatch_ust_registration interact with - * this queue and the wait/wake scheme. + * The thread_registration_apps and thread_dispatch_ust_registration uses this + * queue along with the wait/wake scheme. The thread_manage_apps receives down + * the line new application socket and monitors it for any I/O error or clean + * close that triggers an unregistration of the application. */ static struct ust_cmd_queue ust_cmd_queue; @@ -188,6 +231,10 @@ static const char *consumerd32_bin = CONFIG_CONSUMERD32_BIN; static const char *consumerd64_bin = CONFIG_CONSUMERD64_BIN; static const char *consumerd32_libdir = CONFIG_CONSUMERD32_LIBDIR; static const char *consumerd64_libdir = CONFIG_CONSUMERD64_LIBDIR; +static int consumerd32_bin_override; +static int consumerd64_bin_override; +static int consumerd32_libdir_override; +static int consumerd64_libdir_override; static const char *module_proc_lttng = "/proc/lttng"; @@ -233,6 +280,49 @@ 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 */ + +const char * const config_section_name = "sessiond"; + +/* + * 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 +412,6 @@ 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. * @@ -438,8 +509,8 @@ static void close_consumer_sockets(void) static void cleanup(void) { int ret; - char *cmd = NULL; struct ltt_session *sess, *stmp; + char path[PATH_MAX]; DBG("Cleaning up"); @@ -460,18 +531,65 @@ 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"); @@ -505,6 +623,30 @@ static void cleanup(void) close_consumer_sockets(); + /* + * If the override option is set, the pointer points to a *non* const thus + * freeing it even though the variable type is set to const. + */ + if (tracing_group_name_override) { + free((void *) tracing_group_name); + } + if (consumerd32_bin_override) { + free((void *) consumerd32_bin); + } + if (consumerd64_bin_override) { + free((void *) consumerd64_bin); + } + if (consumerd32_libdir_override) { + free((void *) consumerd32_libdir); + } + if (consumerd64_libdir_override) { + free((void *) consumerd64_libdir); + } + + if (opt_pidfile) { + free(opt_pidfile); + } + /* */ DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", @@ -690,9 +832,6 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) rcu_read_lock(); cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd); - pthread_mutex_lock(socket->lock); ret = kernel_consumer_send_channel_stream(socket, channel, ksess, @@ -758,7 +897,7 @@ 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 @@ -766,13 +905,13 @@ 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; } @@ -841,9 +980,8 @@ 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. @@ -883,7 +1021,7 @@ 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; } @@ -924,7 +1062,7 @@ 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(); @@ -953,7 +1091,7 @@ static void *thread_manage_consumer(void *data) restart: health_poll_entry(); - if (testpoint(thread_manage_consumer)) { + if (testpoint(sessiond_thread_manage_consumer)) { goto error; } @@ -1017,20 +1155,20 @@ restart: } health_code_update(); - if (code == LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { /* Connect both socket, command and metadata. */ consumer_data->cmd_sock = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); consumer_data->metadata_fd = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); - if (consumer_data->cmd_sock < 0 || consumer_data->metadata_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 = &consumer_data->metadata_fd; + 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) { @@ -1143,8 +1281,8 @@ exit: error: /* * We lock here because we are about to close the sockets and some other - * thread might be using them so wait before we are exclusive which will - * abort all other consumer command by other threads. + * 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); @@ -1173,13 +1311,13 @@ error: } consumer_data->cmd_sock = -1; } - if (*consumer_data->metadata_sock.fd >= 0) { - ret = close(*consumer_data->metadata_sock.fd); + if (consumer_data->metadata_sock.fd_ptr && + *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) { @@ -1191,17 +1329,19 @@ error: 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); + /* Cleanup metadata socket mutex. */ + 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; @@ -1213,6 +1353,7 @@ 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; @@ -1221,9 +1362,9 @@ 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; } @@ -1239,7 +1380,7 @@ 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; } @@ -1288,10 +1429,8 @@ 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; } @@ -1309,23 +1448,7 @@ 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 { /* @@ -1341,7 +1464,6 @@ static void *thread_manage_apps(void *data) /* Socket closed on remote end. */ ust_app_unregister(pollfd); - break; } } @@ -1367,7 +1489,7 @@ 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(); @@ -1378,21 +1500,27 @@ 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; @@ -1403,7 +1531,7 @@ 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; } /* @@ -1505,7 +1633,11 @@ 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(); @@ -1556,7 +1688,7 @@ 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; } @@ -1570,7 +1702,7 @@ 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; @@ -1618,7 +1750,7 @@ 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); } @@ -1650,7 +1782,12 @@ 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; } @@ -1675,7 +1812,12 @@ 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; } @@ -1701,12 +1843,13 @@ 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; } @@ -1726,9 +1869,9 @@ 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; } @@ -1805,6 +1948,18 @@ 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. @@ -1877,11 +2032,6 @@ 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); @@ -1906,7 +2056,11 @@ 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; } @@ -2002,19 +2156,23 @@ 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; } @@ -2075,9 +2233,12 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) */ if (opt_verbose_consumer) { verbosity = "--verbose"; - } else { + } else if (lttng_opt_quiet) { verbosity = "--quiet"; + } else { + verbosity = ""; } + switch (consumer_data->type) { case LTTNG_CONSUMER_KERNEL: /* @@ -2100,13 +2261,15 @@ 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: @@ -2145,13 +2308,11 @@ 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: @@ -2190,13 +2351,11 @@ 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: @@ -2204,8 +2363,9 @@ 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; @@ -2275,21 +2435,6 @@ 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. */ @@ -2388,6 +2533,7 @@ 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) { @@ -2431,6 +2577,7 @@ 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: @@ -2451,6 +2598,7 @@ 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 */ @@ -2565,6 +2713,7 @@ 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: @@ -2575,6 +2724,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, case LTTNG_SNAPSHOT_DEL_OUTPUT: case LTTNG_SNAPSHOT_LIST_OUTPUT: case LTTNG_SNAPSHOT_RECORD: + case LTTNG_SAVE_SESSION: need_domain = 0; break; default: @@ -2628,10 +2778,12 @@ 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: case LTTNG_LIST_TRACEPOINT_FIELDS: + case LTTNG_SAVE_SESSION: need_tracing_session = 0; break; default: @@ -2719,6 +2871,7 @@ 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()) { @@ -2810,6 +2963,7 @@ 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; @@ -2891,9 +3045,108 @@ skip_domain: } case LTTNG_ENABLE_EVENT: { + struct lttng_event_exclusion *exclusion = NULL; + struct lttng_filter_bytecode *bytecode = NULL; + char *filter_expression = 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; + } + } + + /* Get filter expression from client. */ + if (cmd_ctx->lsm->u.enable.expression_len > 0) { + size_t expression_len = + cmd_ctx->lsm->u.enable.expression_len; + + if (expression_len > LTTNG_FILTER_MAX_LEN) { + ret = LTTNG_ERR_FILTER_INVAL; + free(exclusion); + goto error; + } + + filter_expression = zmalloc(expression_len); + if (!filter_expression) { + free(exclusion); + ret = LTTNG_ERR_FILTER_NOMEM; + goto error; + } + + /* Receive var. len. data */ + DBG("Receiving var len filter's expression from client ..."); + ret = lttcomm_recv_unix_sock(sock, filter_expression, + expression_len); + if (ret <= 0) { + DBG("Nothing recv() from client car len data... continuing"); + *sock_error = 1; + free(filter_expression); + free(exclusion); + ret = LTTNG_ERR_FILTER_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, + filter_expression, bytecode, exclusion, + kernel_poll_pipe[1]); break; } case LTTNG_ENABLE_ALL_EVENT: @@ -2902,7 +3155,8 @@ skip_domain: ret = cmd_enable_event_all(cmd_ctx->session, &cmd_ctx->lsm->domain, cmd_ctx->lsm->u.enable.channel_name, - cmd_ctx->lsm->u.enable.event.type, NULL, kernel_poll_pipe[1]); + cmd_ctx->lsm->u.enable.event.type, NULL, NULL, + kernel_poll_pipe[1]); break; } case LTTNG_LIST_TRACEPOINTS: @@ -2910,7 +3164,9 @@ 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; @@ -2941,8 +3197,10 @@ 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; @@ -3072,7 +3330,7 @@ skip_domain: } ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, nb_uri, - &cmd_ctx->creds); + &cmd_ctx->creds, 0); free(uris); @@ -3217,46 +3475,6 @@ 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); @@ -3362,6 +3580,51 @@ 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; + } + case LTTNG_SAVE_SESSION: + { + ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr, + &cmd_ctx->creds); + break; + } default: ret = LTTNG_ERR_UND; break; @@ -3395,8 +3658,8 @@ 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"); @@ -3413,6 +3676,27 @@ 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. @@ -3439,6 +3723,8 @@ static void *thread_manage_health(void *data) goto error; } + lttng_sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3503,58 +3789,18 @@ 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) { @@ -3605,11 +3851,7 @@ 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(); @@ -3633,14 +3875,14 @@ 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; } @@ -3815,7 +4057,6 @@ error: error_listen: error_create_poll: -error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3829,7 +4070,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Client thread dying"); @@ -3858,139 +4099,263 @@ 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"); + fprintf(stderr, " -f --config Load daemon configuration file\n"); } /* - * daemon argument parsing + * Take an option from the getopt output and set it in the right variable to be + * used later. + * + * Return 0 on success else a negative value. */ -static int parse_args(int argc, char **argv) +static int set_option(int opt, const char *arg, const char *optname) { - int c; - - static struct option long_options[] = { - { "client-sock", 1, 0, 'c' }, - { "apps-sock", 1, 0, 'a' }, - { "kconsumerd-cmd-sock", 1, 0, 'C' }, - { "kconsumerd-err-sock", 1, 0, 'E' }, - { "ustconsumerd32-cmd-sock", 1, 0, 'G' }, - { "ustconsumerd32-err-sock", 1, 0, 'H' }, - { "ustconsumerd64-cmd-sock", 1, 0, 'D' }, - { "ustconsumerd64-err-sock", 1, 0, 'F' }, - { "consumerd32-path", 1, 0, 'u' }, - { "consumerd32-libdir", 1, 0, 'U' }, - { "consumerd64-path", 1, 0, 't' }, - { "consumerd64-libdir", 1, 0, 'T' }, - { "daemonize", 0, 0, 'd' }, - { "sig-parent", 0, 0, 'S' }, - { "help", 0, 0, 'h' }, - { "group", 1, 0, 'g' }, - { "version", 0, 0, 'V' }, - { "quiet", 0, 0, 'q' }, - { "verbose", 0, 0, 'v' }, - { "verbose-consumer", 0, 0, 'Z' }, - { "no-kernel", 0, 0, 'N' }, - { "pidfile", 1, 0, 'p' }, - { NULL, 0, 0, 0 } - }; + int ret = 0; + + switch (opt) { + case 0: + fprintf(stderr, "option %s", optname); + if (arg) { + fprintf(stderr, " with arg %s\n", arg); + } + break; + case 'c': + snprintf(client_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'a': + snprintf(apps_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'd': + opt_daemon = 1; + break; + case 'b': + opt_background = 1; + break; + case 'g': + tracing_group_name = strdup(arg); + break; + case 'h': + usage(); + exit(EXIT_FAILURE); + case 'V': + fprintf(stdout, "%s\n", VERSION); + exit(EXIT_SUCCESS); + case 'S': + opt_sig_parent = 1; + break; + case 'E': + snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'C': + snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'F': + snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'D': + snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'H': + snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'G': + snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", arg); + break; + case 'N': + opt_no_kernel = 1; + break; + case 'q': + lttng_opt_quiet = 1; + break; + case 'v': + /* Verbose level can increase using multiple -v */ + if (arg) { + lttng_opt_verbose = config_parse_value(arg); + } else { + lttng_opt_verbose += 1; + } + break; + case 'Z': + if (arg) { + opt_verbose_consumer = config_parse_value(arg); + } else { + opt_verbose_consumer += 1; + } + break; + case 'u': + consumerd32_bin = strdup(arg); + consumerd32_bin_override = 1; + break; + case 'U': + consumerd32_libdir = strdup(arg); + consumerd32_libdir_override = 1; + break; + case 't': + consumerd64_bin = strdup(arg); + consumerd64_bin_override = 1; + break; + case 'T': + consumerd64_libdir = strdup(arg); + consumerd64_libdir_override = 1; + break; + case 'p': + opt_pidfile = strdup(arg); + break; + case 'J': /* JUL TCP port. */ + { + unsigned long v; + + errno = 0; + v = strtoul(arg, NULL, 0); + if (errno != 0 || !isdigit(arg[0])) { + ERR("Wrong value in --jul-tcp-port parameter: %s", arg); + return -1; + } + if (v == 0 || v >= 65535) { + ERR("Port overflow in --jul-tcp-port parameter: %s", arg); + 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 */ + ret = -1; + } + + return ret; +} + +/* + * config_entry_handler_cb used to handle options read from a config file. + * See config_entry_handler_cb comment in common/config/config.h for the + * return value conventions. + */ +static int config_entry_handler(const struct config_entry *entry, void *unused) +{ + int ret = 0, i; + + if (!entry || !entry->name || !entry->value) { + ret = -EINVAL; + goto end; + } + + /* Check if the option is to be ignored */ + for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { + if (!strcmp(entry->name, config_ignore_options[i])) { + goto end; + } + } + + for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; + i++) { + + /* Ignore if not fully matched. */ + if (strcmp(entry->name, long_options[i].name)) { + continue; + } + + /* + * If the option takes no argument on the command line, we have to + * check if the value is "true". We support non-zero numeric values, + * true, on and yes. + */ + if (!long_options[i].has_arg) { + ret = config_parse_value(entry->value); + if (ret <= 0) { + if (ret) { + WARN("Invalid configuration value \"%s\" for option %s", + entry->value, entry->name); + } + /* False, skip boolean config option. */ + goto end; + } + } + + ret = set_option(long_options[i].val, entry->value, entry->name); + goto end; + } + + WARN("Unrecognized option \"%s\" in daemon configuration file.", entry->name); + +end: + return ret; +} + +/* + * daemon configuration loading and argument parsing + */ +static int set_options(int argc, char **argv) +{ + int ret = 0, c = 0, option_index = 0; + int orig_optopt = optopt, orig_optind = optind; + char *optstring; + const char *config_path = NULL; + + optstring = utils_generate_optstring(long_options, + sizeof(long_options) / sizeof(struct option)); + if (!optstring) { + ret = -ENOMEM; + goto end; + } + + /* Check for the --config option */ + while ((c = getopt_long(argc, argv, optstring, long_options, + &option_index)) != -1) { + if (c == '?') { + ret = -EINVAL; + goto end; + } else if (c != 'f') { + /* if not equal to --config option. */ + continue; + } + + config_path = utils_expand_path(optarg); + if (!config_path) { + ERR("Failed to resolve path: %s", optarg); + } + } + + ret = config_get_section_entries(config_path, config_section_name, + config_entry_handler, NULL); + if (ret) { + if (ret > 0) { + ERR("Invalid configuration option at line %i", ret); + ret = -1; + } + goto end; + } + /* Reset getopt's global state */ + optopt = orig_optopt; + optind = orig_optind; while (1) { - int option_index = 0; - c = getopt_long(argc, argv, "dhqvVSN" "a:c:g:s:C:E:D:F:Z:u:t:p:", - long_options, &option_index); + c = getopt_long(argc, argv, optstring, long_options, &option_index); if (c == -1) { break; } - switch (c) { - case 0: - fprintf(stderr, "option %s", long_options[option_index].name); - if (optarg) { - fprintf(stderr, " with arg %s\n", optarg); - } - break; - case 'c': - snprintf(client_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'a': - snprintf(apps_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'd': - opt_daemon = 1; - break; - case 'g': - opt_tracing_group = optarg; - break; - case 'h': - usage(); - exit(EXIT_FAILURE); - case 'V': - fprintf(stdout, "%s\n", VERSION); - exit(EXIT_SUCCESS); - case 'S': - opt_sig_parent = 1; - break; - case 'E': - snprintf(kconsumer_data.err_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'C': - snprintf(kconsumer_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'F': - snprintf(ustconsumer64_data.err_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'D': - snprintf(ustconsumer64_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'H': - snprintf(ustconsumer32_data.err_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'G': - snprintf(ustconsumer32_data.cmd_unix_sock_path, PATH_MAX, "%s", optarg); - break; - case 'N': - opt_no_kernel = 1; - break; - case 'q': - lttng_opt_quiet = 1; - break; - case 'v': - /* Verbose level can increase using multiple -v */ - lttng_opt_verbose += 1; - break; - case 'Z': - opt_verbose_consumer += 1; - break; - case 'u': - consumerd32_bin= optarg; - break; - case 'U': - consumerd32_libdir = optarg; - break; - case 't': - consumerd64_bin = optarg; - break; - case 'T': - consumerd64_libdir = optarg; - break; - case 'p': - opt_pidfile = optarg; + ret = set_option(c, optarg, long_options[option_index].name); + if (ret < 0) { break; - default: - /* Unknown option or other error. - * Error is printed by getopt, just return */ - return -1; } } - return 0; +end: + free(optstring); + return ret; } /* @@ -4087,14 +4452,7 @@ 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); @@ -4103,8 +4461,12 @@ 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"); @@ -4118,21 +4480,21 @@ 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"); @@ -4176,7 +4538,7 @@ 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; @@ -4194,7 +4556,7 @@ 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"); @@ -4203,6 +4565,14 @@ 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 = @@ -4256,6 +4626,9 @@ static void sighandler(int sig) DBG("SIGTERM caught"); stop_threads(); break; + case SIGUSR1: + CMM_STORE_SHARED(recv_child_signal, 1); + break; default: break; } @@ -4294,7 +4667,12 @@ 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; } @@ -4350,6 +4728,34 @@ 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; +} + /* * main */ @@ -4363,6 +4769,10 @@ 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); @@ -4372,30 +4782,26 @@ int main(int argc, char **argv) WARN("Fallback page size to %ld", page_size); } - /* Parse arguments */ + /* Parse arguments and load the daemon configuration file */ progname = argv[0]; - if ((ret = parse_args(argc, argv)) < 0) { + if ((ret = set_options(argc, argv)) < 0) { goto error; } /* 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); @@ -4547,6 +4953,12 @@ 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" */ /* @@ -4582,10 +4994,6 @@ 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; @@ -4650,11 +5058,24 @@ int main(int argc, char **argv) } write_pidfile(); + write_julport(); /* Initialize communication library */ lttcomm_init(); + /* This is to get the TCP timeout value. */ + lttcomm_inet_init(); - /* Create thread to manage the client socket */ + /* + * 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) { @@ -4662,7 +5083,7 @@ 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) { @@ -4707,7 +5128,15 @@ 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 */ @@ -4728,12 +5157,27 @@ 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) { @@ -4787,6 +5231,8 @@ 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.