X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=052e32e0c32ec2fb8bfe2a17e71afa0487ebf868;hp=859223a11216b1d342b0899ebdd4e66bc657d657;hb=4421f712a81faad99bb5f20369c276803aeba381;hpb=840cb59cfc335e2ca44841f6fd57972ac3a623be diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 859223a11..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, @@ -20,11 +21,13 @@ #include #include #include +#include #include #include #include #include #include +#include #include #include #include @@ -37,15 +40,17 @@ #include #include -#include #include #include #include #include #include #include +#include +#include #include "lttng-sessiond.h" +#include "buffer-registry.h" #include "channel.h" #include "cmd.h" #include "consumer.h" @@ -59,28 +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_home_dir[] = DEFAULT_HOME_DIR; -const char default_tracing_group[] = DEFAULT_TRACING_GROUP; -const char default_ust_sock_dir[] = DEFAULT_UST_SOCK_DIR; -const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE; - 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. @@ -119,6 +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; @@ -149,20 +188,27 @@ static int thread_quit_pipe[2] = { -1, -1 }; */ static int apps_cmd_pipe[2] = { -1, -1 }; +int apps_cmd_notify_pipe[2] = { -1, -1 }; + /* Pthread, Mutexes and Semaphores */ static pthread_t apps_thread; +static pthread_t apps_notify_thread; static pthread_t reg_apps_thread; static pthread_t client_thread; 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; @@ -185,6 +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"; @@ -227,6 +277,52 @@ static enum consumerd_state kernel_consumerd_state; */ 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) { @@ -279,15 +375,11 @@ void setup_consumerd_path(void) /* * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. */ -static int create_thread_poll_set(struct lttng_poll_event *events, - unsigned int size) +int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size) { int ret; - if (events == NULL || size == 0) { - ret = -1; - goto error; - } + assert(events); ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); if (ret < 0) { @@ -295,7 +387,7 @@ static int create_thread_poll_set(struct lttng_poll_event *events, } /* Add quit pipe */ - ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); + ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); if (ret < 0) { goto error; } @@ -311,7 +403,7 @@ error: * * Return 1 if it was triggered else 0; */ -static int check_thread_quit_pipe(int fd, uint32_t events) +int sessiond_check_thread_quit_pipe(int fd, uint32_t events) { if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { return 1; @@ -320,25 +412,6 @@ static int 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. * @@ -385,18 +458,66 @@ static void stop_threads(void) futex_nto1_wake(&ust_cmd_queue.futex); } +/* + * Close every consumer sockets. + */ +static void close_consumer_sockets(void) +{ + int ret; + + if (kconsumer_data.err_sock >= 0) { + ret = close(kconsumer_data.err_sock); + if (ret < 0) { + PERROR("kernel consumer err_sock close"); + } + } + if (ustconsumer32_data.err_sock >= 0) { + ret = close(ustconsumer32_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd32 err_sock close"); + } + } + if (ustconsumer64_data.err_sock >= 0) { + ret = close(ustconsumer64_data.err_sock); + if (ret < 0) { + PERROR("UST consumerd64 err_sock close"); + } + } + if (kconsumer_data.cmd_sock >= 0) { + ret = close(kconsumer_data.cmd_sock); + if (ret < 0) { + PERROR("kernel consumer cmd_sock close"); + } + } + if (ustconsumer32_data.cmd_sock >= 0) { + ret = close(ustconsumer32_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd32 cmd_sock close"); + } + } + if (ustconsumer64_data.cmd_sock >= 0) { + ret = close(ustconsumer64_data.cmd_sock); + if (ret < 0) { + PERROR("UST consumerd64 cmd_sock close"); + } + } +} + /* * Cleanup the daemon */ static void cleanup(void) { int ret; - char *cmd = NULL; struct ltt_session *sess, *stmp; + char path[PATH_MAX]; DBG("Cleaning up"); - /* First thing first, stop all threads */ + /* + * Close the thread quit pipe. It has already done its job, + * since we are now called. + */ utils_close_pipe(thread_quit_pipe); /* @@ -410,18 +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"); @@ -439,6 +607,7 @@ static void cleanup(void) DBG("Closing all UST sockets"); ust_app_clean_list(); + buffer_reg_destroy_registries(); if (is_root && !opt_no_kernel) { DBG2("Closing kernel fd"); @@ -452,6 +621,32 @@ static void cleanup(void) modprobe_remove_lttng_all(); } + 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", @@ -622,6 +817,8 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) if (ret < 0) { goto error; } + /* Update the stream global counter */ + ksess->stream_count_global += ret; /* * Have we already sent fds to the consumer? If yes, it means @@ -635,12 +832,10 @@ static int update_kernel_stream(struct consumer_data *consumer_data, int fd) rcu_read_lock(); cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter, socket, node.node) { - /* Code flow error */ - assert(socket->fd >= 0); - pthread_mutex_lock(socket->lock); ret = kernel_consumer_send_channel_stream(socket, - channel, ksess); + channel, ksess, + session->output_traces ? 1 : 0); pthread_mutex_unlock(socket->lock); if (ret < 0) { rcu_read_unlock(); @@ -664,13 +859,18 @@ error: } /* - * For each tracing session, update newly registered apps. + * For each tracing session, update newly registered apps. The session list + * lock MUST be acquired before calling this. */ static void update_ust_app(int app_sock) { struct ltt_session *sess, *stmp; - session_lock_list(); + /* Consumer is in an ERROR state. Stop any application update. */ + if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { + /* Stop the update process since the consumer is dead. */ + return; + } /* For all tracing session(s) */ cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) { @@ -680,8 +880,6 @@ static void update_ust_app(int app_sock) } session_unlock(sess); } - - session_unlock_list(); } /* @@ -699,21 +897,21 @@ 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 - * non NULL pointers so zero it before the loop. + * non NULL pointers so initialize it before the loop. */ - memset(&events, 0, sizeof(events)); + 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; } @@ -724,7 +922,7 @@ static void *thread_manage_kernel(void *data) /* Clean events object. We are about to populate it again. */ lttng_poll_clean(&events); - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_poll_create; } @@ -746,9 +944,9 @@ static void *thread_manage_kernel(void *data) /* Poll infinite value of time */ restart: - health_poll_update(); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -774,7 +972,7 @@ static void *thread_manage_kernel(void *data) health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -782,9 +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. @@ -824,7 +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; } @@ -865,32 +1062,15 @@ 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); - /* - * Since the consumer thread can be spawned at any moment in time, we init - * the health to a poll status (1, which is a valid health over time). - * When the thread starts, we update here the health to a "code" path being - * an even value so this thread, when reaching a poll wait, does not - * trigger an error with an even value. - * - * Here is the use case we avoid. - * - * +1: the first poll update during initialization (main()) - * +2 * x: multiple code update once in this thread. - * +1: poll wait in this thread (being a good health state). - * == even number which after the wait period shows as a bad health. - * - * In a nutshell, the following poll update to the health state brings back - * the state to an even value meaning a code path. - */ - health_poll_update(); + health_code_update(); /* - * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock. - * Nothing more will be added to this poll set. + * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the + * metadata_sock. Nothing more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 3); if (ret < 0) { goto error_poll; } @@ -907,16 +1087,16 @@ static void *thread_manage_consumer(void *data) health_code_update(); - /* Inifinite blocking call, waiting for transmission */ + /* Infinite blocking call, waiting for transmission */ restart: - health_poll_update(); + health_poll_entry(); - if (testpoint(thread_manage_consumer)) { + if (testpoint(sessiond_thread_manage_consumer)) { goto error; } ret = lttng_poll_wait(&events, -1); - health_poll_update(); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -937,7 +1117,7 @@ restart: health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -975,91 +1155,137 @@ 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); - if (consumer_data->cmd_sock < 0) { + 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) { + PERROR("consumer connect cmd socket"); /* On error, signal condition and quit. */ signal_consumer_condition(consumer_data, -1); - PERROR("consumer connect"); goto error; } + consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd; + /* Create metadata socket lock. */ + consumer_data->metadata_sock.lock = zmalloc(sizeof(pthread_mutex_t)); + if (consumer_data->metadata_sock.lock == NULL) { + PERROR("zmalloc pthread mutex"); + ret = -1; + goto error; + } + pthread_mutex_init(consumer_data->metadata_sock.lock, NULL); + signal_consumer_condition(consumer_data, 1); - DBG("Consumer command socket ready"); + DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock); + DBG("Consumer metadata socket ready (fd: %d)", + consumer_data->metadata_fd); } else { ERR("consumer error when waiting for SOCK_READY : %s", lttcomm_get_readable_code(-code)); goto error; } - /* Remove the kconsumerd error sock since we've established a connexion */ + /* Remove the consumerd error sock since we've established a connexion */ ret = lttng_poll_del(&events, consumer_data->err_sock); if (ret < 0) { goto error; } + /* Add new accepted error socket. */ ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP); if (ret < 0) { goto error; } + /* Add metadata socket that is successfully connected. */ + ret = lttng_poll_add(&events, consumer_data->metadata_fd, + LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } + health_code_update(); - /* Inifinite blocking call, waiting for transmission */ + /* Infinite blocking call, waiting for transmission */ restart_poll: - health_poll_update(); - ret = lttng_poll_wait(&events, -1); - health_poll_update(); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart_poll; + while (1) { + health_poll_entry(); + ret = lttng_poll_wait(&events, -1); + health_poll_exit(); + if (ret < 0) { + /* + * Restart interrupted system call. + */ + if (errno == EINTR) { + goto restart_poll; + } + goto error; } - goto error; - } - nb_fd = ret; + nb_fd = ret; - for (i = 0; i < nb_fd; i++) { - /* Fetch once the poll data */ - revents = LTTNG_POLL_GETEV(&events, i); - pollfd = LTTNG_POLL_GETFD(&events, i); + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); - health_code_update(); + health_code_update(); - /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; - } + /* Thread quit pipe has been closed. Killing thread. */ + ret = sessiond_check_thread_quit_pipe(pollfd, revents); + if (ret) { + err = 0; + goto exit; + } - /* Event on the kconsumerd socket */ - if (pollfd == sock) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("consumer err socket second poll error"); + if (pollfd == sock) { + /* Event on the consumerd socket */ + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("consumer err socket second poll error"); + goto error; + } + health_code_update(); + /* Wait for any kconsumerd error */ + ret = lttcomm_recv_unix_sock(sock, &code, + sizeof(enum lttcomm_return_code)); + if (ret <= 0) { + ERR("consumer closed the command socket"); + goto error; + } + + ERR("consumer return code : %s", + lttcomm_get_readable_code(-code)); + + goto exit; + } else if (pollfd == consumer_data->metadata_fd) { + /* UST metadata requests */ + ret = ust_consumer_metadata_request( + &consumer_data->metadata_sock); + if (ret < 0) { + ERR("Handling metadata request"); + goto error; + } + break; + } else { + ERR("Unknown pollfd"); goto error; } } + health_code_update(); } - health_code_update(); - - /* Wait for any kconsumerd error */ - ret = lttcomm_recv_unix_sock(sock, &code, - sizeof(enum lttcomm_return_code)); - if (ret <= 0) { - ERR("consumer closed the command socket"); - goto error; - } - - ERR("consumer return code : %s", lttcomm_get_readable_code(-code)); - exit: error: + /* + * We lock here because we are about to close the sockets and some other + * thread might be using them so get exclusive access which will abort all + * other consumer command by other threads. + */ + pthread_mutex_lock(&consumer_data->lock); + /* Immediately set the consumerd state to stopped */ if (consumer_data->type == LTTNG_CONSUMER_KERNEL) { uatomic_set(&kernel_consumerd_state, CONSUMER_ERROR); @@ -1076,12 +1302,21 @@ error: if (ret) { PERROR("close"); } + consumer_data->err_sock = -1; } if (consumer_data->cmd_sock >= 0) { ret = close(consumer_data->cmd_sock); if (ret) { PERROR("close"); } + consumer_data->cmd_sock = -1; + } + if (consumer_data->metadata_sock.fd_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); @@ -1093,14 +1328,20 @@ error: unlink(consumer_data->err_unix_sock_path); unlink(consumer_data->cmd_unix_sock_path); consumer_data->pid = 0; + pthread_mutex_unlock(&consumer_data->lock); + /* Cleanup metadata socket mutex. */ + 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; @@ -1112,8 +1353,8 @@ 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 ust_command ust_cmd; struct lttng_poll_event events; DBG("[thread] Manage application started"); @@ -1121,15 +1362,15 @@ 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; } health_code_update(); - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_poll_create; } @@ -1139,7 +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; } @@ -1150,9 +1391,9 @@ static void *thread_manage_apps(void *data) /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -1173,7 +1414,7 @@ static void *thread_manage_apps(void *data) health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -1185,74 +1426,29 @@ static void *thread_manage_apps(void *data) ERR("Apps command pipe error"); goto error; } else if (revents & LPOLLIN) { + int sock; + /* Empty pipe */ - do { - ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret < sizeof(ust_cmd)) { + size_ret = lttng_read(apps_cmd_pipe[0], &sock, sizeof(sock)); + if (size_ret < sizeof(sock)) { PERROR("read apps cmd pipe"); goto error; } health_code_update(); - /* Register applicaton to the session daemon */ - ret = ust_app_register(&ust_cmd.reg_msg, - ust_cmd.sock); - if (ret == -ENOMEM) { - goto error; - } else if (ret < 0) { - break; - } - - health_code_update(); - /* - * Validate UST version compatibility. + * We only monitor the error events of the socket. This + * thread does not handle any incoming data from UST + * (POLLIN). */ - ret = ust_app_validate_version(ust_cmd.sock); - if (ret >= 0) { - /* - * Add channel(s) and event(s) to newly registered apps - * from lttng global UST domain. - */ - update_ust_app(ust_cmd.sock); - } - - health_code_update(); - - ret = ust_app_register_done(ust_cmd.sock); + ret = lttng_poll_add(&events, sock, + LPOLLERR | LPOLLHUP | LPOLLRDHUP); if (ret < 0) { - /* - * If the registration is not possible, we simply - * unregister the apps and continue - */ - ust_app_unregister(ust_cmd.sock); - } else { - /* - * We only monitor the error events of the socket. This - * thread does not handle any incoming data from UST - * (POLLIN). - */ - ret = lttng_poll_add(&events, ust_cmd.sock, - LPOLLERR & LPOLLHUP & LPOLLRDHUP); - if (ret < 0) { - goto error; - } - - /* Set socket timeout for both receiving and ending */ - (void) lttcomm_setsockopt_rcv_timeout(ust_cmd.sock, - app_socket_timeout); - (void) lttcomm_setsockopt_snd_timeout(ust_cmd.sock, - app_socket_timeout); - - DBG("Apps with sock %d added to poll set", - ust_cmd.sock); + goto error; } - health_code_update(); - - break; + DBG("Apps with sock %d added to poll set", sock); } } else { /* @@ -1268,7 +1464,6 @@ static void *thread_manage_apps(void *data) /* Socket closed on remote end. */ ust_app_unregister(pollfd); - break; } } @@ -1294,7 +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(); @@ -1302,83 +1497,366 @@ error_testpoint: } /* - * Dispatch request from the registration threads to the application - * communication thread. + * 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 void *thread_dispatch_ust_registration(void *data) +static int send_socket_to_thread(int fd, int sock) { - int ret; - struct cds_wfq_node *node; - struct ust_command *ust_cmd = NULL; - - DBG("[thread] Dispatch UST command started"); - - while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { - /* Atomically prepare the queue futex */ - futex_nto1_prepare(&ust_cmd_queue.futex); - - do { - /* Dequeue command for registration */ - node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue); - if (node == NULL) { - DBG("Woken up but nothing in the UST command queue"); - /* Continue thread execution */ - break; - } + ssize_t ret; - ust_cmd = caa_container_of(node, struct ust_command, node); - - DBG("Dispatching UST registration pid:%d ppid:%d uid:%d" - " gid:%d sock:%d name:%s (version %d.%d)", - ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid, - ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid, - ust_cmd->sock, ust_cmd->reg_msg.name, - ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor); - /* - * Inform apps thread of the new application registration. This - * call is blocking so we can be assured that the data will be read - * at some point in time or wait to the end of the world :) - */ - if (apps_cmd_pipe[1] >= 0) { - do { - ret = write(apps_cmd_pipe[1], ust_cmd, - sizeof(struct ust_command)); - } while (ret < 0 && errno == EINTR); - if (ret < 0 || ret != sizeof(struct ust_command)) { - PERROR("write apps cmd pipe"); - if (errno == EBADF) { - /* - * We can't inform the application thread to process - * registration. We will exit or else application - * registration will not occur and tracing will never - * start. - */ - goto error; - } - } - } else { - /* Application manager thread is not available. */ - ret = close(ust_cmd->sock); - if (ret < 0) { - PERROR("close ust_cmd sock"); - } - } - free(ust_cmd); - } while (node != NULL); + /* + * 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; + } - /* Futex wait on queue. Blocking call on futex() */ - futex_nto1_wait(&ust_cmd_queue.futex); + ret = lttng_write(fd, &sock, sizeof(sock)); + if (ret < sizeof(sock)) { + PERROR("write apps pipe %d", fd); + if (ret < 0) { + ret = -errno; + } + goto error; } + /* All good. Don't send back the write positive ret value. */ + ret = 0; error: - DBG("Dispatch thread dying"); - return NULL; + return (int) ret; } /* - * This thread manage application registration. + * Sanitize the wait queue of the dispatch registration thread meaning removing + * invalid nodes from it. This is to avoid memory leaks for the case the UST + * notify socket is never received. */ -static void *thread_registration_apps(void *data) +static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue) +{ + int ret, nb_fd = 0, i; + unsigned int fd_added = 0; + struct lttng_poll_event events; + struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node; + + assert(wait_queue); + + lttng_poll_init(&events); + + /* Just skip everything for an empty queue. */ + if (!wait_queue->count) { + goto end; + } + + ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC); + if (ret < 0) { + goto error_create; + } + + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue->head, head) { + assert(wait_node->app); + ret = lttng_poll_add(&events, wait_node->app->sock, + LPOLLHUP | LPOLLERR); + if (ret < 0) { + goto error; + } + + fd_added = 1; + } + + if (!fd_added) { + goto end; + } + + /* + * Poll but don't block so we can quickly identify the faulty events and + * clean them afterwards from the wait queue. + */ + ret = lttng_poll_wait(&events, 0); + if (ret < 0) { + goto error; + } + nb_fd = ret; + + for (i = 0; i < nb_fd; i++) { + /* Get faulty FD. */ + uint32_t revents = LTTNG_POLL_GETEV(&events, i); + int pollfd = LTTNG_POLL_GETFD(&events, i); + + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue->head, head) { + if (pollfd == wait_node->app->sock && + (revents & (LPOLLHUP | LPOLLERR))) { + cds_list_del(&wait_node->head); + wait_queue->count--; + ust_app_destroy(wait_node->app); + free(wait_node); + break; + } + } + } + + if (nb_fd > 0) { + DBG("Wait queue sanitized, %d node were cleaned up", nb_fd); + } + +end: + lttng_poll_clean(&events); + return; + +error: + lttng_poll_clean(&events); +error_create: + ERR("Unable to sanitize wait queue"); + return; +} + +/* + * Dispatch request from the registration threads to the application + * communication thread. + */ +static void *thread_dispatch_ust_registration(void *data) +{ + int ret, err = -1; + struct cds_wfq_node *node; + struct ust_command *ust_cmd = NULL; + struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node; + struct ust_reg_wait_queue wait_queue = { + .count = 0, + }; + + 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); + + DBG("[thread] Dispatch UST command started"); + + while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { + health_code_update(); + + /* Atomically prepare the queue futex */ + futex_nto1_prepare(&ust_cmd_queue.futex); + + do { + struct ust_app *app = NULL; + ust_cmd = NULL; + + /* + * Make sure we don't have node(s) that have hung up before receiving + * the notify socket. This is to clean the list in order to avoid + * memory leaks from notify socket that are never seen. + */ + sanitize_wait_queue(&wait_queue); + + health_code_update(); + /* Dequeue command for registration */ + node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue); + if (node == NULL) { + DBG("Woken up but nothing in the UST command queue"); + /* Continue thread execution */ + break; + } + + ust_cmd = caa_container_of(node, struct ust_command, node); + + DBG("Dispatching UST registration pid:%d ppid:%d uid:%d" + " gid:%d sock:%d name:%s (version %d.%d)", + ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid, + ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid, + ust_cmd->sock, ust_cmd->reg_msg.name, + ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor); + + if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) { + wait_node = zmalloc(sizeof(*wait_node)); + if (!wait_node) { + PERROR("zmalloc wait_node dispatch"); + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(LTTNG_FD_APPS, 1); + free(ust_cmd); + goto error; + } + CDS_INIT_LIST_HEAD(&wait_node->head); + + /* Create application object if socket is CMD. */ + wait_node->app = ust_app_create(&ust_cmd->reg_msg, + ust_cmd->sock); + if (!wait_node->app) { + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(LTTNG_FD_APPS, 1); + free(wait_node); + free(ust_cmd); + continue; + } + /* + * Add application to the wait queue so we can set the notify + * socket before putting this object in the global ht. + */ + cds_list_add(&wait_node->head, &wait_queue.head); + wait_queue.count++; + + free(ust_cmd); + /* + * We have to continue here since we don't have the notify + * socket and the application MUST be added to the hash table + * only at that moment. + */ + continue; + } else { + /* + * Look for the application in the local wait queue and set the + * notify socket if found. + */ + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue.head, head) { + health_code_update(); + if (wait_node->app->pid == ust_cmd->reg_msg.pid) { + wait_node->app->notify_sock = ust_cmd->sock; + cds_list_del(&wait_node->head); + wait_queue.count--; + app = wait_node->app; + free(wait_node); + DBG3("UST app notify socket %d is set", ust_cmd->sock); + break; + } + } + + /* + * With no application at this stage the received socket is + * basically useless so close it before we free the cmd data + * structure for good. + */ + if (!app) { + ret = close(ust_cmd->sock); + if (ret < 0) { + PERROR("close ust sock dispatch %d", ust_cmd->sock); + } + lttng_fd_put(LTTNG_FD_APPS, 1); + } + free(ust_cmd); + } + + if (app) { + /* + * @session_lock_list + * + * Lock the global session list so from the register up to the + * registration done message, no thread can see the application + * and change its state. + */ + session_lock_list(); + rcu_read_lock(); + + /* + * Add application to the global hash table. This needs to be + * done before the update to the UST registry can locate the + * application. + */ + ust_app_add(app); + + /* Set app version. This call will print an error if needed. */ + (void) ust_app_version(app); + + /* Send notify socket through the notify pipe. */ + ret = send_socket_to_thread(apps_cmd_notify_pipe[1], + app->notify_sock); + if (ret < 0) { + rcu_read_unlock(); + session_unlock_list(); + /* + * 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; + } + + /* + * Update newly registered application with the tracing + * registry info already enabled information. + */ + update_ust_app(app->sock); + + /* + * Don't care about return value. Let the manage apps threads + * handle app unregistration upon socket close. + */ + (void) ust_app_register_done(app->sock); + + /* + * Even if the application socket has been closed, send the app + * to the thread and unregistration will take place at that + * place. + */ + ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock); + if (ret < 0) { + rcu_read_unlock(); + session_unlock_list(); + /* + * 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; + } + + rcu_read_unlock(); + session_unlock_list(); + } + } while (node != NULL); + + health_poll_entry(); + /* Futex wait on queue. Blocking call on futex() */ + futex_nto1_wait(&ust_cmd_queue.futex); + health_poll_exit(); + } + /* Normal exit, no error */ + err = 0; + +error: + /* Clean up wait queue. */ + cds_list_for_each_entry_safe(wait_node, tmp_wait_node, + &wait_queue.head, head) { + cds_list_del(&wait_node->head); + wait_queue.count--; + free(wait_node); + } + +error_testpoint: + DBG("Dispatch thread dying"); + if (err) { + health_error(); + ERR("Health error occurred in %s", __func__); + } + health_unregister(health_sessiond); + return NULL; +} + +/* + * This thread manage application registration. + */ +static void *thread_registration_apps(void *data) { int sock = -1, i, ret, pollfd, err = -1; uint32_t revents, nb_fd; @@ -1391,9 +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; } @@ -1406,7 +1884,7 @@ static void *thread_registration_apps(void *data) * Pass 2 as size here for the thread quit pipe and apps socket. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_create_poll; } @@ -1430,9 +1908,9 @@ static void *thread_registration_apps(void *data) /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -1453,7 +1931,7 @@ static void *thread_registration_apps(void *data) pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -1470,6 +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. @@ -1498,16 +1988,12 @@ static void *thread_registration_apps(void *data) sock = -1; continue; } + health_code_update(); - ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg, - sizeof(struct ust_register_msg)); - if (ret < 0 || ret < sizeof(struct ust_register_msg)) { - if (ret < 0) { - PERROR("lttcomm_recv_unix_sock register apps"); - } else { - ERR("Wrong size received on apps register"); - } + ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg); + if (ret < 0) { free(ust_cmd); + /* Close socket of the application. */ ret = close(sock); if (ret) { PERROR("close"); @@ -1546,11 +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); @@ -1575,7 +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; } @@ -1671,19 +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; } @@ -1744,9 +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: /* @@ -1769,13 +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: @@ -1806,6 +2300,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = putenv(tmpnew); if (ret) { ret = -errno; + free(tmpnew); goto error; } } @@ -1813,13 +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: @@ -1850,6 +2343,7 @@ static pid_t spawn_consumerd(struct consumer_data *consumer_data) ret = putenv(tmpnew); if (ret) { ret = -errno; + free(tmpnew); goto error; } } @@ -1857,13 +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: @@ -1871,8 +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; @@ -1930,7 +2423,7 @@ end: return 0; error: - /* Cleanup already created socket on error. */ + /* Cleanup already created sockets on error. */ if (consumer_data->err_sock >= 0) { int err; @@ -1942,21 +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. */ @@ -2025,6 +2503,8 @@ error: * Copy consumer output from the tracing session to the domain session. The * function also applies the right modification on a per domain basis for the * trace files destination directory. + * + * Should *NOT* be called with RCU read-side lock held. */ static int copy_session_consumer(int domain, struct ltt_session *session) { @@ -2053,6 +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) { @@ -2082,6 +2563,8 @@ error: /* * Create an UST session and add it to the session ust list. + * + * Should *NOT* be called with RCU read-side lock held. */ static int create_ust_session(struct ltt_session *session, struct lttng_domain *domain) @@ -2094,6 +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: @@ -2104,7 +2588,7 @@ static int create_ust_session(struct ltt_session *session, DBG("Creating UST session"); - lus = trace_ust_create_session(session->path, session->id); + lus = trace_ust_create_session(session->id); if (lus == NULL) { ret = LTTNG_ERR_UST_SESS_FAIL; goto error; @@ -2112,6 +2596,9 @@ static int create_ust_session(struct ltt_session *session, lus->uid = session->uid; lus->gid = session->gid; + lus->output_traces = session->output_traces; + lus->snapshot_mode = session->snapshot_mode; + lus->live_timer_interval = session->live_timer; session->ust_session = lus; /* Copy session output to the newly created UST session */ @@ -2168,6 +2655,8 @@ static int create_kernel_session(struct ltt_session *session) session->kernel_session->uid = session->uid; session->kernel_session->gid = session->gid; + session->kernel_session->output_traces = session->output_traces; + session->kernel_session->snapshot_mode = session->snapshot_mode; return LTTNG_OK; @@ -2207,6 +2696,8 @@ static unsigned int lttng_sessions_count(uid_t uid, gid_t gid) * Return any error encountered or 0 for success. * * "sock" is only used for special-case var. len data. + * + * Should *NOT* be called with RCU read-side lock held. */ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, int *sock_error) @@ -2221,12 +2712,19 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: + case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_LIVE: case LTTNG_DESTROY_SESSION: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_DOMAINS: case LTTNG_START_TRACE: case LTTNG_STOP_TRACE: case LTTNG_DATA_PENDING: + case LTTNG_SNAPSHOT_ADD_OUTPUT: + case LTTNG_SNAPSHOT_DEL_OUTPUT: + case LTTNG_SNAPSHOT_LIST_OUTPUT: + case LTTNG_SNAPSHOT_RECORD: + case LTTNG_SAVE_SESSION: need_domain = 0; break; default: @@ -2279,10 +2777,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Commands that DO NOT need a session. */ switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: + case LTTNG_CREATE_SESSION_SNAPSHOT: + case LTTNG_CREATE_SESSION_LIVE: case LTTNG_CALIBRATE: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: case LTTNG_LIST_TRACEPOINT_FIELDS: + case LTTNG_SAVE_SESSION: need_tracing_session = 0; break; default: @@ -2295,12 +2796,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, session_lock_list(); cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name); if (cmd_ctx->session == NULL) { - if (cmd_ctx->lsm->session.name != NULL) { - ret = LTTNG_ERR_SESS_NOT_FOUND; - } else { - /* If no session name specified */ - ret = LTTNG_ERR_SELECT_SESS; - } + ret = LTTNG_ERR_SESS_NOT_FOUND; goto error; } else { /* Acquire lock for the session */ @@ -2351,8 +2847,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* Start the kernel consumer daemon */ pthread_mutex_lock(&kconsumer_data.pid_mutex); if (kconsumer_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&kconsumer_data.pid_mutex); ret = start_consumerd(&kconsumer_data); if (ret < 0) { @@ -2376,8 +2871,13 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, } break; + case LTTNG_DOMAIN_JUL: case LTTNG_DOMAIN_UST: { + if (!ust_app_supported()) { + ret = LTTNG_ERR_NO_UST; + goto error; + } /* Consumer is in an ERROR state. Report back to client */ if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) { ret = LTTNG_ERR_NO_USTCONSUMERD; @@ -2399,8 +2899,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, pthread_mutex_lock(&ustconsumer64_data.pid_mutex); if (consumerd64_bin[0] != '\0' && ustconsumer64_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&ustconsumer64_data.pid_mutex); ret = start_consumerd(&ustconsumer64_data); if (ret < 0) { @@ -2428,8 +2927,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx, int sock, /* 32-bit */ if (consumerd32_bin[0] != '\0' && ustconsumer32_data.pid == 0 && - cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER && - cmd_ctx->session->start_consumer) { + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { pthread_mutex_unlock(&ustconsumer32_data.pid_mutex); ret = start_consumerd(&ustconsumer32_data); if (ret < 0) { @@ -2465,6 +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; @@ -2493,6 +2992,21 @@ skip_domain: } } + /* + * Send relayd information to consumer as soon as we have a domain and a + * session defined. + */ + if (cmd_ctx->session && need_domain) { + /* + * Setup relayd if not done yet. If the relayd information was already + * sent to the consumer, this call will gracefully return. + */ + ret = cmd_setup_relayd(cmd_ctx->session); + if (ret != LTTNG_OK) { + goto error; + } + } + /* Process by command type */ switch (cmd_ctx->lsm->cmd_type) { case LTTNG_ADD_CONTEXT: @@ -2523,55 +3037,126 @@ skip_domain: cmd_ctx->lsm->u.disable.channel_name); break; } - case LTTNG_DISABLE_CONSUMER: - { - ret = cmd_disable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session); - break; - } case LTTNG_ENABLE_CHANNEL: { - ret = cmd_enable_channel(cmd_ctx->session, cmd_ctx->lsm->domain.type, + ret = cmd_enable_channel(cmd_ctx->session, &cmd_ctx->lsm->domain, &cmd_ctx->lsm->u.channel.chan, kernel_poll_pipe[1]); break; } - case LTTNG_ENABLE_CONSUMER: + case LTTNG_ENABLE_EVENT: { - /* - * XXX: 0 means that this URI should be applied on the session. Should - * be a DOMAIN enuam. - */ - ret = cmd_enable_consumer(cmd_ctx->lsm->domain.type, cmd_ctx->session); - if (ret != LTTNG_OK) { - goto error; - } + 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; + } - if (cmd_ctx->lsm->domain.type == 0) { - /* Add the URI for the UST session if a consumer is present. */ - if (cmd_ctx->session->ust_session && - cmd_ctx->session->ust_session->consumer) { - ret = cmd_enable_consumer(LTTNG_DOMAIN_UST, cmd_ctx->session); - } else if (cmd_ctx->session->kernel_session && - cmd_ctx->session->kernel_session->consumer) { - ret = cmd_enable_consumer(LTTNG_DOMAIN_KERNEL, - cmd_ctx->session); + 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; } } - break; - } - case LTTNG_ENABLE_EVENT: - { - ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, NULL, kernel_poll_pipe[1]); - break; - } + + /* 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, + filter_expression, bytecode, exclusion, + kernel_poll_pipe[1]); + break; + } case LTTNG_ENABLE_ALL_EVENT: { DBG("Enabling all events"); - ret = cmd_enable_event_all(cmd_ctx->session, cmd_ctx->lsm->domain.type, + 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: @@ -2579,7 +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; @@ -2610,8 +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; @@ -2741,7 +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); @@ -2769,6 +3358,7 @@ skip_domain: ret = setup_lttng_msg(cmd_ctx, nb_dom * sizeof(struct lttng_domain)); if (ret < 0) { + free(domains); goto setup_error; } @@ -2796,6 +3386,7 @@ skip_domain: ret = setup_lttng_msg(cmd_ctx, nb_chan * sizeof(struct lttng_channel)); if (ret < 0) { + free(channels); goto setup_error; } @@ -2823,6 +3414,7 @@ skip_domain: ret = setup_lttng_msg(cmd_ctx, nb_event * sizeof(struct lttng_event)); if (ret < 0) { + free(events); goto setup_error; } @@ -2883,49 +3475,154 @@ skip_domain: cmd_ctx->lsm->u.reg.path, cdata); break; } - case LTTNG_ENABLE_EVENT_WITH_FILTER: + case LTTNG_DATA_PENDING: + { + ret = cmd_data_pending(cmd_ctx->session); + break; + } + case LTTNG_SNAPSHOT_ADD_OUTPUT: { - struct lttng_filter_bytecode *bytecode; + struct lttcomm_lttng_output_id reply; - if (cmd_ctx->lsm->u.enable.bytecode_len > LTTNG_FILTER_MAX_LEN) { - ret = LTTNG_ERR_FILTER_INVAL; + ret = cmd_snapshot_add_output(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_output.output, &reply.id); + if (ret != LTTNG_OK) { goto error; } - if (cmd_ctx->lsm->u.enable.bytecode_len == 0) { - ret = LTTNG_ERR_FILTER_INVAL; - goto error; + + ret = setup_lttng_msg(cmd_ctx, sizeof(reply)); + if (ret < 0) { + goto setup_error; } - bytecode = zmalloc(cmd_ctx->lsm->u.enable.bytecode_len); - if (!bytecode) { - ret = LTTNG_ERR_FILTER_NOMEM; + + /* Copy output list into message payload */ + memcpy(cmd_ctx->llm->payload, &reply, sizeof(reply)); + ret = LTTNG_OK; + break; + } + case LTTNG_SNAPSHOT_DEL_OUTPUT: + { + ret = cmd_snapshot_del_output(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_output.output); + break; + } + case LTTNG_SNAPSHOT_LIST_OUTPUT: + { + ssize_t nb_output; + struct lttng_snapshot_output *outputs = NULL; + + nb_output = cmd_snapshot_list_outputs(cmd_ctx->session, &outputs); + if (nb_output < 0) { + ret = -nb_output; goto error; } - /* 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; + + ret = setup_lttng_msg(cmd_ctx, + nb_output * sizeof(struct lttng_snapshot_output)); + if (ret < 0) { + free(outputs); + goto setup_error; } - if (bytecode->len + sizeof(*bytecode) - != cmd_ctx->lsm->u.enable.bytecode_len) { - free(bytecode); - ret = LTTNG_ERR_FILTER_INVAL; - goto error; + if (outputs) { + /* Copy output list into message payload */ + memcpy(cmd_ctx->llm->payload, outputs, + nb_output * sizeof(struct lttng_snapshot_output)); + free(outputs); } - ret = cmd_enable_event(cmd_ctx->session, cmd_ctx->lsm->domain.type, - cmd_ctx->lsm->u.enable.channel_name, - &cmd_ctx->lsm->u.enable.event, bytecode, kernel_poll_pipe[1]); + ret = LTTNG_OK; break; } - case LTTNG_DATA_PENDING: + case LTTNG_SNAPSHOT_RECORD: { - ret = cmd_data_pending(cmd_ctx->session); + ret = cmd_snapshot_record(cmd_ctx->session, + &cmd_ctx->lsm->u.snapshot_record.output, + cmd_ctx->lsm->u.snapshot_record.wait); + break; + } + case LTTNG_CREATE_SESSION_SNAPSHOT: + { + size_t nb_uri, len; + struct lttng_uri *uris = NULL; + + nb_uri = cmd_ctx->lsm->u.uri.size; + len = nb_uri * sizeof(struct lttng_uri); + + if (nb_uri > 0) { + uris = zmalloc(len); + if (uris == NULL) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + /* Receive variable len data */ + DBG("Waiting for %zu URIs from client ...", nb_uri); + ret = lttcomm_recv_unix_sock(sock, uris, len); + if (ret <= 0) { + DBG("No URIs received from client... continuing"); + *sock_error = 1; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + + if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) { + DBG("Creating session with ONE network URI is a bad call"); + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + } + + ret = cmd_create_session_snapshot(cmd_ctx->lsm->session.name, uris, + nb_uri, &cmd_ctx->creds); + free(uris); + break; + } + case LTTNG_CREATE_SESSION_LIVE: + { + size_t nb_uri, len; + struct lttng_uri *uris = NULL; + + nb_uri = cmd_ctx->lsm->u.uri.size; + len = nb_uri * sizeof(struct lttng_uri); + + if (nb_uri > 0) { + uris = zmalloc(len); + if (uris == NULL) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + /* Receive variable len data */ + DBG("Waiting for %zu URIs from client ...", nb_uri); + ret = lttcomm_recv_unix_sock(sock, uris, len); + if (ret <= 0) { + DBG("No URIs received from client... continuing"); + *sock_error = 1; + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + + if (nb_uri == 1 && uris[0].dtype != LTTNG_DST_PATH) { + DBG("Creating session with ONE network URI is a bad call"); + ret = LTTNG_ERR_SESSION_FAIL; + free(uris); + goto error; + } + } + + ret = cmd_create_session_uri(cmd_ctx->lsm->session.name, uris, + nb_uri, &cmd_ctx->creds, cmd_ctx->lsm->u.session_live.timer_interval); + free(uris); + break; + } + case LTTNG_SAVE_SESSION: + { + ret = cmd_save_sessions(&cmd_ctx->lsm->u.save_session.attr, + &cmd_ctx->creds); break; } default: @@ -2961,13 +3658,16 @@ 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"); rcu_register_thread(); + /* We might hit an error path before this is created. */ + lttng_poll_init(&events); + /* Create unix socket */ sock = lttcomm_create_unix_sock(health_unix_sock_path); if (sock < 0) { @@ -2976,6 +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. @@ -2991,7 +3712,7 @@ static void *thread_manage_health(void *data) * Pass 2 as size here for the thread quit pipe and client_sock. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error; } @@ -3002,6 +3723,8 @@ static void *thread_manage_health(void *data) goto error; } + lttng_sessiond_notify_ready(); + while (1) { DBG("Health check ready"); @@ -3026,7 +3749,7 @@ restart: pollfd = LTTNG_POLL_GETFD(&events, i); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -3066,46 +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_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(); - 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) { @@ -3133,12 +3828,6 @@ error: PERROR("close"); } } - if (new_sock >= 0) { - ret = close(new_sock); - if (ret) { - PERROR("close"); - } - } lttng_poll_clean(&events); @@ -3162,11 +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(); @@ -3179,7 +3864,7 @@ static void *thread_manage_clients(void *data) * Pass 2 as size here for the thread quit pipe and client_sock. Nothing * more will be added to this poll set. */ - ret = create_thread_poll_set(&events, 2); + ret = sessiond_set_thread_pollset(&events, 2); if (ret < 0) { goto error_create_poll; } @@ -3190,14 +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; } @@ -3208,9 +3893,9 @@ static void *thread_manage_clients(void *data) /* Inifinite blocking call, waiting for transmission */ restart: - health_poll_update(); + health_poll_entry(); ret = lttng_poll_wait(&events, -1); - health_poll_update(); + health_poll_exit(); if (ret < 0) { /* * Restart interrupted system call. @@ -3231,7 +3916,7 @@ static void *thread_manage_clients(void *data) health_code_update(); /* Thread quit pipe has been closed. Killing thread. */ - ret = check_thread_quit_pipe(pollfd, revents); + ret = sessiond_check_thread_quit_pipe(pollfd, revents); if (ret) { err = 0; goto exit; @@ -3320,13 +4005,11 @@ static void *thread_manage_clients(void *data) ret = process_client_msg(cmd_ctx, sock, &sock_error); rcu_thread_offline(); if (ret < 0) { - if (sock_error) { - ret = close(sock); - if (ret) { - PERROR("close"); - } - sock = -1; + ret = close(sock); + if (ret) { + PERROR("close"); } + sock = -1; /* * TODO: Inform client somehow of the fatal error. At * this point, ret < 0 means that a zmalloc failed @@ -3374,7 +4057,6 @@ error: error_listen: error_create_poll: -error_testpoint: unlink(client_unix_sock_path); if (client_sock >= 0) { ret = close(client_sock); @@ -3388,7 +4070,7 @@ error_testpoint: ERR("Health error occurred in %s", __func__); } - health_unregister(); + health_unregister(health_sessiond); DBG("Client thread dying"); @@ -3417,139 +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; } /* @@ -3646,14 +4452,7 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - ret = allowed_group(); - if (ret < 0) { - WARN("No tracing group detected"); - ret = 0; - goto end; - } - - gid = ret; + gid = utils_get_group_id(tracing_group_name); /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -3662,8 +4461,12 @@ static int set_permissions(char *rundir) PERROR("chown"); } - /* Ensure tracing group can search the run dir */ - ret = chmod(rundir, S_IRWXU | S_IXGRP | S_IXOTH); + /* + * Ensure all applications and tracing group can search the run + * dir. Allow everyone to read the directory, since it does not + * buy us anything to hide its content. + */ + ret = chmod(rundir, S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH); if (ret < 0) { ERR("Unable to set permissions on %s", rundir); PERROR("chmod"); @@ -3677,21 +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"); @@ -3699,7 +4502,6 @@ static int set_permissions(char *rundir) DBG("All permissions are set"); -end: return ret; } @@ -3736,7 +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; @@ -3754,7 +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"); @@ -3763,6 +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 = @@ -3773,6 +4583,16 @@ static int set_consumer_sockets(struct consumer_data *consumer_data, goto error; } + /* + * Set the CLOEXEC flag. Return code is useless because either way, the + * show must go on. + */ + ret = utils_set_fd_cloexec(consumer_data->err_sock); + if (ret < 0) { + PERROR("utils_set_fd_cloexec"); + /* continue anyway */ + } + /* File permission MUST be 660 */ ret = chmod(consumer_data->err_unix_sock_path, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP); @@ -3806,6 +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; } @@ -3844,7 +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; } @@ -3900,6 +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 */ @@ -3913,32 +4769,39 @@ int main(int argc, char **argv) rcu_register_thread(); + if ((ret = set_signal_handler()) < 0) { + goto error; + } + setup_consumerd_path(); - /* Parse arguments */ + page_size = sysconf(_SC_PAGESIZE); + if (page_size < 0) { + PERROR("sysconf _SC_PAGESIZE"); + page_size = LONG_MAX; + WARN("Fallback page size to %ld", page_size); + } + + /* 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); @@ -3994,7 +4857,7 @@ int main(int argc, char **argv) DBG2("Kernel consumer cmd path: %s", kconsumer_data.cmd_unix_sock_path); } else { - home_path = get_home_dir(); + home_path = utils_get_home_dir(); if (home_path == NULL) { /* TODO: Add --socket PATH option */ ERR("Can't get HOME directory for sockets creation."); @@ -4031,7 +4894,7 @@ int main(int argc, char **argv) /* Set global SHM for ust */ if (strlen(wait_shm_path) == 0) { snprintf(wait_shm_path, PATH_MAX, - DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid()); + DEFAULT_HOME_APPS_WAIT_SHM_PATH, getuid()); } /* Set health check Unix path */ @@ -4047,6 +4910,7 @@ int main(int argc, char **argv) DBG("Client socket path %s", client_unix_sock_path); DBG("Application socket path %s", apps_unix_sock_path); + DBG("Application wait path %s", wait_shm_path); DBG("LTTng run directory path: %s", rundir); /* 32 bits consumerd path setup */ @@ -4089,6 +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" */ /* @@ -4124,10 +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; @@ -4150,11 +5016,25 @@ int main(int argc, char **argv) } } + /* Setup the thread ht_cleanup communication pipe. */ + if (utils_create_pipe_cloexec(ht_cleanup_pipe) < 0) { + goto exit; + } + /* Setup the thread apps communication pipe. */ if ((ret = utils_create_pipe_cloexec(apps_cmd_pipe)) < 0) { goto exit; } + /* Setup the thread apps notify communication pipe. */ + if (utils_create_pipe_cloexec(apps_cmd_notify_pipe) < 0) { + goto exit; + } + + /* Initialize global buffer per UID and PID registry. */ + buffer_reg_init_uid_registry(); + buffer_reg_init_pid_registry(); + /* Init UST command queue. */ cds_wfq_init(&ust_cmd_queue.queue); @@ -4178,8 +5058,32 @@ int main(int argc, char **argv) } write_pidfile(); + write_julport(); - /* Create thread to manage the client socket */ + /* Initialize communication library */ + lttcomm_init(); + /* This is to get the TCP timeout value. */ + lttcomm_inet_init(); + + /* + * Initialize the health check subsystem. This call should set the + * appropriate time values. + */ + health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES); + if (!health_sessiond) { + PERROR("health_app_create error"); + goto exit_health_sessiond_cleanup; + } + + /* Create thread to clean up RCU hash tables */ + ret = pthread_create(&ht_cleanup_thread, NULL, + thread_ht_cleanup, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create ht_cleanup"); + goto exit_ht_cleanup; + } + + /* Create health-check thread */ ret = pthread_create(&health_thread, NULL, thread_manage_health, (void *) NULL); if (ret != 0) { @@ -4219,6 +5123,22 @@ int main(int argc, char **argv) goto exit_apps; } + /* Create thread to manage application notify socket */ + ret = pthread_create(&apps_notify_thread, NULL, + ust_thread_manage_notify, (void *) NULL); + if (ret != 0) { + PERROR("pthread_create 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 */ if (is_root && !opt_no_kernel) { /* Create kernel thread to manage kernel event */ @@ -4237,12 +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) { @@ -4290,6 +5225,14 @@ exit_client: } exit_health: + ret = pthread_join(ht_cleanup_thread, &status); + if (ret != 0) { + PERROR("pthread_join ht cleanup thread"); + 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.