X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=40a418d0cfd0a79c453a8637df7322e15aca286e;hp=b211370d538f03a0de8ba00121a4333160053b29;hb=b7fc068d873bcfc93761f418bfefe8c928c33a59;hpb=1785d7f29bf3f162f37152ee9ea8907b9bc6d9d2 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index b211370d5..40a418d0c 100644 --- a/src/bin/lttng-sessiond/main.c +++ b/src/bin/lttng-sessiond/main.c @@ -1,20 +1,10 @@ /* - * Copyright (C) 2011 - David Goulet - * Mathieu Desnoyers - * 2013 - Jérémie Galarneau + * Copyright (C) 2011 David Goulet + * Copyright (C) 2011 Mathieu Desnoyers + * Copyright (C) 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, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #define _LGPL_SOURCE @@ -61,22 +51,18 @@ #include "event.h" #include "kernel.h" #include "kernel-consumer.h" -#include "modprobe.h" -#include "shm.h" -#include "ust-ctl.h" +#include "lttng-ust-ctl.h" #include "ust-consumer.h" #include "utils.h" #include "fd-limit.h" #include "health-sessiond.h" #include "testpoint.h" -#include "ust-thread.h" +#include "notify-apps.h" #include "agent-thread.h" #include "save.h" -#include "load-session-thread.h" #include "notification-thread.h" #include "notification-thread-commands.h" #include "rotation-thread.h" -#include "lttng-syscall.h" #include "agent.h" #include "ht-cleanup.h" #include "sessiond-config.h" @@ -85,6 +71,8 @@ #include "client.h" #include "dispatch.h" #include "register.h" +#include "manage-apps.h" +#include "manage-kernel.h" static const char *help_msg = #ifdef LTTNG_EMBED_HELP @@ -96,6 +84,7 @@ NULL const char *progname; static int lockfile_fd = -1; +static int opt_print_version; /* Set to 1 when a SIGUSR1 signal is received. */ static int recv_child_signal; @@ -141,13 +130,7 @@ static const char *config_ignore_options[] = { "help", "version", "config" }; * that a command is queued and ready to be processed. */ static int apps_cmd_pipe[2] = { -1, -1 }; - -/* Pthread, Mutexes and Semaphores */ -static pthread_t apps_thread; -static pthread_t apps_notify_thread; -static pthread_t kernel_thread; -static pthread_t agent_reg_thread; -static pthread_t load_session_thread; +static int apps_cmd_notify_pipe[2] = { -1, -1 }; /* * UST registration command queue. This queue is tied with a futex and uses a N @@ -160,11 +143,6 @@ static pthread_t load_session_thread; */ static struct ust_cmd_queue ust_cmd_queue; -static const char *module_proc_lttng = "/proc/lttng"; - -/* Load session thread information to operate. */ -static struct load_session_thread_data *load_info; - /* * Section name to look for in the daemon configuration file. */ @@ -293,6 +271,9 @@ static void sessiond_cleanup(void) * since we are now called. */ sessiond_close_quit_pipe(); + utils_close_pipe(apps_cmd_pipe); + utils_close_pipe(apps_cmd_notify_pipe); + utils_close_pipe(kernel_poll_pipe); ret = remove(config.pid_file_path.value); if (ret < 0) { @@ -332,35 +313,23 @@ static void sessiond_cleanup(void) pthread_mutex_destroy(&session_list->lock); - wait_consumer(&kconsumer_data); - wait_consumer(&ustconsumer64_data); - wait_consumer(&ustconsumer32_data); + DBG("Cleaning up all per-event notifier domain agents"); + agent_by_event_notifier_domain_ht_destroy(); DBG("Cleaning up all agent apps"); agent_app_ht_clean(); - DBG("Closing all UST sockets"); ust_app_clean_list(); buffer_reg_destroy_registries(); - if (is_root && !config.no_kernel) { - DBG2("Closing kernel fd"); - if (kernel_tracer_fd >= 0) { - ret = close(kernel_tracer_fd); - if (ret) { - PERROR("close"); - } - } - DBG("Unloading kernel modules"); - modprobe_remove_lttng_all(); - free(syscall_table); - } - close_consumer_sockets(); - if (load_info) { - load_session_destroy_data(load_info); - free(load_info); + wait_consumer(&kconsumer_data); + wait_consumer(&ustconsumer64_data); + wait_consumer(&ustconsumer32_data); + + if (is_root && !config.no_kernel) { + cleanup_kernel_tracer(); } /* @@ -382,920 +351,6 @@ static void sessiond_cleanup_options(void) run_as_destroy_worker(); } -/* - * Update the kernel poll set of all channel fd available over all tracing - * session. Add the wakeup pipe at the end of the set. - */ -static int update_kernel_poll(struct lttng_poll_event *events) -{ - int ret; - struct ltt_kernel_channel *channel; - struct ltt_session *session; - const struct ltt_session_list *session_list = session_get_list(); - - DBG("Updating kernel poll set"); - - session_lock_list(); - cds_list_for_each_entry(session, &session_list->head, list) { - if (!session_get(session)) { - continue; - } - session_lock(session); - if (session->kernel_session == NULL) { - session_unlock(session); - session_put(session); - continue; - } - - cds_list_for_each_entry(channel, - &session->kernel_session->channel_list.head, list) { - /* Add channel fd to the kernel poll set */ - ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM); - if (ret < 0) { - session_unlock(session); - session_put(session); - goto error; - } - DBG("Channel fd %d added to kernel set", channel->fd); - } - session_unlock(session); - } - session_unlock_list(); - - return 0; - -error: - session_unlock_list(); - return -1; -} - -/* - * Find the channel fd from 'fd' over all tracing session. When found, check - * for new channel stream and send those stream fds to the kernel consumer. - * - * Useful for CPU hotplug feature. - */ -static int update_kernel_stream(int fd) -{ - int ret = 0; - struct ltt_session *session; - struct ltt_kernel_session *ksess; - struct ltt_kernel_channel *channel; - const struct ltt_session_list *session_list = session_get_list(); - - DBG("Updating kernel streams for channel fd %d", fd); - - session_lock_list(); - cds_list_for_each_entry(session, &session_list->head, list) { - if (!session_get(session)) { - continue; - } - session_lock(session); - if (session->kernel_session == NULL) { - session_unlock(session); - session_put(session); - continue; - } - ksess = session->kernel_session; - - cds_list_for_each_entry(channel, - &ksess->channel_list.head, list) { - struct lttng_ht_iter iter; - struct consumer_socket *socket; - - if (channel->fd != fd) { - continue; - } - DBG("Channel found, updating kernel streams"); - ret = kernel_open_channel_stream(channel); - 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 that tracing is started so it is safe to send - * our updated stream fds. - */ - if (ksess->consumer_fds_sent != 1 - || ksess->consumer == NULL) { - ret = -1; - goto error; - } - - rcu_read_lock(); - cds_lfht_for_each_entry(ksess->consumer->socks->ht, - &iter.iter, socket, node.node) { - pthread_mutex_lock(socket->lock); - ret = kernel_consumer_send_channel_streams(socket, - channel, ksess, - session->output_traces ? 1 : 0); - pthread_mutex_unlock(socket->lock); - if (ret < 0) { - rcu_read_unlock(); - goto error; - } - } - rcu_read_unlock(); - } - session_unlock(session); - session_put(session); - } - session_unlock_list(); - return ret; - -error: - session_unlock(session); - session_put(session); - session_unlock_list(); - return ret; -} - -/* - * This thread manage event coming from the kernel. - * - * Features supported in this thread: - * -) CPU Hotplug - */ -static void *thread_manage_kernel(void *data) -{ - int ret, i, pollfd, update_poll_flag = 1, err = -1; - uint32_t revents, nb_fd; - char tmp; - struct lttng_poll_event events; - - DBG("[thread] Thread manage kernel started"); - - 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 initialize it before the loop. - */ - lttng_poll_init(&events); - - if (testpoint(sessiond_thread_manage_kernel)) { - goto error_testpoint; - } - - health_code_update(); - - if (testpoint(sessiond_thread_manage_kernel_before_loop)) { - goto error_testpoint; - } - - while (1) { - health_code_update(); - - if (update_poll_flag == 1) { - /* Clean events object. We are about to populate it again. */ - lttng_poll_clean(&events); - - ret = sessiond_set_thread_pollset(&events, 2); - if (ret < 0) { - goto error_poll_create; - } - - ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); - if (ret < 0) { - goto error; - } - - /* This will add the available kernel channel if any. */ - ret = update_kernel_poll(&events); - if (ret < 0) { - goto error; - } - update_poll_flag = 0; - } - - DBG("Thread kernel polling"); - - /* Poll infinite value of time */ - restart: - health_poll_entry(); - ret = lttng_poll_wait(&events, -1); - DBG("Thread kernel return from poll on %d fds", - LTTNG_POLL_GETNB(&events)); - health_poll_exit(); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart; - } - goto error; - } else if (ret == 0) { - /* Should not happen since timeout is infinite */ - ERR("Return value of poll is 0 with an infinite timeout.\n" - "This should not have happened! Continuing..."); - continue; - } - - 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); - - health_code_update(); - - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - - /* Thread quit pipe has been closed. Killing thread. */ - ret = sessiond_check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; - } - - /* Check for data on kernel pipe */ - if (revents & LPOLLIN) { - if (pollfd == kernel_poll_pipe[0]) { - (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. - */ - update_poll_flag = 1; - continue; - } else { - /* - * New CPU detected by the kernel. Adding kernel stream to - * kernel session and updating the kernel consumer - */ - ret = update_kernel_stream(pollfd); - if (ret < 0) { - continue; - } - break; - } - } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - update_poll_flag = 1; - continue; - } else { - ERR("Unexpected poll events %u for sock %d", revents, pollfd); - goto error; - } - } - } - -exit: -error: - lttng_poll_clean(&events); -error_poll_create: -error_testpoint: - utils_close_pipe(kernel_poll_pipe); - kernel_poll_pipe[0] = kernel_poll_pipe[1] = -1; - if (err) { - health_error(); - ERR("Health error occurred in %s", __func__); - WARN("Kernel thread died unexpectedly. " - "Kernel tracing can continue but CPU hotplug is disabled."); - } - health_unregister(health_sessiond); - DBG("Kernel thread dying"); - return NULL; -} - -/* - * Signal pthread condition of the consumer data that the thread. - */ -static void signal_consumer_condition(struct consumer_data *data, int state) -{ - pthread_mutex_lock(&data->cond_mutex); - - /* - * The state is set before signaling. It can be any value, it's the waiter - * job to correctly interpret this condition variable associated to the - * consumer pthread_cond. - * - * A value of 0 means that the corresponding thread of the consumer data - * was not started. 1 indicates that the thread has started and is ready - * for action. A negative value means that there was an error during the - * thread bootstrap. - */ - data->consumer_thread_is_ready = state; - (void) pthread_cond_signal(&data->cond); - - pthread_mutex_unlock(&data->cond_mutex); -} - -/* - * This thread manage the consumer error sent back to the session daemon. - */ -void *thread_manage_consumer(void *data) -{ - int sock = -1, i, ret, pollfd, err = -1, should_quit = 0; - uint32_t revents, nb_fd; - enum lttcomm_return_code code; - struct lttng_poll_event events; - struct consumer_data *consumer_data = data; - struct consumer_socket *cmd_socket_wrapper = NULL; - - DBG("[thread] Manage consumer started"); - - rcu_register_thread(); - rcu_thread_online(); - - health_register(health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); - - health_code_update(); - - /* - * 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 = sessiond_set_thread_pollset(&events, 3); - if (ret < 0) { - goto error_poll; - } - - /* - * The error socket here is already in a listening state which was done - * just before spawning this thread to avoid a race between the consumer - * daemon exec trying to connect and the listen() call. - */ - ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP); - if (ret < 0) { - goto error; - } - - health_code_update(); - - /* Infinite blocking call, waiting for transmission */ -restart: - health_poll_entry(); - - if (testpoint(sessiond_thread_manage_consumer)) { - goto error; - } - - ret = lttng_poll_wait(&events, -1); - health_poll_exit(); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart; - } - goto error; - } - - 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); - - health_code_update(); - - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - - /* Thread quit pipe has been closed. Killing thread. */ - ret = sessiond_check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; - } - - /* Event on the registration socket */ - if (pollfd == consumer_data->err_sock) { - if (revents & LPOLLIN) { - continue; - } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("consumer err socket poll error"); - goto error; - } else { - ERR("Unexpected poll events %u for sock %d", revents, pollfd); - goto error; - } - } - } - - sock = lttcomm_accept_unix_sock(consumer_data->err_sock); - if (sock < 0) { - goto error; - } - - /* - * Set the CLOEXEC flag. Return code is useless because either way, the - * show must go on. - */ - (void) utils_set_fd_cloexec(sock); - - health_code_update(); - - DBG2("Receiving code from consumer err_sock"); - - /* Getting status code from kconsumerd */ - ret = lttcomm_recv_unix_sock(sock, &code, - sizeof(enum lttcomm_return_code)); - if (ret <= 0) { - goto error; - } - - health_code_update(); - if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { - ERR("consumer error when waiting for SOCK_READY : %s", - lttcomm_get_readable_code(-code)); - goto error; - } - - /* Connect both command and metadata sockets. */ - 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) { - PERROR("consumer connect cmd socket"); - /* On error, signal condition and quit. */ - signal_consumer_condition(consumer_data, -1); - 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"); - goto error; - } - pthread_mutex_init(consumer_data->metadata_sock.lock, NULL); - - DBG("Consumer command socket ready (fd: %d", consumer_data->cmd_sock); - DBG("Consumer metadata socket ready (fd: %d)", - consumer_data->metadata_fd); - - /* - * Remove the consumerd error sock since we've established a connection. - */ - 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(); - - /* - * Transfer the write-end of the channel monitoring and rotate pipe - * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE command. - */ - cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock); - if (!cmd_socket_wrapper) { - goto error; - } - cmd_socket_wrapper->lock = &consumer_data->lock; - - ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper, - consumer_data->channel_monitor_pipe); - if (ret) { - goto error; - } - - /* Discard the socket wrapper as it is no longer needed. */ - consumer_destroy_socket(cmd_socket_wrapper); - cmd_socket_wrapper = NULL; - - /* The thread is completely initialized, signal that it is ready. */ - signal_consumer_condition(consumer_data, 1); - - /* Infinite blocking call, waiting for transmission */ -restart_poll: - while (1) { - health_code_update(); - - /* Exit the thread because the thread quit pipe has been triggered. */ - if (should_quit) { - /* Not a health error. */ - err = 0; - goto exit; - } - - health_poll_entry(); - ret = lttng_poll_wait(&events, -1); - health_poll_exit(); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart_poll; - } - goto error; - } - - 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); - - health_code_update(); - - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - - /* - * Thread quit pipe has been triggered, flag that we should stop - * but continue the current loop to handle potential data from - * consumer. - */ - should_quit = sessiond_check_thread_quit_pipe(pollfd, revents); - - if (pollfd == sock) { - /* Event on the consumerd socket */ - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) - && !(revents & LPOLLIN)) { - 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) { - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) - && !(revents & LPOLLIN)) { - ERR("consumer err metadata socket second poll error"); - goto error; - } - /* UST metadata requests */ - ret = ust_consumer_metadata_request( - &consumer_data->metadata_sock); - if (ret < 0) { - ERR("Handling metadata request"); - goto error; - } - } - /* No need for an else branch all FDs are tested prior. */ - } - health_code_update(); - } - -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); - } else if (consumer_data->type == LTTNG_CONSUMER64_UST || - consumer_data->type == LTTNG_CONSUMER32_UST) { - uatomic_set(&ust_consumerd_state, CONSUMER_ERROR); - } else { - /* Code flow error... */ - assert(0); - } - - if (consumer_data->err_sock >= 0) { - ret = close(consumer_data->err_sock); - 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); - if (ret) { - PERROR("close"); - } - } - - unlink(consumer_data->err_unix_sock_path); - unlink(consumer_data->cmd_unix_sock_path); - 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); - - if (cmd_socket_wrapper) { - consumer_destroy_socket(cmd_socket_wrapper); - } -error_poll: - if (err) { - health_error(); - ERR("Health error occurred in %s", __func__); - } - health_unregister(health_sessiond); - DBG("consumer thread cleanup completed"); - - rcu_thread_offline(); - rcu_unregister_thread(); - - return NULL; -} - -/* - * This thread receives application command sockets (FDs) on the - * apps_cmd_pipe and waits (polls) on them until they are closed - * or an error occurs. - * - * At that point, it flushes the data (tracing and metadata) associated - * with this application and tears down ust app sessions and other - * associated data structures through ust_app_unregister(). - * - * Note that this thread never sends commands to the applications - * through the command sockets; it merely listens for hang-ups - * and errors on those sockets and cleans-up as they occur. - */ -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; - - DBG("[thread] Manage application started"); - - rcu_register_thread(); - rcu_thread_online(); - - health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_MANAGE); - - if (testpoint(sessiond_thread_manage_apps)) { - goto error_testpoint; - } - - health_code_update(); - - ret = sessiond_set_thread_pollset(&events, 2); - if (ret < 0) { - goto error_poll_create; - } - - ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); - if (ret < 0) { - goto error; - } - - if (testpoint(sessiond_thread_manage_apps_before_loop)) { - goto error; - } - - health_code_update(); - - while (1) { - DBG("Apps thread polling"); - - /* Inifinite blocking call, waiting for transmission */ - restart: - health_poll_entry(); - ret = lttng_poll_wait(&events, -1); - DBG("Apps thread return from poll on %d fds", - LTTNG_POLL_GETNB(&events)); - health_poll_exit(); - if (ret < 0) { - /* - * Restart interrupted system call. - */ - if (errno == EINTR) { - goto restart; - } - goto error; - } - - 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); - - health_code_update(); - - if (!revents) { - /* No activity for this FD (poll implementation). */ - continue; - } - - /* Thread quit pipe has been closed. Killing thread. */ - ret = sessiond_check_thread_quit_pipe(pollfd, revents); - if (ret) { - err = 0; - goto exit; - } - - /* Inspect the apps cmd pipe */ - if (pollfd == apps_cmd_pipe[0]) { - if (revents & LPOLLIN) { - int sock; - - /* Empty pipe */ - 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(); - - /* - * Since this is a command socket (write then read), - * we only monitor the error events of the socket. - */ - ret = lttng_poll_add(&events, sock, - LPOLLERR | LPOLLHUP | LPOLLRDHUP); - if (ret < 0) { - goto error; - } - - DBG("Apps with sock %d added to poll set", sock); - } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - ERR("Apps command pipe error"); - goto error; - } else { - ERR("Unknown poll events %u for sock %d", revents, pollfd); - goto error; - } - } else { - /* - * At this point, we know that a registered application made - * the event at poll_wait. - */ - if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { - /* Removing from the poll set */ - ret = lttng_poll_del(&events, pollfd); - if (ret < 0) { - goto error; - } - - /* Socket closed on remote end. */ - ust_app_unregister(pollfd); - } else { - ERR("Unexpected poll events %u for sock %d", revents, pollfd); - goto error; - } - } - - health_code_update(); - } - } - -exit: -error: - lttng_poll_clean(&events); -error_poll_create: -error_testpoint: - utils_close_pipe(apps_cmd_pipe); - apps_cmd_pipe[0] = apps_cmd_pipe[1] = -1; - - /* - * We don't clean the UST app hash table here since already registered - * applications can still be controlled so let them be until the session - * daemon dies or the applications stop. - */ - - if (err) { - health_error(); - ERR("Health error occurred in %s", __func__); - } - health_unregister(health_sessiond); - DBG("Application communication apps thread cleanup complete"); - rcu_thread_offline(); - rcu_unregister_thread(); - return NULL; -} - -/* - * Setup necessary data for kernel tracer action. - */ -static int init_kernel_tracer(void) -{ - int ret; - - /* Modprobe lttng kernel modules */ - ret = modprobe_lttng_control(); - if (ret < 0) { - goto error; - } - - /* Open debugfs lttng */ - kernel_tracer_fd = open(module_proc_lttng, O_RDWR); - if (kernel_tracer_fd < 0) { - DBG("Failed to open %s", module_proc_lttng); - goto error_open; - } - - /* Validate kernel version */ - ret = kernel_validate_version(kernel_tracer_fd, &kernel_tracer_version, - &kernel_tracer_abi_version); - if (ret < 0) { - goto error_version; - } - - ret = modprobe_lttng_data(); - if (ret < 0) { - goto error_modules; - } - - ret = kernel_supports_ring_buffer_snapshot_sample_positions( - kernel_tracer_fd); - if (ret < 0) { - goto error_modules; - } - - if (ret < 1) { - WARN("Kernel tracer does not support buffer monitoring. " - "The monitoring timer of channels in the kernel domain " - "will be set to 0 (disabled)."); - } - - DBG("Kernel tracer fd %d", kernel_tracer_fd); - return 0; - -error_version: - modprobe_remove_lttng_control(); - ret = close(kernel_tracer_fd); - if (ret) { - PERROR("close"); - } - kernel_tracer_fd = -1; - return LTTNG_ERR_KERN_VERSION; - -error_modules: - ret = close(kernel_tracer_fd); - if (ret) { - PERROR("close"); - } - -error_open: - modprobe_remove_lttng_control(); - -error: - WARN("No kernel tracer available"); - kernel_tracer_fd = -1; - if (!is_root) { - return LTTNG_ERR_NEED_ROOT_SESSIOND; - } else { - return LTTNG_ERR_KERN_NA; - } -} - static int string_match(const char *str1, const char *str2) { return (str1 && str2) && !strcmp(str1, str2); @@ -1371,8 +426,7 @@ static int set_option(int opt, const char *arg, const char *optname) } exit(ret ? EXIT_FAILURE : EXIT_SUCCESS); } else if (string_match(optname, "version") || opt == 'V') { - fprintf(stdout, "%s\n", VERSION); - exit(EXIT_SUCCESS); + opt_print_version = 1; } else if (string_match(optname, "sig-parent") || opt == 'S') { config.sig_parent = true; } else if (string_match(optname, "kconsumerd-err-sock")) { @@ -1605,7 +659,7 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-l, --load"); } else { - config_string_set(&config.load_session_path, strdup(arg)); + config_string_set(&config.load_session_path, strdup(arg)); if (!config.load_session_path.value) { PERROR("strdup"); ret = -ENOMEM; @@ -1727,6 +781,10 @@ end: return ret; } +static void print_version(void) { + fprintf(stdout, "%s\n", VERSION); +} + /* * daemon configuration loading and argument parsing */ @@ -1813,7 +871,7 @@ end: */ static int create_lockfile(void) { - return utils_create_lock_file(config.lock_file_path.value); + return utils_create_lock_file(config.lock_file_path.value); } /* @@ -1872,7 +930,11 @@ static int set_permissions(char *rundir) int ret; gid_t gid; - gid = utils_get_group_id(config.tracing_group_name.value); + ret = utils_get_group_id(config.tracing_group_name.value, true, &gid); + if (ret) { + /* Default to root group. */ + gid = 0; + } /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -1983,7 +1045,16 @@ static int set_consumer_sockets(struct consumer_data *consumer_data) goto error; } if (is_root) { - ret = chown(path, 0, utils_get_group_id(config.tracing_group_name.value)); + gid_t gid; + + ret = utils_get_group_id(config.tracing_group_name.value, true, + &gid); + if (ret) { + /* Default to root group. */ + gid = 0; + } + + ret = chown(path, 0, gid); if (ret < 0) { ERR("Unable to set group on %s", path); PERROR("chown"); @@ -2114,7 +1185,7 @@ static void set_ulimit(void) static int write_pidfile(void) { - return utils_create_pid_file(getpid(), config.pid_file_path.value); + return utils_create_pid_file(getpid(), config.pid_file_path.value); } static int set_clock_plugin_env(void) @@ -2126,7 +1197,7 @@ static int set_clock_plugin_env(void) goto end; } - ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s", + ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s", config.lttng_ust_clock_plugin.value); if (ret < 0) { PERROR("asprintf"); @@ -2170,8 +1241,9 @@ static void destroy_all_sessions_and_wait(void) if (session->destroyed) { goto unlock_session; } - (void) cmd_destroy_session(session, - notification_thread_handle); + (void) cmd_stop_trace(session); + (void) cmd_destroy_session(session, notification_thread_handle, + NULL); unlock_session: session_unlock(session); session_put(session); @@ -2184,13 +1256,97 @@ static void destroy_all_sessions_and_wait(void) DBG("Destruction of all sessions completed"); } +static void unregister_all_triggers(void) +{ + enum lttng_error_code ret_code; + enum lttng_trigger_status trigger_status; + struct lttng_triggers *triggers = NULL; + unsigned int trigger_count, i; + const struct lttng_credentials creds = { + .uid = LTTNG_OPTIONAL_INIT_VALUE(0), + }; + + DBG("Unregistering all triggers"); + + /* + * List all triggers as "root" since we wish to unregister all triggers. + */ + ret_code = notification_thread_command_list_triggers( + notification_thread_handle, creds.uid.value, &triggers); + if (ret_code != LTTNG_OK) { + ERR("Failed to list triggers while unregistering all triggers"); + goto end; + } + + trigger_status = lttng_triggers_get_count(triggers, &trigger_count); + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + + for (i = 0; i < trigger_count; i++) { + enum lttng_error_code ret_code; + uid_t trigger_owner; + const char *trigger_name; + const struct lttng_trigger *trigger = + lttng_triggers_get_at_index(triggers, i); + + assert(trigger); + + trigger_status = lttng_trigger_get_owner_uid( + trigger, &trigger_owner); + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + + trigger_status = lttng_trigger_get_name(trigger, &trigger_name); + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + + DBG("Unregistering trigger: trigger owner uid = %d, trigger name = '%s'", + (int) trigger_owner, trigger_name); + + ret_code = cmd_unregister_trigger( + &creds, trigger, notification_thread_handle); + if (ret_code != LTTNG_OK) { + ERR("Failed to unregister trigger: trigger owner uid = %d, trigger name = '%s', error: '%s'", + (int) trigger_owner, trigger_name, + lttng_strerror(-ret_code)); + /* Continue to unregister the remaining triggers. */ + } + } +end: + lttng_triggers_destroy(triggers); +} + +static int run_as_worker_post_fork_cleanup(void *data) +{ + struct sessiond_config *sessiond_config = data; + + sessiond_config_fini(sessiond_config); + return 0; +} + +static int launch_run_as_worker(const char *procname) +{ + /* + * Clean-up before forking the run-as worker. Any dynamically + * allocated memory of which the worker is not aware will + * be leaked as the process forks a run-as worker (and performs + * no exec*()). The same would apply to any opened fd. + */ + return run_as_create_worker(procname, run_as_worker_post_fork_cleanup, + &config); +} + +static void sessiond_uuid_log(void) +{ + char uuid_str[LTTNG_UUID_STR_LEN]; + + lttng_uuid_to_str(sessiond_uuid, uuid_str); + DBG("Starting lttng-sessiond {%s}", uuid_str); +} + /* * main */ int main(int argc, char **argv) { int ret = 0, retval = 0; - void *status; const char *env_app_timeout; struct lttng_pipe *ust32_channel_monitor_pipe = NULL, *ust64_channel_monitor_pipe = NULL, @@ -2202,7 +1358,10 @@ int main(int argc, char **argv) /* Queue of rotation jobs populated by the sessiond-timer. */ struct rotation_thread_timer_queue *rotation_timer_queue = NULL; struct lttng_thread *client_thread = NULL; + struct lttng_thread *notification_thread = NULL; + struct lttng_thread *register_apps_thread = NULL; + logger_set_thread_name("Main", false); init_kernel_workarounds(); rcu_register_thread(); @@ -2279,6 +1438,13 @@ int main(int argc, char **argv) set_clock_plugin_env(); sessiond_config_log(&config); + sessiond_uuid_log(); + + if (opt_print_version) { + print_version(); + retval = 0; + goto exit_options; + } if (create_lttng_rundir()) { retval = -1; @@ -2316,7 +1482,7 @@ int main(int argc, char **argv) } } - if (run_as_create_worker(argv[0]) < 0) { + if (launch_run_as_worker(argv[0]) < 0) { goto exit_create_run_as_worker_cleanup; } @@ -2333,20 +1499,20 @@ int main(int argc, char **argv) if (!health_sessiond) { PERROR("health_app_create error"); retval = -1; - goto exit_health_sessiond_cleanup; + goto stop_threads; } /* Create thread to clean up RCU hash tables */ ht_cleanup_thread = launch_ht_cleanup_thread(); if (!ht_cleanup_thread) { retval = -1; - goto exit_ht_cleanup; + goto stop_threads; } /* Create thread quit pipe */ if (sessiond_init_thread_quit_pipe()) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* Check if daemon is UID = 0 */ @@ -2358,14 +1524,14 @@ int main(int argc, char **argv) if (!kernel_channel_monitor_pipe) { ERR("Failed to create kernel consumer channel monitor pipe"); retval = -1; - goto exit_init_data; + goto stop_threads; } kconsumer_data.channel_monitor_pipe = lttng_pipe_release_writefd( kernel_channel_monitor_pipe); if (kconsumer_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } } @@ -2377,13 +1543,13 @@ int main(int argc, char **argv) if (!ust32_channel_monitor_pipe) { ERR("Failed to create 32-bit user space consumer channel monitor pipe"); retval = -1; - goto exit_init_data; + goto stop_threads; } ustconsumer32_data.channel_monitor_pipe = lttng_pipe_release_writefd( ust32_channel_monitor_pipe); if (ustconsumer32_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* @@ -2394,7 +1560,7 @@ int main(int argc, char **argv) rotation_timer_queue = rotation_thread_timer_queue_create(); if (!rotation_timer_queue) { retval = -1; - goto exit_init_data; + goto stop_threads; } timer_thread_parameters.rotation_thread_job_queue = rotation_timer_queue; @@ -2403,13 +1569,13 @@ int main(int argc, char **argv) if (!ust64_channel_monitor_pipe) { ERR("Failed to create 64-bit user space consumer channel monitor pipe"); retval = -1; - goto exit_init_data; + goto stop_threads; } ustconsumer64_data.channel_monitor_pipe = lttng_pipe_release_writefd( ust64_channel_monitor_pipe); if (ustconsumer64_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* @@ -2419,7 +1585,7 @@ int main(int argc, char **argv) if (ust_app_ht_alloc()) { ERR("Failed to allocate UST app hash table"); retval = -1; - goto exit_init_data; + goto stop_threads; } /* @@ -2429,9 +1595,14 @@ int main(int argc, char **argv) if (agent_app_ht_alloc()) { ERR("Failed to allocate Agent app hash table"); retval = -1; - goto exit_init_data; + goto stop_threads; } + if (agent_by_event_notifier_domain_ht_create()) { + ERR("Failed to allocate per-event notifier domain agent hash table"); + retval = -1; + goto stop_threads; + } /* * These actions must be executed as root. We do that *after* setting up * the sockets path because we MUST make the check for another daemon using @@ -2441,20 +1612,12 @@ int main(int argc, char **argv) if (is_root) { if (set_consumer_sockets(&kconsumer_data)) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* Setup kernel tracer */ if (!config.no_kernel) { init_kernel_tracer(); - if (kernel_tracer_fd >= 0) { - ret = syscall_init_table(); - if (ret < 0) { - ERR("Unable to populate syscall table. " - "Syscall tracing won't work " - "for this session daemon."); - } - } } /* Set ulimit for open files */ @@ -2465,18 +1628,12 @@ int main(int argc, char **argv) if (set_consumer_sockets(&ustconsumer64_data)) { retval = -1; - goto exit_init_data; + goto stop_threads; } if (set_consumer_sockets(&ustconsumer32_data)) { retval = -1; - goto exit_init_data; - } - - /* Set credentials to socket */ - if (is_root && set_permissions(config.rundir.value)) { - retval = -1; - goto exit_init_data; + goto stop_threads; } /* Get parent pid if -S, --sig-parent is specified. */ @@ -2488,20 +1645,20 @@ int main(int argc, char **argv) if (is_root && !config.no_kernel) { if (utils_create_pipe_cloexec(kernel_poll_pipe)) { retval = -1; - goto exit_init_data; + goto stop_threads; } } /* Setup the thread apps communication pipe. */ if (utils_create_pipe_cloexec(apps_cmd_pipe)) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* Setup the thread apps notify communication pipe. */ if (utils_create_pipe_cloexec(apps_cmd_notify_pipe)) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* Initialize global buffer per UID and PID registry. */ @@ -2525,7 +1682,7 @@ int main(int argc, char **argv) if (ret) { ERR("Error in write_pidfile"); retval = -1; - goto exit_init_data; + goto stop_threads; } /* Initialize communication library */ @@ -2533,16 +1690,10 @@ int main(int argc, char **argv) /* Initialize TCP timeout values */ lttcomm_inet_init(); - if (load_session_init_data(&load_info) < 0) { - retval = -1; - goto exit_init_data; - } - load_info->path = config.load_session_path.value; - /* Create health-check thread. */ if (!launch_health_management_thread()) { retval = -1; - goto exit_health; + goto stop_threads; } /* notification_thread_data acquires the pipes' read side. */ @@ -2553,19 +1704,21 @@ int main(int argc, char **argv) if (!notification_thread_handle) { retval = -1; ERR("Failed to create notification thread shared data"); - goto exit_notification; + goto stop_threads; } /* Create notification thread. */ - if (!launch_notification_thread(notification_thread_handle)) { + notification_thread = launch_notification_thread( + notification_thread_handle); + if (!notification_thread) { retval = -1; - goto exit_notification; + goto stop_threads; } /* Create timer thread. */ if (!launch_timer_thread(&timer_thread_parameters)) { retval = -1; - goto exit_notification; + goto stop_threads; } /* rotation_thread_data acquires the pipes' read side. */ @@ -2576,164 +1729,125 @@ int main(int argc, char **argv) retval = -1; ERR("Failed to create rotation thread shared data"); stop_threads(); - goto exit_rotation; + goto stop_threads; } /* Create rotation thread. */ if (!launch_rotation_thread(rotation_thread_handle)) { retval = -1; - goto exit_rotation; + goto stop_threads; } /* Create thread to manage the client socket */ client_thread = launch_client_thread(); if (!client_thread) { retval = -1; - goto exit_client; + goto stop_threads; + } + + /* Set credentials of the client socket and rundir */ + if (is_root && set_permissions(config.rundir.value)) { + retval = -1; + goto stop_threads; } if (!launch_ust_dispatch_thread(&ust_cmd_queue, apps_cmd_pipe[1], apps_cmd_notify_pipe[1])) { retval = -1; - goto exit_dispatch; + goto stop_threads; } /* Create thread to manage application registration. */ - if (!launch_application_registration_thread(&ust_cmd_queue)) { + register_apps_thread = launch_application_registration_thread( + &ust_cmd_queue); + if (!register_apps_thread) { retval = -1; - goto exit_reg_apps; + goto stop_threads; } /* Create thread to manage application socket */ - ret = pthread_create(&apps_thread, default_pthread_attr(), - thread_manage_apps, (void *) NULL); - if (ret) { - errno = ret; - PERROR("pthread_create apps"); + if (!launch_application_management_thread(apps_cmd_pipe[0])) { retval = -1; - stop_threads(); - goto exit_apps; + goto stop_threads; } /* Create thread to manage application notify socket */ - ret = pthread_create(&apps_notify_thread, default_pthread_attr(), - ust_thread_manage_notify, (void *) NULL); - if (ret) { - errno = ret; - PERROR("pthread_create notify"); + if (!launch_application_notification_thread(apps_cmd_notify_pipe[0])) { retval = -1; - stop_threads(); - goto exit_apps_notify; + goto stop_threads; } - /* Create agent registration thread. */ - ret = pthread_create(&agent_reg_thread, default_pthread_attr(), - agent_thread_manage_registration, (void *) NULL); - if (ret) { - errno = ret; - PERROR("pthread_create agent"); + /* Create agent management thread. */ + if (!launch_agent_management_thread()) { retval = -1; - stop_threads(); - goto exit_agent_reg; + goto stop_threads; } /* Don't start this thread if kernel tracing is not requested nor root */ if (is_root && !config.no_kernel) { /* Create kernel thread to manage kernel event */ - ret = pthread_create(&kernel_thread, default_pthread_attr(), - thread_manage_kernel, (void *) NULL); - if (ret) { - errno = ret; - PERROR("pthread_create kernel"); + if (!launch_kernel_management_thread(kernel_poll_pipe[0])) { retval = -1; - stop_threads(); - goto exit_kernel; + goto stop_threads; + } + + if (kernel_get_notification_fd() >= 0) { + ret = notification_thread_command_add_tracer_event_source( + notification_thread_handle, + kernel_get_notification_fd(), + LTTNG_DOMAIN_KERNEL); + if (ret != LTTNG_OK) { + ERR("Failed to add kernel trigger event source to notification thread"); + retval = -1; + goto stop_threads; + } } } - /* Create session loading thread. */ - ret = pthread_create(&load_session_thread, default_pthread_attr(), - thread_load_session, load_info); + /* Load sessions. */ + ret = config_load_session(config.load_session_path.value, + NULL, 1, 1, NULL); if (ret) { - errno = ret; - PERROR("pthread_create load_session_thread"); + ERR("Session load failed: %s", error_get_str(ret)); retval = -1; - stop_threads(); - goto exit_load_session; + goto stop_threads; } + /* Initialization completed. */ + sessiond_signal_parents(); + /* * This is where we start awaiting program completion (e.g. through * signal that asks threads to teardown). */ - ret = pthread_join(load_session_thread, &status); - if (ret) { - errno = ret; - PERROR("pthread_join load_session_thread"); - retval = -1; - } - /* Initiate teardown once activity occurs on the quit pipe. */ - sessiond_wait_for_quit_pipe(-1U); + sessiond_wait_for_quit_pipe(-1); +stop_threads: /* * Ensure that the client thread is no longer accepting new commands, * which could cause new sessions to be created. */ - if (!lttng_thread_shutdown(client_thread)) { - ERR("Failed to shutdown the client thread, continuing teardown"); + if (client_thread) { + lttng_thread_shutdown(client_thread); lttng_thread_put(client_thread); - client_thread = NULL; } destroy_all_sessions_and_wait(); -exit_load_session: - - if (is_root && !config.no_kernel) { - ret = pthread_join(kernel_thread, &status); - if (ret) { - errno = ret; - PERROR("pthread_join"); - retval = -1; - } - } -exit_kernel: - ret = pthread_join(agent_reg_thread, &status); - if (ret) { - errno = ret; - PERROR("pthread_join agent"); - retval = -1; - } -exit_agent_reg: - - ret = pthread_join(apps_notify_thread, &status); - if (ret) { - errno = ret; - PERROR("pthread_join apps notify"); - retval = -1; - } -exit_apps_notify: + /* + * At this point no new trigger can be registered (no sessions are + * running/rotating) and clients can't connect to the session daemon + * anymore. Unregister all triggers. + */ + unregister_all_triggers(); - ret = pthread_join(apps_thread, &status); - if (ret) { - errno = ret; - PERROR("pthread_join apps"); - retval = -1; + if (register_apps_thread) { + lttng_thread_shutdown(register_apps_thread); + lttng_thread_put(register_apps_thread); } -exit_apps: -exit_reg_apps: -exit_dispatch: -exit_client: -exit_rotation: -exit_notification: lttng_thread_list_shutdown_orphans(); -exit_health: -exit_init_data: - if (client_thread) { - lttng_thread_put(client_thread); - } /* * Wait for all pending call_rcu work to complete before tearing @@ -2748,6 +1862,18 @@ exit_init_data: rcu_thread_online(); sessiond_cleanup(); + /* + * Wait for all pending call_rcu work to complete tearing shutting down + * the notification thread. This call_rcu work includes shutting down + * UST apps and event notifier pipes. + */ + rcu_barrier(); + + if (notification_thread) { + lttng_thread_shutdown(notification_thread); + lttng_thread_put(notification_thread); + } + /* * Ensure all prior call_rcu are done. call_rcu callbacks may push * hash tables to the ht_cleanup thread. Therefore, we ensure that @@ -2783,12 +1909,11 @@ exit_init_data: lttng_pipe_destroy(ust32_channel_monitor_pipe); lttng_pipe_destroy(ust64_channel_monitor_pipe); lttng_pipe_destroy(kernel_channel_monitor_pipe); -exit_ht_cleanup: - health_app_destroy(health_sessiond); -exit_health_sessiond_cleanup: + if (health_sessiond) { + health_app_destroy(health_sessiond); + } exit_create_run_as_worker_cleanup: - exit_options: sessiond_cleanup_lock_file(); sessiond_cleanup_options();