X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fmain.c;h=bca33a4f8a0f669d6e4f278ce15a9c65c444478d;hp=56d932da9f7a961a2274d63d7ee32b6f8ce877ed;hb=49bac6bf673c3a156f27313d875c89df9717dd2f;hpb=8a7e45909a46d4aab3c3debde8f5ce2006c7d306 diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c index 56d932da9..bca33a4f8 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 @@ -59,24 +49,21 @@ #include "consumer.h" #include "context.h" #include "event.h" +#include "event-notifier-error-accounting.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" @@ -86,6 +73,8 @@ #include "dispatch.h" #include "register.h" #include "manage-apps.h" +#include "manage-kernel.h" +#include "modprobe.h" static const char *help_msg = #ifdef LTTNG_EMBED_HELP @@ -95,8 +84,11 @@ NULL #endif ; +#define EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX 65535 + 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; @@ -131,6 +123,7 @@ static const struct option long_options[] = { { "load", required_argument, 0, 'l' }, { "kmod-probes", required_argument, 0, '\0' }, { "extra-kmod-probes", required_argument, 0, '\0' }, + { "event-notifier-error-number-of-bucket", required_argument, 0, '\0' }, { NULL, 0, 0, 0 } }; @@ -144,10 +137,6 @@ static const char *config_ignore_options[] = { "help", "version", "config" }; static int apps_cmd_pipe[2] = { -1, -1 }; static int apps_cmd_notify_pipe[2] = { -1, -1 }; -/* Pthread, Mutexes and Semaphores */ -static pthread_t kernel_thread; -static pthread_t load_session_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 @@ -159,11 +148,6 @@ 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. */ @@ -194,56 +178,56 @@ static void close_consumer_sockets(void) { int ret; - if (kconsumer_data.err_sock >= 0) { - ret = close(kconsumer_data.err_sock); + if (the_kconsumer_data.err_sock >= 0) { + ret = close(the_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 (the_ustconsumer32_data.err_sock >= 0) { + ret = close(the_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 (the_ustconsumer64_data.err_sock >= 0) { + ret = close(the_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 (the_kconsumer_data.cmd_sock >= 0) { + ret = close(the_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 (the_ustconsumer32_data.cmd_sock >= 0) { + ret = close(the_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 (the_ustconsumer64_data.cmd_sock >= 0) { + ret = close(the_ustconsumer64_data.cmd_sock); if (ret < 0) { PERROR("UST consumerd64 cmd_sock close"); } } - if (kconsumer_data.channel_monitor_pipe >= 0) { - ret = close(kconsumer_data.channel_monitor_pipe); + if (the_kconsumer_data.channel_monitor_pipe >= 0) { + ret = close(the_kconsumer_data.channel_monitor_pipe); if (ret < 0) { PERROR("kernel consumer channel monitor pipe close"); } } - if (ustconsumer32_data.channel_monitor_pipe >= 0) { - ret = close(ustconsumer32_data.channel_monitor_pipe); + if (the_ustconsumer32_data.channel_monitor_pipe >= 0) { + ret = close(the_ustconsumer32_data.channel_monitor_pipe); if (ret < 0) { PERROR("UST consumerd32 channel monitor pipe close"); } } - if (ustconsumer64_data.channel_monitor_pipe >= 0) { - ret = close(ustconsumer64_data.channel_monitor_pipe); + if (the_ustconsumer64_data.channel_monitor_pipe >= 0) { + ret = close(the_ustconsumer64_data.channel_monitor_pipe); if (ret < 0) { PERROR("UST consumerd64 channel monitor pipe close"); } @@ -294,74 +278,63 @@ static void sessiond_cleanup(void) sessiond_close_quit_pipe(); utils_close_pipe(apps_cmd_pipe); utils_close_pipe(apps_cmd_notify_pipe); + utils_close_pipe(the_kernel_poll_pipe); - ret = remove(config.pid_file_path.value); + ret = remove(the_config.pid_file_path.value); if (ret < 0) { - PERROR("remove pidfile %s", config.pid_file_path.value); + PERROR("remove pidfile %s", the_config.pid_file_path.value); } DBG("Removing sessiond and consumerd content of directory %s", - config.rundir.value); + the_config.rundir.value); /* sessiond */ - DBG("Removing %s", config.pid_file_path.value); - (void) unlink(config.pid_file_path.value); + DBG("Removing %s", the_config.pid_file_path.value); + (void) unlink(the_config.pid_file_path.value); - DBG("Removing %s", config.agent_port_file_path.value); - (void) unlink(config.agent_port_file_path.value); + DBG("Removing %s", the_config.agent_port_file_path.value); + (void) unlink(the_config.agent_port_file_path.value); /* kconsumerd */ - DBG("Removing %s", kconsumer_data.err_unix_sock_path); - (void) unlink(kconsumer_data.err_unix_sock_path); + DBG("Removing %s", the_kconsumer_data.err_unix_sock_path); + (void) unlink(the_kconsumer_data.err_unix_sock_path); - DBG("Removing directory %s", config.kconsumerd_path.value); - (void) rmdir(config.kconsumerd_path.value); + DBG("Removing directory %s", the_config.kconsumerd_path.value); + (void) rmdir(the_config.kconsumerd_path.value); /* ust consumerd 32 */ - DBG("Removing %s", config.consumerd32_err_unix_sock_path.value); - (void) unlink(config.consumerd32_err_unix_sock_path.value); + DBG("Removing %s", the_config.consumerd32_err_unix_sock_path.value); + (void) unlink(the_config.consumerd32_err_unix_sock_path.value); - DBG("Removing directory %s", config.consumerd32_path.value); - (void) rmdir(config.consumerd32_path.value); + DBG("Removing directory %s", the_config.consumerd32_path.value); + (void) rmdir(the_config.consumerd32_path.value); /* ust consumerd 64 */ - DBG("Removing %s", config.consumerd64_err_unix_sock_path.value); - (void) unlink(config.consumerd64_err_unix_sock_path.value); + DBG("Removing %s", the_config.consumerd64_err_unix_sock_path.value); + (void) unlink(the_config.consumerd64_err_unix_sock_path.value); - DBG("Removing directory %s", config.consumerd64_path.value); - (void) rmdir(config.consumerd64_path.value); + DBG("Removing directory %s", the_config.consumerd64_path.value); + (void) rmdir(the_config.consumerd64_path.value); 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(&the_kconsumer_data); + wait_consumer(&the_ustconsumer64_data); + wait_consumer(&the_ustconsumer32_data); + + if (is_root && !the_config.no_kernel) { + cleanup_kernel_tracer(); } /* @@ -378,754 +351,11 @@ static void sessiond_cleanup_options(void) { DBG("Cleaning up options"); - sessiond_config_fini(&config); + sessiond_config_fini(&the_config); 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; -} - -/* - * 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); @@ -1150,9 +380,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-c, --client-sock"); } else { - config_string_set(&config.client_unix_sock_path, + config_string_set(&the_config.client_unix_sock_path, strdup(arg)); - if (!config.client_unix_sock_path.value) { + if (!the_config.client_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1166,17 +396,17 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-a, --apps-sock"); } else { - config_string_set(&config.apps_unix_sock_path, + config_string_set(&the_config.apps_unix_sock_path, strdup(arg)); - if (!config.apps_unix_sock_path.value) { + if (!the_config.apps_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } } } else if (string_match(optname, "daemonize") || opt == 'd') { - config.daemonize = true; + the_config.daemonize = true; } else if (string_match(optname, "background") || opt == 'b') { - config.background = true; + the_config.background = true; } else if (string_match(optname, "group") || opt == 'g') { if (!arg || *arg == '\0') { ret = -EINVAL; @@ -1186,9 +416,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-g, --group"); } else { - config_string_set(&config.tracing_group_name, + config_string_set(&the_config.tracing_group_name, strdup(arg)); - if (!config.tracing_group_name.value) { + if (!the_config.tracing_group_name.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1201,10 +431,9 @@ 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; + the_config.sig_parent = true; } else if (string_match(optname, "kconsumerd-err-sock")) { if (!arg || *arg == '\0') { ret = -EINVAL; @@ -1214,9 +443,10 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kconsumerd-err-sock"); } else { - config_string_set(&config.kconsumerd_err_unix_sock_path, + config_string_set( + &the_config.kconsumerd_err_unix_sock_path, strdup(arg)); - if (!config.kconsumerd_err_unix_sock_path.value) { + if (!the_config.kconsumerd_err_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1230,9 +460,10 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kconsumerd-cmd-sock"); } else { - config_string_set(&config.kconsumerd_cmd_unix_sock_path, + config_string_set( + &the_config.kconsumerd_cmd_unix_sock_path, strdup(arg)); - if (!config.kconsumerd_cmd_unix_sock_path.value) { + if (!the_config.kconsumerd_cmd_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1246,9 +477,10 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd64-err-sock"); } else { - config_string_set(&config.consumerd64_err_unix_sock_path, + config_string_set( + &the_config.consumerd64_err_unix_sock_path, strdup(arg)); - if (!config.consumerd64_err_unix_sock_path.value) { + if (!the_config.consumerd64_err_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1262,9 +494,10 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd64-cmd-sock"); } else { - config_string_set(&config.consumerd64_cmd_unix_sock_path, + config_string_set( + &the_config.consumerd64_cmd_unix_sock_path, strdup(arg)); - if (!config.consumerd64_cmd_unix_sock_path.value) { + if (!the_config.consumerd64_cmd_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1278,9 +511,10 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd32-err-sock"); } else { - config_string_set(&config.consumerd32_err_unix_sock_path, + config_string_set( + &the_config.consumerd32_err_unix_sock_path, strdup(arg)); - if (!config.consumerd32_err_unix_sock_path.value) { + if (!the_config.consumerd32_err_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } @@ -1294,34 +528,37 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--ustconsumerd32-cmd-sock"); } else { - config_string_set(&config.consumerd32_cmd_unix_sock_path, + config_string_set( + &the_config.consumerd32_cmd_unix_sock_path, strdup(arg)); - if (!config.consumerd32_cmd_unix_sock_path.value) { + if (!the_config.consumerd32_cmd_unix_sock_path.value) { ret = -ENOMEM; PERROR("strdup"); } } } else if (string_match(optname, "no-kernel")) { - config.no_kernel = true; + the_config.no_kernel = true; } else if (string_match(optname, "quiet") || opt == 'q') { - config.quiet = true; + the_config.quiet = true; } else if (string_match(optname, "verbose") || opt == 'v') { /* Verbose level can increase using multiple -v */ if (arg) { /* Value obtained from config file */ - config.verbose = config_parse_value(arg); + the_config.verbose = config_parse_value(arg); } else { /* -v used on command line */ - config.verbose++; + the_config.verbose++; } /* Clamp value to [0, 3] */ - config.verbose = config.verbose < 0 ? 0 : - (config.verbose <= 3 ? config.verbose : 3); + the_config.verbose = the_config.verbose < 0 ? + 0 : + (the_config.verbose <= 3 ? the_config.verbose : + 3); } else if (string_match(optname, "verbose-consumer")) { if (arg) { - config.verbose_consumer = config_parse_value(arg); + the_config.verbose_consumer = config_parse_value(arg); } else { - config.verbose_consumer++; + the_config.verbose_consumer++; } } else if (string_match(optname, "consumerd32-path")) { if (!arg || *arg == '\0') { @@ -1332,9 +569,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd32-path"); } else { - config_string_set(&config.consumerd32_bin_path, + config_string_set(&the_config.consumerd32_bin_path, strdup(arg)); - if (!config.consumerd32_bin_path.value) { + if (!the_config.consumerd32_bin_path.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1348,9 +585,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd32-libdir"); } else { - config_string_set(&config.consumerd32_lib_dir, + config_string_set(&the_config.consumerd32_lib_dir, strdup(arg)); - if (!config.consumerd32_lib_dir.value) { + if (!the_config.consumerd32_lib_dir.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1364,9 +601,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd64-path"); } else { - config_string_set(&config.consumerd64_bin_path, + config_string_set(&the_config.consumerd64_bin_path, strdup(arg)); - if (!config.consumerd64_bin_path.value) { + if (!the_config.consumerd64_bin_path.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1380,9 +617,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--consumerd64-libdir"); } else { - config_string_set(&config.consumerd64_lib_dir, + config_string_set(&the_config.consumerd64_lib_dir, strdup(arg)); - if (!config.consumerd64_lib_dir.value) { + if (!the_config.consumerd64_lib_dir.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1396,8 +633,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "-p, --pidfile"); } else { - config_string_set(&config.pid_file_path, strdup(arg)); - if (!config.pid_file_path.value) { + config_string_set( + &the_config.pid_file_path, strdup(arg)); + if (!the_config.pid_file_path.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1423,7 +661,8 @@ static int set_option(int opt, const char *arg, const char *optname) ERR("Port overflow in --agent-tcp-port parameter: %s", arg); return -1; } - config.agent_tcp_port.begin = config.agent_tcp_port.end = (int) v; + the_config.agent_tcp_port.begin = + the_config.agent_tcp_port.end = (int) v; DBG3("Agent TCP port set to non default: %i", (int) v); } } else if (string_match(optname, "load") || opt == 'l') { @@ -1435,8 +674,9 @@ 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)); - if (!config.load_session_path.value) { + config_string_set(&the_config.load_session_path, + strdup(arg)); + if (!the_config.load_session_path.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1450,8 +690,9 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--kmod-probes"); } else { - config_string_set(&config.kmod_probes_list, strdup(arg)); - if (!config.kmod_probes_list.value) { + config_string_set(&the_config.kmod_probes_list, + strdup(arg)); + if (!the_config.kmod_probes_list.value) { PERROR("strdup"); ret = -ENOMEM; } @@ -1465,13 +706,30 @@ static int set_option(int opt, const char *arg, const char *optname) WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.", "--extra-kmod-probes"); } else { - config_string_set(&config.kmod_extra_probes_list, + config_string_set(&the_config.kmod_extra_probes_list, strdup(arg)); - if (!config.kmod_extra_probes_list.value) { + if (!the_config.kmod_extra_probes_list.value) { PERROR("strdup"); ret = -ENOMEM; } } + } else if (string_match(optname, "event-notifier-error-number-of-bucket")) { + unsigned long v; + + errno = 0; + v = strtoul(arg, NULL, 0); + if (errno != 0 || !isdigit(arg[0])) { + ERR("Wrong value in --event-notifier-error-number-of-bucket parameter: %s", arg); + return -1; + } + if (v == 0 || v >= EVENT_NOTIFIER_ERROR_COUNTER_NUMBER_OF_BUCKET_MAX) { + ERR("Value out of range for --event-notifier-error-number-of-bucket parameter: %s", arg); + return -1; + } + the_config.event_notifier_error_counter_bucket = (int) v; + DBG3("Number of event notifier error counter set to non default: %i", + the_config.event_notifier_error_counter_bucket); + goto end; } else if (string_match(optname, "config") || opt == 'f') { /* This is handled in set_options() thus silent skip. */ goto end; @@ -1557,6 +815,10 @@ end: return ret; } +static void print_version(void) { + fprintf(stdout, "%s\n", VERSION); +} + /* * daemon configuration loading and argument parsing */ @@ -1643,7 +905,7 @@ end: */ static int create_lockfile(void) { - return utils_create_lock_file(config.lock_file_path.value); + return utils_create_lock_file(the_config.lock_file_path.value); } /* @@ -1680,7 +942,7 @@ static void sessiond_cleanup_lock_file(void) * release the file system lock. */ if (lockfile_fd >= 0) { - ret = remove(config.lock_file_path.value); + ret = remove(the_config.lock_file_path.value); if (ret < 0) { PERROR("remove lock file"); } @@ -1702,7 +964,12 @@ 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( + the_config.tracing_group_name.value, true, &gid); + if (ret) { + /* Default to root group. */ + gid = 0; + } /* Set lttng run dir */ ret = chown(rundir, 0, gid); @@ -1723,30 +990,34 @@ static int set_permissions(char *rundir) } /* lttng client socket path */ - ret = chown(config.client_unix_sock_path.value, 0, gid); + ret = chown(the_config.client_unix_sock_path.value, 0, gid); if (ret < 0) { - ERR("Unable to set group on %s", config.client_unix_sock_path.value); + ERR("Unable to set group on %s", + the_config.client_unix_sock_path.value); PERROR("chown"); } /* kconsumer error socket path */ - ret = chown(kconsumer_data.err_unix_sock_path, 0, 0); + ret = chown(the_kconsumer_data.err_unix_sock_path, 0, 0); if (ret < 0) { - ERR("Unable to set group on %s", kconsumer_data.err_unix_sock_path); + ERR("Unable to set group on %s", + the_kconsumer_data.err_unix_sock_path); PERROR("chown"); } /* 64-bit ustconsumer error socket path */ - ret = chown(ustconsumer64_data.err_unix_sock_path, 0, 0); + ret = chown(the_ustconsumer64_data.err_unix_sock_path, 0, 0); if (ret < 0) { - ERR("Unable to set group on %s", ustconsumer64_data.err_unix_sock_path); + ERR("Unable to set group on %s", + the_ustconsumer64_data.err_unix_sock_path); PERROR("chown"); } /* 32-bit ustconsumer compat32 error socket path */ - ret = chown(ustconsumer32_data.err_unix_sock_path, 0, 0); + ret = chown(the_ustconsumer32_data.err_unix_sock_path, 0, 0); if (ret < 0) { - ERR("Unable to set group on %s", ustconsumer32_data.err_unix_sock_path); + ERR("Unable to set group on %s", + the_ustconsumer32_data.err_unix_sock_path); PERROR("chown"); } @@ -1762,12 +1033,12 @@ static int create_lttng_rundir(void) { int ret; - DBG3("Creating LTTng run directory: %s", config.rundir.value); + DBG3("Creating LTTng run directory: %s", the_config.rundir.value); - ret = mkdir(config.rundir.value, S_IRWXU); + ret = mkdir(the_config.rundir.value, S_IRWXU); if (ret < 0) { if (errno != EEXIST) { - ERR("Unable to create %s", config.rundir.value); + ERR("Unable to create %s", the_config.rundir.value); goto error; } else { ret = 0; @@ -1789,13 +1060,13 @@ static int set_consumer_sockets(struct consumer_data *consumer_data) switch (consumer_data->type) { case LTTNG_CONSUMER_KERNEL: - path = config.kconsumerd_path.value; + path = the_config.kconsumerd_path.value; break; case LTTNG_CONSUMER64_UST: - path = config.consumerd64_path.value; + path = the_config.consumerd64_path.value; break; case LTTNG_CONSUMER32_UST: - path = config.consumerd32_path.value; + path = the_config.consumerd32_path.value; break; default: ERR("Consumer type unknown"); @@ -1813,7 +1084,16 @@ 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(the_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"); @@ -1944,7 +1224,7 @@ 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(), the_config.pid_file_path.value); } static int set_clock_plugin_env(void) @@ -1952,12 +1232,12 @@ static int set_clock_plugin_env(void) int ret = 0; char *env_value = NULL; - if (!config.lttng_ust_clock_plugin.value) { + if (!the_config.lttng_ust_clock_plugin.value) { goto end; } - ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s", - config.lttng_ust_clock_plugin.value); + ret = asprintf(&env_value, "LTTNG_UST_CLOCK_PLUGIN=%s", + the_config.lttng_ust_clock_plugin.value); if (ret < 0) { PERROR("asprintf"); goto end; @@ -1971,7 +1251,7 @@ static int set_clock_plugin_env(void) } DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"", - config.lttng_ust_clock_plugin.value); + the_config.lttng_ust_clock_plugin.value); end: return ret; } @@ -2000,8 +1280,9 @@ 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, the_notification_thread_handle, NULL); unlock_session: session_unlock(session); session_put(session); @@ -2014,13 +1295,97 @@ 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( + the_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++) { + 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, + the_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, &the_config); +} + +static void sessiond_uuid_log(void) +{ + char uuid_str[LTTNG_UUID_STR_LEN]; + + lttng_uuid_to_str(the_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, @@ -2032,7 +1397,10 @@ 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(); @@ -2047,14 +1415,14 @@ int main(int argc, char **argv) goto exit_set_signal_handler; } - page_size = sysconf(_SC_PAGESIZE); - if (page_size < 0) { + the_page_size = sysconf(_SC_PAGESIZE); + if (the_page_size < 0) { PERROR("sysconf _SC_PAGESIZE"); - page_size = LONG_MAX; - WARN("Fallback page size to %ld", page_size); + the_page_size = LONG_MAX; + WARN("Fallback page size to %ld", the_page_size); } - ret = sessiond_config_init(&config); + ret = sessiond_config_init(&the_config); if (ret) { retval = -1; goto exit_set_signal_handler; @@ -2064,7 +1432,7 @@ int main(int argc, char **argv) * Init config from environment variables. * Command line option override env configuration per-doc. Do env first. */ - sessiond_config_apply_env_config(&config); + sessiond_config_apply_env_config(&the_config); /* * Parse arguments and load the daemon configuration file. @@ -2086,29 +1454,36 @@ int main(int argc, char **argv) * since daemonizing causes the sessiond's current working directory * to '/'. */ - ret = sessiond_config_resolve_paths(&config); + ret = sessiond_config_resolve_paths(&the_config); if (ret) { goto exit_options; } /* Apply config. */ - lttng_opt_verbose = config.verbose; - lttng_opt_quiet = config.quiet; - kconsumer_data.err_unix_sock_path = - config.kconsumerd_err_unix_sock_path.value; - kconsumer_data.cmd_unix_sock_path = - config.kconsumerd_cmd_unix_sock_path.value; - ustconsumer32_data.err_unix_sock_path = - config.consumerd32_err_unix_sock_path.value; - ustconsumer32_data.cmd_unix_sock_path = - config.consumerd32_cmd_unix_sock_path.value; - ustconsumer64_data.err_unix_sock_path = - config.consumerd64_err_unix_sock_path.value; - ustconsumer64_data.cmd_unix_sock_path = - config.consumerd64_cmd_unix_sock_path.value; + lttng_opt_verbose = the_config.verbose; + lttng_opt_quiet = the_config.quiet; + the_kconsumer_data.err_unix_sock_path = + the_config.kconsumerd_err_unix_sock_path.value; + the_kconsumer_data.cmd_unix_sock_path = + the_config.kconsumerd_cmd_unix_sock_path.value; + the_ustconsumer32_data.err_unix_sock_path = + the_config.consumerd32_err_unix_sock_path.value; + the_ustconsumer32_data.cmd_unix_sock_path = + the_config.consumerd32_cmd_unix_sock_path.value; + the_ustconsumer64_data.err_unix_sock_path = + the_config.consumerd64_err_unix_sock_path.value; + the_ustconsumer64_data.cmd_unix_sock_path = + the_config.consumerd64_cmd_unix_sock_path.value; set_clock_plugin_env(); - sessiond_config_log(&config); + sessiond_config_log(&the_config); + sessiond_uuid_log(); + + if (opt_print_version) { + print_version(); + retval = 0; + goto exit_options; + } if (create_lttng_rundir()) { retval = -1; @@ -2123,11 +1498,11 @@ int main(int argc, char **argv) } /* Daemonize */ - if (config.daemonize || config.background) { + if (the_config.daemonize || the_config.background) { int i; - ret = lttng_daemonize(&child_ppid, &recv_child_signal, - !config.background); + ret = lttng_daemonize(&the_child_ppid, &recv_child_signal, + !the_config.background); if (ret < 0) { retval = -1; goto exit_options; @@ -2146,7 +1521,7 @@ 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; } @@ -2159,24 +1534,24 @@ int main(int argc, char **argv) * 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) { + the_health_sessiond = health_app_create(NR_HEALTH_SESSIOND_TYPES); + if (!the_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 */ @@ -2188,32 +1563,32 @@ 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 = + the_kconsumer_data.channel_monitor_pipe = lttng_pipe_release_writefd( - kernel_channel_monitor_pipe); - if (kconsumer_data.channel_monitor_pipe < 0) { + kernel_channel_monitor_pipe); + if (the_kconsumer_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } } /* Set consumer initial state */ - kernel_consumerd_state = CONSUMER_STOPPED; - ust_consumerd_state = CONSUMER_STOPPED; + the_kernel_consumerd_state = CONSUMER_STOPPED; + the_ust_consumerd_state = CONSUMER_STOPPED; ust32_channel_monitor_pipe = lttng_pipe_open(0); 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) { + the_ustconsumer32_data.channel_monitor_pipe = + lttng_pipe_release_writefd(ust32_channel_monitor_pipe); + if (the_ustconsumer32_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* @@ -2224,7 +1599,7 @@ 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; @@ -2233,13 +1608,13 @@ 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) { + the_ustconsumer64_data.channel_monitor_pipe = + lttng_pipe_release_writefd(ust64_channel_monitor_pipe); + if (the_ustconsumer64_data.channel_monitor_pipe < 0) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* @@ -2249,9 +1624,11 @@ 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; } + event_notifier_error_accounting_init(the_config.event_notifier_error_counter_bucket); + /* * Initialize agent app hash table. We allocate the hash table here * since cleanup() can get called after this point. @@ -2259,9 +1636,14 @@ 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 @@ -2269,22 +1651,14 @@ int main(int argc, char **argv) * kernel tracer. */ if (is_root) { - if (set_consumer_sockets(&kconsumer_data)) { + if (set_consumer_sockets(&the_kconsumer_data)) { retval = -1; - goto exit_init_data; + goto stop_threads; } /* Setup kernel tracer */ - if (!config.no_kernel) { + if (!the_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 */ @@ -2293,45 +1667,39 @@ int main(int argc, char **argv) /* init lttng_fd tracking must be done after set_ulimit. */ lttng_fd_init(); - if (set_consumer_sockets(&ustconsumer64_data)) { + if (set_consumer_sockets(&the_ustconsumer64_data)) { retval = -1; - goto exit_init_data; + goto stop_threads; } - if (set_consumer_sockets(&ustconsumer32_data)) { + if (set_consumer_sockets(&the_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. */ - if (config.sig_parent) { - ppid = getppid(); + if (the_config.sig_parent) { + the_ppid = getppid(); } /* Setup the kernel pipe for waking up the kernel thread */ - if (is_root && !config.no_kernel) { - if (utils_create_pipe_cloexec(kernel_poll_pipe)) { + if (is_root && !the_config.no_kernel) { + if (utils_create_pipe_cloexec(the_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. */ @@ -2346,16 +1714,16 @@ int main(int argc, char **argv) /* Check for the application socket timeout env variable. */ env_app_timeout = getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV); if (env_app_timeout) { - config.app_socket_timeout = atoi(env_app_timeout); + the_config.app_socket_timeout = atoi(env_app_timeout); } else { - config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT; + the_config.app_socket_timeout = DEFAULT_APP_SOCKET_RW_TIMEOUT; } ret = write_pidfile(); if (ret) { ERR("Error in write_pidfile"); retval = -1; - goto exit_init_data; + goto stop_threads; } /* Initialize communication library */ @@ -2363,171 +1731,163 @@ 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. */ - notification_thread_handle = notification_thread_handle_create( - ust32_channel_monitor_pipe, - ust64_channel_monitor_pipe, + the_notification_thread_handle = notification_thread_handle_create( + ust32_channel_monitor_pipe, ust64_channel_monitor_pipe, kernel_channel_monitor_pipe); - if (!notification_thread_handle) { + if (!the_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( + the_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. */ rotation_thread_handle = rotation_thread_handle_create( - rotation_timer_queue, - notification_thread_handle); + rotation_timer_queue, the_notification_thread_handle); if (!rotation_thread_handle) { 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(the_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 */ if (!launch_application_management_thread(apps_cmd_pipe[0])) { retval = -1; - goto exit_apps; + goto stop_threads; } /* Create thread to manage application notify socket */ if (!launch_application_notification_thread(apps_cmd_notify_pipe[0])) { retval = -1; - goto exit_apps_notify; + goto stop_threads; } /* Create agent management thread. */ if (!launch_agent_management_thread()) { retval = -1; - 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) { + if (is_root && !the_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(the_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( + the_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( + the_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; - } + /* + * 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(); + + if (register_apps_thread) { + lttng_thread_shutdown(register_apps_thread); + lttng_thread_put(register_apps_thread); } -exit_kernel: -exit_agent_reg: -exit_apps_notify: -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 @@ -2542,6 +1902,39 @@ exit_init_data: rcu_thread_online(); sessiond_cleanup(); + /* + * Wait for all pending call_rcu work to complete before 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); + } + + /* + * Error accounting teardown has to be done after the teardown of all + * event notifier pipes to ensure that no tracer may try to use the + * error accounting facilities. + */ + event_notifier_error_accounting_fini(); + + /* + * Unloading the kernel modules needs to be done after all kernel + * ressources have been released. In our case, this includes the + * notification fd, the event notifier group fd, error accounting fd, + * all event and event notifier fds, etc. + * + * In short, at this point, we need to have called close() on all fds + * received from the kernel tracer. + */ + if (is_root && !the_config.no_kernel) { + DBG("Unloading kernel modules"); + modprobe_remove_lttng_all(); + } + /* * Ensure all prior call_rcu are done. call_rcu callbacks may push * hash tables to the ht_cleanup thread. Therefore, we ensure that @@ -2571,18 +1964,18 @@ exit_init_data: * session daemon's teardown in order to allow it to be notified * of the active session and channels at the moment of the teardown. */ - if (notification_thread_handle) { - notification_thread_handle_destroy(notification_thread_handle); + if (the_notification_thread_handle) { + notification_thread_handle_destroy( + the_notification_thread_handle); } 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 (the_health_sessiond) { + health_app_destroy(the_health_sessiond); + } exit_create_run_as_worker_cleanup: - exit_options: sessiond_cleanup_lock_file(); sessiond_cleanup_options();