Rename sessiond-timer.[hc] to timer.[hc]

[lttng-tools.git] / src / bin / lttng-sessiond / main.c

/*
 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
 *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *               2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
 *
 * 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.
 *
 * 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
#include <getopt.h>
#include <grp.h>
#include <limits.h>
#include <paths.h>
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <inttypes.h>
#include <sys/mman.h>
#include <sys/mount.h>
#include <sys/resource.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <urcu/uatomic.h>
#include <unistd.h>
#include <ctype.h>

#include <common/common.h>
#include <common/compat/socket.h>
#include <common/compat/getenv.h>
#include <common/defaults.h>
#include <common/kernel-consumer/kernel-consumer.h>
#include <common/futex.h>
#include <common/relayd/relayd.h>
#include <common/utils.h>
#include <common/daemonize.h>
#include <common/config/session-config.h>
#include <common/dynamic-buffer.h>
#include <lttng/userspace-probe-internal.h>
#include <lttng/event-internal.h>

#include "lttng-sessiond.h"
#include "buffer-registry.h"
#include "channel.h"
#include "cmd.h"
#include "consumer.h"
#include "context.h"
#include "event.h"
#include "kernel.h"
#include "kernel-consumer.h"
#include "modprobe.h"
#include "shm.h"
#include "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 "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"
#include "timer.h"

static const char *help_msg =
#ifdef LTTNG_EMBED_HELP
#include <lttng-sessiond.8.h>
#else
NULL
#endif
;

const char *progname;
static pid_t ppid;          /* Parent PID for --sig-parent option */
static pid_t child_ppid;    /* Internal parent PID use with daemonize. */
static int lockfile_fd = -1;

/* Set to 1 when a SIGUSR1 signal is received. */
static int recv_child_signal;

static struct lttng_kernel_tracer_version kernel_tracer_version;
static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version;

/*
 * Consumer daemon specific control data. Every value not initialized here is
 * set to 0 by the static definition.
 */
static struct consumer_data kconsumer_data = {
        .type = LTTNG_CONSUMER_KERNEL,
        .err_sock = -1,
        .cmd_sock = -1,
        .channel_monitor_pipe = -1,
        .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
        .lock = PTHREAD_MUTEX_INITIALIZER,
        .cond = PTHREAD_COND_INITIALIZER,
        .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer64_data = {
        .type = LTTNG_CONSUMER64_UST,
        .err_sock = -1,
        .cmd_sock = -1,
        .channel_monitor_pipe = -1,
        .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
        .lock = PTHREAD_MUTEX_INITIALIZER,
        .cond = PTHREAD_COND_INITIALIZER,
        .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};
static struct consumer_data ustconsumer32_data = {
        .type = LTTNG_CONSUMER32_UST,
        .err_sock = -1,
        .cmd_sock = -1,
        .channel_monitor_pipe = -1,
        .pid_mutex = PTHREAD_MUTEX_INITIALIZER,
        .lock = PTHREAD_MUTEX_INITIALIZER,
        .cond = PTHREAD_COND_INITIALIZER,
        .cond_mutex = PTHREAD_MUTEX_INITIALIZER,
};

/* Command line options */
static const struct option long_options[] = {
        { "client-sock", required_argument, 0, 'c' },
        { "apps-sock", required_argument, 0, 'a' },
        { "kconsumerd-cmd-sock", required_argument, 0, '\0' },
        { "kconsumerd-err-sock", required_argument, 0, '\0' },
        { "ustconsumerd32-cmd-sock", required_argument, 0, '\0' },
        { "ustconsumerd32-err-sock", required_argument, 0, '\0' },
        { "ustconsumerd64-cmd-sock", required_argument, 0, '\0' },
        { "ustconsumerd64-err-sock", required_argument, 0, '\0' },
        { "consumerd32-path", required_argument, 0, '\0' },
        { "consumerd32-libdir", required_argument, 0, '\0' },
        { "consumerd64-path", required_argument, 0, '\0' },
        { "consumerd64-libdir", required_argument, 0, '\0' },
        { "daemonize", no_argument, 0, 'd' },
        { "background", no_argument, 0, 'b' },
        { "sig-parent", no_argument, 0, 'S' },
        { "help", no_argument, 0, 'h' },
        { "group", required_argument, 0, 'g' },
        { "version", no_argument, 0, 'V' },
        { "quiet", no_argument, 0, 'q' },
        { "verbose", no_argument, 0, 'v' },
        { "verbose-consumer", no_argument, 0, '\0' },
        { "no-kernel", no_argument, 0, '\0' },
        { "pidfile", required_argument, 0, 'p' },
        { "agent-tcp-port", required_argument, 0, '\0' },
        { "config", required_argument, 0, 'f' },
        { "load", required_argument, 0, 'l' },
        { "kmod-probes", required_argument, 0, '\0' },
        { "extra-kmod-probes", required_argument, 0, '\0' },
        { NULL, 0, 0, 0 }
};

struct sessiond_config config;

/* Command line options to ignore from configuration file */
static const char *config_ignore_options[] = { "help", "version", "config" };

/* Shared between threads */
static int dispatch_thread_exit;

/* Sockets and FDs */
static int client_sock = -1;
static int apps_sock = -1;
int kernel_tracer_fd = -1;
static int kernel_poll_pipe[2] = { -1, -1 };

/*
 * Quit pipe for all threads. This permits a single cancellation point
 * for all threads when receiving an event on the pipe.
 */
static int thread_quit_pipe[2] = { -1, -1 };

/*
 * This pipe is used to inform the thread managing application communication
 * that a command is queued and ready to be processed.
 */
static int apps_cmd_pipe[2] = { -1, -1 };

int apps_cmd_notify_pipe[2] = { -1, -1 };

/* Pthread, Mutexes and Semaphores */
static pthread_t apps_thread;
static pthread_t apps_notify_thread;
static pthread_t reg_apps_thread;
static pthread_t client_thread;
static pthread_t kernel_thread;
static pthread_t dispatch_thread;
static pthread_t health_thread;
static pthread_t ht_cleanup_thread;
static pthread_t agent_reg_thread;
static pthread_t load_session_thread;
static pthread_t notification_thread;
static pthread_t rotation_thread;
static pthread_t timer_thread;

/*
 * UST registration command queue. This queue is tied with a futex and uses a N
 * wakers / 1 waiter implemented and detailed in futex.c/.h
 *
 * The thread_registration_apps and thread_dispatch_ust_registration uses this
 * queue along with the wait/wake scheme. The thread_manage_apps receives down
 * the line new application socket and monitors it for any I/O error or clean
 * close that triggers an unregistration of the application.
 */
static struct ust_cmd_queue ust_cmd_queue;

/*
 * Pointer initialized before thread creation.
 *
 * This points to the tracing session list containing the session count and a
 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
 * MUST NOT be taken if you call a public function in session.c.
 *
 * The lock is nested inside the structure: session_list_ptr->lock. Please use
 * session_lock_list and session_unlock_list for lock acquisition.
 */
static struct ltt_session_list *session_list_ptr;

int ust_consumerd64_fd = -1;
int ust_consumerd32_fd = -1;

static const char *module_proc_lttng = "/proc/lttng";

/*
 * Consumer daemon state which is changed when spawning it, killing it or in
 * case of a fatal error.
 */
enum consumerd_state {
        CONSUMER_STARTED = 1,
        CONSUMER_STOPPED = 2,
        CONSUMER_ERROR   = 3,
};

/*
 * This consumer daemon state is used to validate if a client command will be
 * able to reach the consumer. If not, the client is informed. For instance,
 * doing a "lttng start" when the consumer state is set to ERROR will return an
 * error to the client.
 *
 * The following example shows a possible race condition of this scheme:
 *
 * consumer thread error happens
 *                                    client cmd arrives
 *                                    client cmd checks state -> still OK
 * consumer thread exit, sets error
 *                                    client cmd try to talk to consumer
 *                                    ...
 *
 * However, since the consumer is a different daemon, we have no way of making
 * sure the command will reach it safely even with this state flag. This is why
 * we consider that up to the state validation during command processing, the
 * command is safe. After that, we can not guarantee the correctness of the
 * client request vis-a-vis the consumer.
 */
static enum consumerd_state ust_consumerd_state;
static enum consumerd_state kernel_consumerd_state;

/* Set in main() with the current page size. */
long page_size;

/* Application health monitoring */
struct health_app *health_sessiond;

/* Am I root or not. */
int is_root;                    /* Set to 1 if the daemon is running as root */

const char * const config_section_name = "sessiond";

/* Load session thread information to operate. */
struct load_session_thread_data *load_info;

/* Notification thread handle. */
struct notification_thread_handle *notification_thread_handle;

/* Rotation thread handle. */
struct rotation_thread_handle *rotation_thread_handle;

/* Global hash tables */
struct lttng_ht *agent_apps_ht_by_sock = NULL;

/*
 * The initialization of the session daemon is done in multiple phases.
 *
 * While all threads are launched near-simultaneously, only some of them
 * are needed to ensure the session daemon can start to respond to client
 * requests.
 *
 * There are two important guarantees that we wish to offer with respect
 * to the initialisation of the session daemon:
 *   - When the daemonize/background launcher process exits, the sessiond
 *     is fully able to respond to client requests,
 *   - Auto-loaded sessions are visible to clients.
 *
 * In order to achieve this, a number of support threads have to be launched
 * to allow the "client" thread to function properly. Moreover, since the
 * "load session" thread needs the client thread, we must provide a way
 * for the "load session" thread to know that the "client" thread is up
 * and running.
 *
 * Hence, the support threads decrement the lttng_sessiond_ready counter
 * while the "client" threads waits for it to reach 0. Once the "client" thread
 * unblocks, it posts the message_thread_ready semaphore which allows the
 * "load session" thread to progress.
 *
 * This implies that the "load session" thread is the last to be initialized
 * and will explicitly call sessiond_signal_parents(), which signals the parents
 * that the session daemon is fully initialized.
 *
 * The four (4) support threads are:
 *  - agent_thread
 *  - notification_thread
 *  - rotation_thread
 *  - health_thread
 */
#define NR_LTTNG_SESSIOND_SUPPORT_THREADS 4
int lttng_sessiond_ready = NR_LTTNG_SESSIOND_SUPPORT_THREADS;

int sessiond_check_thread_quit_pipe(int fd, uint32_t events)
{
        return (fd == thread_quit_pipe[0] && (events & LPOLLIN)) ? 1 : 0;
}

/* Notify parents that we are ready for cmd and health check */
LTTNG_HIDDEN
void sessiond_signal_parents(void)
{
        /*
         * Notify parent pid that we are ready to accept command
         * for client side.  This ppid is the one from the
         * external process that spawned us.
         */
        if (config.sig_parent) {
                kill(ppid, SIGUSR1);
        }

        /*
         * Notify the parent of the fork() process that we are
         * ready.
         */
        if (config.daemonize || config.background) {
                kill(child_ppid, SIGUSR1);
        }
}

LTTNG_HIDDEN
void sessiond_notify_ready(void)
{
        /*
         * This memory barrier is paired with the one performed by
         * the client thread after it has seen that 'lttng_sessiond_ready' is 0.
         *
         * The purpose of these memory barriers is to ensure that all
         * initialization operations of the various threads that call this
         * function to signal that they are ready are commited/published
         * before the client thread can see the 'lttng_sessiond_ready' counter
         * reach 0.
         *
         * Note that this could be a 'write' memory barrier, but a full barrier
         * is used in case the code using this utility changes. The performance
         * implications of this choice are minimal since this is a slow path.
         */
        cmm_smp_mb();
        uatomic_sub(&lttng_sessiond_ready, 1);
}

static
int __sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size,
                int *a_pipe)
{
        int ret;

        assert(events);

        ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
        if (ret < 0) {
                goto error;
        }

        /* Add quit pipe */
        ret = lttng_poll_add(events, a_pipe[0], LPOLLIN | LPOLLERR);
        if (ret < 0) {
                goto error;
        }

        return 0;

error:
        return ret;
}

/*
 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 */
int sessiond_set_thread_pollset(struct lttng_poll_event *events, size_t size)
{
        return __sessiond_set_thread_pollset(events, size, thread_quit_pipe);
}

/*
 * Init thread quit pipe.
 *
 * Return -1 on error or 0 if all pipes are created.
 */
static int __init_thread_quit_pipe(int *a_pipe)
{
        int ret, i;

        ret = pipe(a_pipe);
        if (ret < 0) {
                PERROR("thread quit pipe");
                goto error;
        }

        for (i = 0; i < 2; i++) {
                ret = fcntl(a_pipe[i], F_SETFD, FD_CLOEXEC);
                if (ret < 0) {
                        PERROR("fcntl");
                        goto error;
                }
        }

error:
        return ret;
}

static int init_thread_quit_pipe(void)
{
        return __init_thread_quit_pipe(thread_quit_pipe);
}

/*
 * Stop all threads by closing the thread quit pipe.
 */
static void stop_threads(void)
{
        int ret;

        /* Stopping all threads */
        DBG("Terminating all threads");
        ret = notify_thread_pipe(thread_quit_pipe[1]);
        if (ret < 0) {
                ERR("write error on thread quit pipe");
        }

        /* Dispatch thread */
        CMM_STORE_SHARED(dispatch_thread_exit, 1);
        futex_nto1_wake(&ust_cmd_queue.futex);
}

/*
 * Close every consumer sockets.
 */
static void close_consumer_sockets(void)
{
        int ret;

        if (kconsumer_data.err_sock >= 0) {
                ret = close(kconsumer_data.err_sock);
                if (ret < 0) {
                        PERROR("kernel consumer err_sock close");
                }
        }
        if (ustconsumer32_data.err_sock >= 0) {
                ret = close(ustconsumer32_data.err_sock);
                if (ret < 0) {
                        PERROR("UST consumerd32 err_sock close");
                }
        }
        if (ustconsumer64_data.err_sock >= 0) {
                ret = close(ustconsumer64_data.err_sock);
                if (ret < 0) {
                        PERROR("UST consumerd64 err_sock close");
                }
        }
        if (kconsumer_data.cmd_sock >= 0) {
                ret = close(kconsumer_data.cmd_sock);
                if (ret < 0) {
                        PERROR("kernel consumer cmd_sock close");
                }
        }
        if (ustconsumer32_data.cmd_sock >= 0) {
                ret = close(ustconsumer32_data.cmd_sock);
                if (ret < 0) {
                        PERROR("UST consumerd32 cmd_sock close");
                }
        }
        if (ustconsumer64_data.cmd_sock >= 0) {
                ret = close(ustconsumer64_data.cmd_sock);
                if (ret < 0) {
                        PERROR("UST consumerd64 cmd_sock close");
                }
        }
        if (kconsumer_data.channel_monitor_pipe >= 0) {
                ret = close(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 (ret < 0) {
                        PERROR("UST consumerd32 channel monitor pipe close");
                }
        }
        if (ustconsumer64_data.channel_monitor_pipe >= 0) {
                ret = close(ustconsumer64_data.channel_monitor_pipe);
                if (ret < 0) {
                        PERROR("UST consumerd64 channel monitor pipe close");
                }
        }
}

/*
 * Wait on consumer process termination.
 *
 * Need to be called with the consumer data lock held or from a context
 * ensuring no concurrent access to data (e.g: cleanup).
 */
static void wait_consumer(struct consumer_data *consumer_data)
{
        pid_t ret;
        int status;

        if (consumer_data->pid <= 0) {
                return;
        }

        DBG("Waiting for complete teardown of consumerd (PID: %d)",
                        consumer_data->pid);
        ret = waitpid(consumer_data->pid, &status, 0);
        if (ret == -1) {
                PERROR("consumerd waitpid pid: %d", consumer_data->pid)
        } else  if (!WIFEXITED(status)) {
                ERR("consumerd termination with error: %d",
                                WEXITSTATUS(ret));
        }
        consumer_data->pid = 0;
}

/*
 * Cleanup the session daemon's data structures.
 */
static void sessiond_cleanup(void)
{
        int ret;
        struct ltt_session *sess, *stmp;

        DBG("Cleanup sessiond");

        /*
         * Close the thread quit pipe. It has already done its job,
         * since we are now called.
         */
        utils_close_pipe(thread_quit_pipe);

        ret = remove(config.pid_file_path.value);
        if (ret < 0) {
                PERROR("remove pidfile %s", config.pid_file_path.value);
        }

        DBG("Removing sessiond and consumerd content of directory %s",
                config.rundir.value);

        /* sessiond */
        DBG("Removing %s", config.pid_file_path.value);
        (void) unlink(config.pid_file_path.value);

        DBG("Removing %s", config.agent_port_file_path.value);
        (void) unlink(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 directory %s", config.kconsumerd_path.value);
        (void) rmdir(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 directory %s", config.consumerd32_path.value);
        (void) rmdir(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 directory %s", config.consumerd64_path.value);
        (void) rmdir(config.consumerd64_path.value);

        DBG("Cleaning up all sessions");

        /* Destroy session list mutex */
        if (session_list_ptr != NULL) {
                pthread_mutex_destroy(&session_list_ptr->lock);

                /* Cleanup ALL session */
                cds_list_for_each_entry_safe(sess, stmp,
                                &session_list_ptr->head, list) {
                        cmd_destroy_session(sess, kernel_poll_pipe[1],
                                        notification_thread_handle);
                }
        }

        wait_consumer(&kconsumer_data);
        wait_consumer(&ustconsumer64_data);
        wait_consumer(&ustconsumer32_data);

        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);
        }

        /*
         * We do NOT rmdir rundir because there are other processes
         * using it, for instance lttng-relayd, which can start in
         * parallel with this teardown.
         */
}

/*
 * Cleanup the daemon's option data structures.
 */
static void sessiond_cleanup_options(void)
{
        DBG("Cleaning up options");

        sessiond_config_fini(&config);

        run_as_destroy_worker();
}

/*
 * Send data on a unix socket using the liblttsessiondcomm API.
 *
 * Return lttcomm error code.
 */
static int send_unix_sock(int sock, void *buf, size_t len)
{
        /* Check valid length */
        if (len == 0) {
                return -1;
        }

        return lttcomm_send_unix_sock(sock, buf, len);
}

/*
 * Free memory of a command context structure.
 */
static void clean_command_ctx(struct command_ctx **cmd_ctx)
{
        DBG("Clean command context structure");
        if (*cmd_ctx) {
                if ((*cmd_ctx)->llm) {
                        free((*cmd_ctx)->llm);
                }
                if ((*cmd_ctx)->lsm) {
                        free((*cmd_ctx)->lsm);
                }
                free(*cmd_ctx);
                *cmd_ctx = NULL;
        }
}

/*
 * Notify UST applications using the shm mmap futex.
 */
static int notify_ust_apps(int active)
{
        char *wait_shm_mmap;

        DBG("Notifying applications of session daemon state: %d", active);

        /* See shm.c for this call implying mmap, shm and futex calls */
        wait_shm_mmap = shm_ust_get_mmap(config.wait_shm_path.value, is_root);
        if (wait_shm_mmap == NULL) {
                goto error;
        }

        /* Wake waiting process */
        futex_wait_update((int32_t *) wait_shm_mmap, active);

        /* Apps notified successfully */
        return 0;

error:
        return -1;
}

/*
 * Setup the outgoing data buffer for the response (llm) by allocating the
 * right amount of memory and copying the original information from the lsm
 * structure.
 *
 * Return 0 on success, negative value on error.
 */
static int setup_lttng_msg(struct command_ctx *cmd_ctx,
        const void *payload_buf, size_t payload_len,
        const void *cmd_header_buf, size_t cmd_header_len)
{
        int ret = 0;
        const size_t header_len = sizeof(struct lttcomm_lttng_msg);
        const size_t cmd_header_offset = header_len;
        const size_t payload_offset = cmd_header_offset + cmd_header_len;
        const size_t total_msg_size = header_len + cmd_header_len + payload_len;

        cmd_ctx->llm = zmalloc(total_msg_size);

        if (cmd_ctx->llm == NULL) {
                PERROR("zmalloc");
                ret = -ENOMEM;
                goto end;
        }

        /* Copy common data */
        cmd_ctx->llm->cmd_type = cmd_ctx->lsm->cmd_type;
        cmd_ctx->llm->pid = cmd_ctx->lsm->domain.attr.pid;
        cmd_ctx->llm->cmd_header_size = cmd_header_len;
        cmd_ctx->llm->data_size = payload_len;
        cmd_ctx->lttng_msg_size = total_msg_size;

        /* Copy command header */
        if (cmd_header_len) {
                memcpy(((uint8_t *) cmd_ctx->llm) + cmd_header_offset, cmd_header_buf,
                        cmd_header_len);
        }

        /* Copy payload */
        if (payload_len) {
                memcpy(((uint8_t *) cmd_ctx->llm) + payload_offset, payload_buf,
                        payload_len);
        }

end:
        return ret;
}

/*
 * Version of setup_lttng_msg() without command header.
 */
static int setup_lttng_msg_no_cmd_header(struct command_ctx *cmd_ctx,
        void *payload_buf, size_t payload_len)
{
        return setup_lttng_msg(cmd_ctx, payload_buf, payload_len, NULL, 0);
}
/*
 * 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_session *session;
        struct ltt_kernel_channel *channel;

        DBG("Updating kernel poll set");

        session_lock_list();
        cds_list_for_each_entry(session, &session_list_ptr->head, list) {
                session_lock(session);
                if (session->kernel_session == NULL) {
                        session_unlock(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);
                                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;

        DBG("Updating kernel streams for channel fd %d", fd);

        session_lock_list();
        cds_list_for_each_entry(session, &session_list_ptr->head, list) {
                session_lock(session);
                if (session->kernel_session == NULL) {
                        session_unlock(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_unlock_list();
        return ret;

error:
        session_unlock(session);
        session_unlock_list();
        return ret;
}

/*
 * For each tracing session, update newly registered apps. The session list
 * lock MUST be acquired before calling this.
 */
static void update_ust_app(int app_sock)
{
        struct ltt_session *sess, *stmp;

        /* Consumer is in an ERROR state. Stop any application update. */
        if (uatomic_read(&ust_consumerd_state) == CONSUMER_ERROR) {
                /* Stop the update process since the consumer is dead. */
                return;
        }

        /* For all tracing session(s) */
        cds_list_for_each_entry_safe(sess, stmp, &session_list_ptr->head, list) {
                struct ust_app *app;

                session_lock(sess);
                if (!sess->ust_session) {
                        goto unlock_session;
                }

                rcu_read_lock();
                assert(app_sock >= 0);
                app = ust_app_find_by_sock(app_sock);
                if (app == NULL) {
                        /*
                         * Application can be unregistered before so
                         * this is possible hence simply stopping the
                         * update.
                         */
                        DBG3("UST app update failed to find app sock %d",
                                app_sock);
                        goto unlock_rcu;
                }
                ust_app_global_update(sess->ust_session, app);
        unlock_rcu:
                rcu_read_unlock();
        unlock_session:
                session_unlock(sess);
        }
}

/*
 * 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.
 */
static 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;
}

/*
 * Send a socket to a thread This is called from the dispatch UST registration
 * thread once all sockets are set for the application.
 *
 * The sock value can be invalid, we don't really care, the thread will handle
 * it and make the necessary cleanup if so.
 *
 * On success, return 0 else a negative value being the errno message of the
 * write().
 */
static int send_socket_to_thread(int fd, int sock)
{
        ssize_t ret;

        /*
         * It's possible that the FD is set as invalid with -1 concurrently just
         * before calling this function being a shutdown state of the thread.
         */
        if (fd < 0) {
                ret = -EBADF;
                goto error;
        }

        ret = lttng_write(fd, &sock, sizeof(sock));
        if (ret < sizeof(sock)) {
                PERROR("write apps pipe %d", fd);
                if (ret < 0) {
                        ret = -errno;
                }
                goto error;
        }

        /* All good. Don't send back the write positive ret value. */
        ret = 0;
error:
        return (int) ret;
}

/*
 * Sanitize the wait queue of the dispatch registration thread meaning removing
 * invalid nodes from it. This is to avoid memory leaks for the case the UST
 * notify socket is never received.
 */
static void sanitize_wait_queue(struct ust_reg_wait_queue *wait_queue)
{
        int ret, nb_fd = 0, i;
        unsigned int fd_added = 0;
        struct lttng_poll_event events;
        struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;

        assert(wait_queue);

        lttng_poll_init(&events);

        /* Just skip everything for an empty queue. */
        if (!wait_queue->count) {
                goto end;
        }

        ret = lttng_poll_create(&events, wait_queue->count, LTTNG_CLOEXEC);
        if (ret < 0) {
                goto error_create;
        }

        cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
                        &wait_queue->head, head) {
                assert(wait_node->app);
                ret = lttng_poll_add(&events, wait_node->app->sock,
                                LPOLLHUP | LPOLLERR);
                if (ret < 0) {
                        goto error;
                }

                fd_added = 1;
        }

        if (!fd_added) {
                goto end;
        }

        /*
         * Poll but don't block so we can quickly identify the faulty events and
         * clean them afterwards from the wait queue.
         */
        ret = lttng_poll_wait(&events, 0);
        if (ret < 0) {
                goto error;
        }
        nb_fd = ret;

        for (i = 0; i < nb_fd; i++) {
                /* Get faulty FD. */
                uint32_t revents = LTTNG_POLL_GETEV(&events, i);
                int pollfd = LTTNG_POLL_GETFD(&events, i);

                if (!revents) {
                        /* No activity for this FD (poll implementation). */
                        continue;
                }

                cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
                                &wait_queue->head, head) {
                        if (pollfd == wait_node->app->sock &&
                                        (revents & (LPOLLHUP | LPOLLERR))) {
                                cds_list_del(&wait_node->head);
                                wait_queue->count--;
                                ust_app_destroy(wait_node->app);
                                free(wait_node);
                                /*
                                 * Silence warning of use-after-free in
                                 * cds_list_for_each_entry_safe which uses
                                 * __typeof__(*wait_node).
                                 */
                                wait_node = NULL;
                                break;
                        } else {
                                ERR("Unexpected poll events %u for sock %d", revents, pollfd);
                                goto error;
                        }
                }
        }

        if (nb_fd > 0) {
                DBG("Wait queue sanitized, %d node were cleaned up", nb_fd);
        }

end:
        lttng_poll_clean(&events);
        return;

error:
        lttng_poll_clean(&events);
error_create:
        ERR("Unable to sanitize wait queue");
        return;
}

/*
 * Dispatch request from the registration threads to the application
 * communication thread.
 */
static void *thread_dispatch_ust_registration(void *data)
{
        int ret, err = -1;
        struct cds_wfcq_node *node;
        struct ust_command *ust_cmd = NULL;
        struct ust_reg_wait_node *wait_node = NULL, *tmp_wait_node;
        struct ust_reg_wait_queue wait_queue = {
                .count = 0,
        };

        rcu_register_thread();

        health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH);

        if (testpoint(sessiond_thread_app_reg_dispatch)) {
                goto error_testpoint;
        }

        health_code_update();

        CDS_INIT_LIST_HEAD(&wait_queue.head);

        DBG("[thread] Dispatch UST command started");

        for (;;) {
                health_code_update();

                /* Atomically prepare the queue futex */
                futex_nto1_prepare(&ust_cmd_queue.futex);

                if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
                        break;
                }

                do {
                        struct ust_app *app = NULL;
                        ust_cmd = NULL;

                        /*
                         * Make sure we don't have node(s) that have hung up before receiving
                         * the notify socket. This is to clean the list in order to avoid
                         * memory leaks from notify socket that are never seen.
                         */
                        sanitize_wait_queue(&wait_queue);

                        health_code_update();
                        /* Dequeue command for registration */
                        node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
                        if (node == NULL) {
                                DBG("Woken up but nothing in the UST command queue");
                                /* Continue thread execution */
                                break;
                        }

                        ust_cmd = caa_container_of(node, struct ust_command, node);

                        DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
                                        " gid:%d sock:%d name:%s (version %d.%d)",
                                        ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
                                        ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
                                        ust_cmd->sock, ust_cmd->reg_msg.name,
                                        ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);

                        if (ust_cmd->reg_msg.type == USTCTL_SOCKET_CMD) {
                                wait_node = zmalloc(sizeof(*wait_node));
                                if (!wait_node) {
                                        PERROR("zmalloc wait_node dispatch");
                                        ret = close(ust_cmd->sock);
                                        if (ret < 0) {
                                                PERROR("close ust sock dispatch %d", ust_cmd->sock);
                                        }
                                        lttng_fd_put(LTTNG_FD_APPS, 1);
                                        free(ust_cmd);
                                        goto error;
                                }
                                CDS_INIT_LIST_HEAD(&wait_node->head);

                                /* Create application object if socket is CMD. */
                                wait_node->app = ust_app_create(&ust_cmd->reg_msg,
                                                ust_cmd->sock);
                                if (!wait_node->app) {
                                        ret = close(ust_cmd->sock);
                                        if (ret < 0) {
                                                PERROR("close ust sock dispatch %d", ust_cmd->sock);
                                        }
                                        lttng_fd_put(LTTNG_FD_APPS, 1);
                                        free(wait_node);
                                        free(ust_cmd);
                                        continue;
                                }
                                /*
                                 * Add application to the wait queue so we can set the notify
                                 * socket before putting this object in the global ht.
                                 */
                                cds_list_add(&wait_node->head, &wait_queue.head);
                                wait_queue.count++;

                                free(ust_cmd);
                                /*
                                 * We have to continue here since we don't have the notify
                                 * socket and the application MUST be added to the hash table
                                 * only at that moment.
                                 */
                                continue;
                        } else {
                                /*
                                 * Look for the application in the local wait queue and set the
                                 * notify socket if found.
                                 */
                                cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
                                                &wait_queue.head, head) {
                                        health_code_update();
                                        if (wait_node->app->pid == ust_cmd->reg_msg.pid) {
                                                wait_node->app->notify_sock = ust_cmd->sock;
                                                cds_list_del(&wait_node->head);
                                                wait_queue.count--;
                                                app = wait_node->app;
                                                free(wait_node);
                                                DBG3("UST app notify socket %d is set", ust_cmd->sock);
                                                break;
                                        }
                                }

                                /*
                                 * With no application at this stage the received socket is
                                 * basically useless so close it before we free the cmd data
                                 * structure for good.
                                 */
                                if (!app) {
                                        ret = close(ust_cmd->sock);
                                        if (ret < 0) {
                                                PERROR("close ust sock dispatch %d", ust_cmd->sock);
                                        }
                                        lttng_fd_put(LTTNG_FD_APPS, 1);
                                }
                                free(ust_cmd);
                        }

                        if (app) {
                                /*
                                 * @session_lock_list
                                 *
                                 * Lock the global session list so from the register up to the
                                 * registration done message, no thread can see the application
                                 * and change its state.
                                 */
                                session_lock_list();
                                rcu_read_lock();

                                /*
                                 * Add application to the global hash table. This needs to be
                                 * done before the update to the UST registry can locate the
                                 * application.
                                 */
                                ust_app_add(app);

                                /* Set app version. This call will print an error if needed. */
                                (void) ust_app_version(app);

                                /* Send notify socket through the notify pipe. */
                                ret = send_socket_to_thread(apps_cmd_notify_pipe[1],
                                                app->notify_sock);
                                if (ret < 0) {
                                        rcu_read_unlock();
                                        session_unlock_list();
                                        /*
                                         * No notify thread, stop the UST tracing. However, this is
                                         * not an internal error of the this thread thus setting
                                         * the health error code to a normal exit.
                                         */
                                        err = 0;
                                        goto error;
                                }

                                /*
                                 * Update newly registered application with the tracing
                                 * registry info already enabled information.
                                 */
                                update_ust_app(app->sock);

                                /*
                                 * Don't care about return value. Let the manage apps threads
                                 * handle app unregistration upon socket close.
                                 */
                                (void) ust_app_register_done(app);

                                /*
                                 * Even if the application socket has been closed, send the app
                                 * to the thread and unregistration will take place at that
                                 * place.
                                 */
                                ret = send_socket_to_thread(apps_cmd_pipe[1], app->sock);
                                if (ret < 0) {
                                        rcu_read_unlock();
                                        session_unlock_list();
                                        /*
                                         * No apps. thread, stop the UST tracing. However, this is
                                         * not an internal error of the this thread thus setting
                                         * the health error code to a normal exit.
                                         */
                                        err = 0;
                                        goto error;
                                }

                                rcu_read_unlock();
                                session_unlock_list();
                        }
                } while (node != NULL);

                health_poll_entry();
                /* Futex wait on queue. Blocking call on futex() */
                futex_nto1_wait(&ust_cmd_queue.futex);
                health_poll_exit();
        }
        /* Normal exit, no error */
        err = 0;

error:
        /* Clean up wait queue. */
        cds_list_for_each_entry_safe(wait_node, tmp_wait_node,
                        &wait_queue.head, head) {
                cds_list_del(&wait_node->head);
                wait_queue.count--;
                free(wait_node);
        }

        /* Empty command queue. */
        for (;;) {
                /* Dequeue command for registration */
                node = cds_wfcq_dequeue_blocking(&ust_cmd_queue.head, &ust_cmd_queue.tail);
                if (node == NULL) {
                        break;
                }
                ust_cmd = caa_container_of(node, struct ust_command, node);
                ret = close(ust_cmd->sock);
                if (ret < 0) {
                        PERROR("close ust sock exit dispatch %d", ust_cmd->sock);
                }
                lttng_fd_put(LTTNG_FD_APPS, 1);
                free(ust_cmd);
        }

error_testpoint:
        DBG("Dispatch thread dying");
        if (err) {
                health_error();
                ERR("Health error occurred in %s", __func__);
        }
        health_unregister(health_sessiond);
        rcu_unregister_thread();
        return NULL;
}

/*
 * This thread manage application registration.
 */
static void *thread_registration_apps(void *data)
{
        int sock = -1, i, ret, pollfd, err = -1;
        uint32_t revents, nb_fd;
        struct lttng_poll_event events;
        /*
         * Get allocated in this thread, enqueued to a global queue, dequeued and
         * freed in the manage apps thread.
         */
        struct ust_command *ust_cmd = NULL;

        DBG("[thread] Manage application registration started");

        health_register(health_sessiond, HEALTH_SESSIOND_TYPE_APP_REG);

        if (testpoint(sessiond_thread_registration_apps)) {
                goto error_testpoint;
        }

        ret = lttcomm_listen_unix_sock(apps_sock);
        if (ret < 0) {
                goto error_listen;
        }

        /*
         * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
         * more will be added to this poll set.
         */
        ret = sessiond_set_thread_pollset(&events, 2);
        if (ret < 0) {
                goto error_create_poll;
        }

        /* Add the application registration socket */
        ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP);
        if (ret < 0) {
                goto error_poll_add;
        }

        /* Notify all applications to register */
        ret = notify_ust_apps(1);
        if (ret < 0) {
                ERR("Failed to notify applications or create the wait shared memory.\n"
                        "Execution continues but there might be problem for already\n"
                        "running applications that wishes to register.");
        }

        while (1) {
                DBG("Accepting application registration");

                /* Inifinite blocking call, waiting for transmission */
        restart:
                health_poll_entry();
                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++) {
                        health_code_update();

                        /* Fetch once the poll data */
                        revents = LTTNG_POLL_GETEV(&events, i);
                        pollfd = LTTNG_POLL_GETFD(&events, i);

                        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 == apps_sock) {
                                if (revents & LPOLLIN) {
                                        sock = lttcomm_accept_unix_sock(apps_sock);
                                        if (sock < 0) {
                                                goto error;
                                        }

                                        /*
                                         * Set socket timeout for both receiving and ending.
                                         * app_socket_timeout is in seconds, whereas
                                         * lttcomm_setsockopt_rcv_timeout and
                                         * lttcomm_setsockopt_snd_timeout expect msec as
                                         * parameter.
                                         */
                                        if (config.app_socket_timeout >= 0) {
                                                (void) lttcomm_setsockopt_rcv_timeout(sock,
                                                                config.app_socket_timeout * 1000);
                                                (void) lttcomm_setsockopt_snd_timeout(sock,
                                                                config.app_socket_timeout * 1000);
                                        }

                                        /*
                                         * Set the CLOEXEC flag. Return code is useless because
                                         * either way, the show must go on.
                                         */
                                        (void) utils_set_fd_cloexec(sock);

                                        /* Create UST registration command for enqueuing */
                                        ust_cmd = zmalloc(sizeof(struct ust_command));
                                        if (ust_cmd == NULL) {
                                                PERROR("ust command zmalloc");
                                                ret = close(sock);
                                                if (ret) {
                                                        PERROR("close");
                                                }
                                                goto error;
                                        }

                                        /*
                                         * Using message-based transmissions to ensure we don't
                                         * have to deal with partially received messages.
                                         */
                                        ret = lttng_fd_get(LTTNG_FD_APPS, 1);
                                        if (ret < 0) {
                                                ERR("Exhausted file descriptors allowed for applications.");
                                                free(ust_cmd);
                                                ret = close(sock);
                                                if (ret) {
                                                        PERROR("close");
                                                }
                                                sock = -1;
                                                continue;
                                        }

                                        health_code_update();
                                        ret = ust_app_recv_registration(sock, &ust_cmd->reg_msg);
                                        if (ret < 0) {
                                                free(ust_cmd);
                                                /* Close socket of the application. */
                                                ret = close(sock);
                                                if (ret) {
                                                        PERROR("close");
                                                }
                                                lttng_fd_put(LTTNG_FD_APPS, 1);
                                                sock = -1;
                                                continue;
                                        }
                                        health_code_update();

                                        ust_cmd->sock = sock;
                                        sock = -1;

                                        DBG("UST registration received with pid:%d ppid:%d uid:%d"
                                                        " gid:%d sock:%d name:%s (version %d.%d)",
                                                        ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid,
                                                        ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid,
                                                        ust_cmd->sock, ust_cmd->reg_msg.name,
                                                        ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor);

                                        /*
                                         * Lock free enqueue the registration request. The red pill
                                         * has been taken! This apps will be part of the *system*.
                                         */
                                        cds_wfcq_enqueue(&ust_cmd_queue.head, &ust_cmd_queue.tail, &ust_cmd->node);

                                        /*
                                         * Wake the registration queue futex. Implicit memory
                                         * barrier with the exchange in cds_wfcq_enqueue.
                                         */
                                        futex_nto1_wake(&ust_cmd_queue.futex);
                                } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
                                        ERR("Register apps socket poll error");
                                        goto error;
                                } else {
                                        ERR("Unexpected poll events %u for sock %d", revents, pollfd);
                                        goto error;
                                }
                        }
                }
        }

exit:
error:
        /* Notify that the registration thread is gone */
        notify_ust_apps(0);

        if (apps_sock >= 0) {
                ret = close(apps_sock);
                if (ret) {
                        PERROR("close");
                }
        }
        if (sock >= 0) {
                ret = close(sock);
                if (ret) {
                        PERROR("close");
                }
                lttng_fd_put(LTTNG_FD_APPS, 1);
        }
        unlink(config.apps_unix_sock_path.value);

error_poll_add:
        lttng_poll_clean(&events);
error_listen:
error_create_poll:
error_testpoint:
        DBG("UST Registration thread cleanup complete");
        if (err) {
                health_error();
                ERR("Health error occurred in %s", __func__);
        }
        health_unregister(health_sessiond);

        return NULL;
}

/*
 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
 * exec or it will fails.
 */
static int spawn_consumer_thread(struct consumer_data *consumer_data)
{
        int ret, clock_ret;
        struct timespec timeout;

        /*
         * Make sure we set the readiness flag to 0 because we are NOT ready.
         * This access to consumer_thread_is_ready does not need to be
         * protected by consumer_data.cond_mutex (yet) since the consumer
         * management thread has not been started at this point.
         */
        consumer_data->consumer_thread_is_ready = 0;

        /* Setup pthread condition */
        ret = pthread_condattr_init(&consumer_data->condattr);
        if (ret) {
                errno = ret;
                PERROR("pthread_condattr_init consumer data");
                goto error;
        }

        /*
         * Set the monotonic clock in order to make sure we DO NOT jump in time
         * between the clock_gettime() call and the timedwait call. See bug #324
         * for a more details and how we noticed it.
         */
        ret = pthread_condattr_setclock(&consumer_data->condattr, CLOCK_MONOTONIC);
        if (ret) {
                errno = ret;
                PERROR("pthread_condattr_setclock consumer data");
                goto error;
        }

        ret = pthread_cond_init(&consumer_data->cond, &consumer_data->condattr);
        if (ret) {
                errno = ret;
                PERROR("pthread_cond_init consumer data");
                goto error;
        }

        ret = pthread_create(&consumer_data->thread, default_pthread_attr(),
                        thread_manage_consumer, consumer_data);
        if (ret) {
                errno = ret;
                PERROR("pthread_create consumer");
                ret = -1;
                goto error;
        }

        /* We are about to wait on a pthread condition */
        pthread_mutex_lock(&consumer_data->cond_mutex);

        /* Get time for sem_timedwait absolute timeout */
        clock_ret = lttng_clock_gettime(CLOCK_MONOTONIC, &timeout);
        /*
         * Set the timeout for the condition timed wait even if the clock gettime
         * call fails since we might loop on that call and we want to avoid to
         * increment the timeout too many times.
         */
        timeout.tv_sec += DEFAULT_SEM_WAIT_TIMEOUT;

        /*
         * The following loop COULD be skipped in some conditions so this is why we
         * set ret to 0 in order to make sure at least one round of the loop is
         * done.
         */
        ret = 0;

        /*
         * Loop until the condition is reached or when a timeout is reached. Note
         * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
         * be returned but the pthread_cond(3), from the glibc-doc, says that it is
         * possible. This loop does not take any chances and works with both of
         * them.
         */
        while (!consumer_data->consumer_thread_is_ready && ret != ETIMEDOUT) {
                if (clock_ret < 0) {
                        PERROR("clock_gettime spawn consumer");
                        /* Infinite wait for the consumerd thread to be ready */
                        ret = pthread_cond_wait(&consumer_data->cond,
                                        &consumer_data->cond_mutex);
                } else {
                        ret = pthread_cond_timedwait(&consumer_data->cond,
                                        &consumer_data->cond_mutex, &timeout);
                }
        }

        /* Release the pthread condition */
        pthread_mutex_unlock(&consumer_data->cond_mutex);

        if (ret != 0) {
                errno = ret;
                if (ret == ETIMEDOUT) {
                        int pth_ret;

                        /*
                         * Call has timed out so we kill the kconsumerd_thread and return
                         * an error.
                         */
                        ERR("Condition timed out. The consumer thread was never ready."
                                        " Killing it");
                        pth_ret = pthread_cancel(consumer_data->thread);
                        if (pth_ret < 0) {
                                PERROR("pthread_cancel consumer thread");
                        }
                } else {
                        PERROR("pthread_cond_wait failed consumer thread");
                }
                /* Caller is expecting a negative value on failure. */
                ret = -1;
                goto error;
        }

        pthread_mutex_lock(&consumer_data->pid_mutex);
        if (consumer_data->pid == 0) {
                ERR("Consumerd did not start");
                pthread_mutex_unlock(&consumer_data->pid_mutex);
                goto error;
        }
        pthread_mutex_unlock(&consumer_data->pid_mutex);

        return 0;

error:
        return ret;
}

/*
 * Join consumer thread
 */
static int join_consumer_thread(struct consumer_data *consumer_data)
{
        void *status;

        /* Consumer pid must be a real one. */
        if (consumer_data->pid > 0) {
                int ret;
                ret = kill(consumer_data->pid, SIGTERM);
                if (ret) {
                        PERROR("Error killing consumer daemon");
                        return ret;
                }
                return pthread_join(consumer_data->thread, &status);
        } else {
                return 0;
        }
}

/*
 * Fork and exec a consumer daemon (consumerd).
 *
 * Return pid if successful else -1.
 */
static pid_t spawn_consumerd(struct consumer_data *consumer_data)
{
        int ret;
        pid_t pid;
        const char *consumer_to_use;
        const char *verbosity;
        struct stat st;

        DBG("Spawning consumerd");

        pid = fork();
        if (pid == 0) {
                /*
                 * Exec consumerd.
                 */
                if (config.verbose_consumer) {
                        verbosity = "--verbose";
                } else if (lttng_opt_quiet) {
                        verbosity = "--quiet";
                } else {
                        verbosity = "";
                }

                switch (consumer_data->type) {
                case LTTNG_CONSUMER_KERNEL:
                        /*
                         * Find out which consumerd to execute. We will first try the
                         * 64-bit path, then the sessiond's installation directory, and
                         * fallback on the 32-bit one,
                         */
                        DBG3("Looking for a kernel consumer at these locations:");
                        DBG3("  1) %s", config.consumerd64_bin_path.value ? : "NULL");
                        DBG3("  2) %s/%s", INSTALL_BIN_PATH, DEFAULT_CONSUMERD_FILE);
                        DBG3("  3) %s", config.consumerd32_bin_path.value ? : "NULL");
                        if (stat(config.consumerd64_bin_path.value, &st) == 0) {
                                DBG3("Found location #1");
                                consumer_to_use = config.consumerd64_bin_path.value;
                        } else if (stat(INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE, &st) == 0) {
                                DBG3("Found location #2");
                                consumer_to_use = INSTALL_BIN_PATH "/" DEFAULT_CONSUMERD_FILE;
                        } else if (config.consumerd32_bin_path.value &&
                                        stat(config.consumerd32_bin_path.value, &st) == 0) {
                                DBG3("Found location #3");
                                consumer_to_use = config.consumerd32_bin_path.value;
                        } else {
                                DBG("Could not find any valid consumerd executable");
                                ret = -EINVAL;
                                goto error;
                        }
                        DBG("Using kernel consumer at: %s",  consumer_to_use);
                        (void) execl(consumer_to_use,
                                "lttng-consumerd", verbosity, "-k",
                                "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
                                "--consumerd-err-sock", consumer_data->err_unix_sock_path,
                                "--group", config.tracing_group_name.value,
                                NULL);
                        break;
                case LTTNG_CONSUMER64_UST:
                {
                        if (config.consumerd64_lib_dir.value) {
                                char *tmp;
                                size_t tmplen;
                                char *tmpnew;

                                tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
                                if (!tmp) {
                                        tmp = "";
                                }
                                tmplen = strlen(config.consumerd64_lib_dir.value) + 1 /* : */ + strlen(tmp);
                                tmpnew = zmalloc(tmplen + 1 /* \0 */);
                                if (!tmpnew) {
                                        ret = -ENOMEM;
                                        goto error;
                                }
                                strcat(tmpnew, config.consumerd64_lib_dir.value);
                                if (tmp[0] != '\0') {
                                        strcat(tmpnew, ":");
                                        strcat(tmpnew, tmp);
                                }
                                ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
                                free(tmpnew);
                                if (ret) {
                                        ret = -errno;
                                        goto error;
                                }
                        }
                        DBG("Using 64-bit UST consumer at: %s",  config.consumerd64_bin_path.value);
                        (void) execl(config.consumerd64_bin_path.value, "lttng-consumerd", verbosity, "-u",
                                        "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
                                        "--consumerd-err-sock", consumer_data->err_unix_sock_path,
                                        "--group", config.tracing_group_name.value,
                                        NULL);
                        break;
                }
                case LTTNG_CONSUMER32_UST:
                {
                        if (config.consumerd32_lib_dir.value) {
                                char *tmp;
                                size_t tmplen;
                                char *tmpnew;

                                tmp = lttng_secure_getenv("LD_LIBRARY_PATH");
                                if (!tmp) {
                                        tmp = "";
                                }
                                tmplen = strlen(config.consumerd32_lib_dir.value) + 1 /* : */ + strlen(tmp);
                                tmpnew = zmalloc(tmplen + 1 /* \0 */);
                                if (!tmpnew) {
                                        ret = -ENOMEM;
                                        goto error;
                                }
                                strcat(tmpnew, config.consumerd32_lib_dir.value);
                                if (tmp[0] != '\0') {
                                        strcat(tmpnew, ":");
                                        strcat(tmpnew, tmp);
                                }
                                ret = setenv("LD_LIBRARY_PATH", tmpnew, 1);
                                free(tmpnew);
                                if (ret) {
                                        ret = -errno;
                                        goto error;
                                }
                        }
                        DBG("Using 32-bit UST consumer at: %s",  config.consumerd32_bin_path.value);
                        (void) execl(config.consumerd32_bin_path.value, "lttng-consumerd", verbosity, "-u",
                                        "--consumerd-cmd-sock", consumer_data->cmd_unix_sock_path,
                                        "--consumerd-err-sock", consumer_data->err_unix_sock_path,
                                        "--group", config.tracing_group_name.value,
                                        NULL);
                        break;
                }
                default:
                        ERR("unknown consumer type");
                        errno = 0;
                }
                if (errno != 0) {
                        PERROR("Consumer execl()");
                }
                /* Reaching this point, we got a failure on our execl(). */
                exit(EXIT_FAILURE);
        } else if (pid > 0) {
                ret = pid;
        } else {
                PERROR("start consumer fork");
                ret = -errno;
        }
error:
        return ret;
}

/*
 * Spawn the consumerd daemon and session daemon thread.
 */
static int start_consumerd(struct consumer_data *consumer_data)
{
        int ret;

        /*
         * Set the listen() state on the socket since there is a possible race
         * between the exec() of the consumer daemon and this call if place in the
         * consumer thread. See bug #366 for more details.
         */
        ret = lttcomm_listen_unix_sock(consumer_data->err_sock);
        if (ret < 0) {
                goto error;
        }

        pthread_mutex_lock(&consumer_data->pid_mutex);
        if (consumer_data->pid != 0) {
                pthread_mutex_unlock(&consumer_data->pid_mutex);
                goto end;
        }

        ret = spawn_consumerd(consumer_data);
        if (ret < 0) {
                ERR("Spawning consumerd failed");
                pthread_mutex_unlock(&consumer_data->pid_mutex);
                goto error;
        }

        /* Setting up the consumer_data pid */
        consumer_data->pid = ret;
        DBG2("Consumer pid %d", consumer_data->pid);
        pthread_mutex_unlock(&consumer_data->pid_mutex);

        DBG2("Spawning consumer control thread");
        ret = spawn_consumer_thread(consumer_data);
        if (ret < 0) {
                ERR("Fatal error spawning consumer control thread");
                goto error;
        }

end:
        return 0;

error:
        /* Cleanup already created sockets on error. */
        if (consumer_data->err_sock >= 0) {
                int err;

                err = close(consumer_data->err_sock);
                if (err < 0) {
                        PERROR("close consumer data error socket");
                }
        }
        return ret;
}

/*
 * 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;
        }
}


/*
 * Copy consumer output from the tracing session to the domain session. The
 * function also applies the right modification on a per domain basis for the
 * trace files destination directory.
 *
 * Should *NOT* be called with RCU read-side lock held.
 */
static int copy_session_consumer(int domain, struct ltt_session *session)
{
        int ret;
        const char *dir_name;
        struct consumer_output *consumer;

        assert(session);
        assert(session->consumer);

        switch (domain) {
        case LTTNG_DOMAIN_KERNEL:
                DBG3("Copying tracing session consumer output in kernel session");
                /*
                 * XXX: We should audit the session creation and what this function
                 * does "extra" in order to avoid a destroy since this function is used
                 * in the domain session creation (kernel and ust) only. Same for UST
                 * domain.
                 */
                if (session->kernel_session->consumer) {
                        consumer_output_put(session->kernel_session->consumer);
                }
                session->kernel_session->consumer =
                        consumer_copy_output(session->consumer);
                /* Ease our life a bit for the next part */
                consumer = session->kernel_session->consumer;
                dir_name = DEFAULT_KERNEL_TRACE_DIR;
                break;
        case LTTNG_DOMAIN_JUL:
        case LTTNG_DOMAIN_LOG4J:
        case LTTNG_DOMAIN_PYTHON:
        case LTTNG_DOMAIN_UST:
                DBG3("Copying tracing session consumer output in UST session");
                if (session->ust_session->consumer) {
                        consumer_output_put(session->ust_session->consumer);
                }
                session->ust_session->consumer =
                        consumer_copy_output(session->consumer);
                /* Ease our life a bit for the next part */
                consumer = session->ust_session->consumer;
                dir_name = DEFAULT_UST_TRACE_DIR;
                break;
        default:
                ret = LTTNG_ERR_UNKNOWN_DOMAIN;
                goto error;
        }

        /* Append correct directory to subdir */
        strncat(consumer->subdir, dir_name,
                        sizeof(consumer->subdir) - strlen(consumer->subdir) - 1);
        DBG3("Copy session consumer subdir %s", consumer->subdir);

        ret = LTTNG_OK;

error:
        return ret;
}

/*
 * Create an UST session and add it to the session ust list.
 *
 * Should *NOT* be called with RCU read-side lock held.
 */
static int create_ust_session(struct ltt_session *session,
                struct lttng_domain *domain)
{
        int ret;
        struct ltt_ust_session *lus = NULL;

        assert(session);
        assert(domain);
        assert(session->consumer);

        switch (domain->type) {
        case LTTNG_DOMAIN_JUL:
        case LTTNG_DOMAIN_LOG4J:
        case LTTNG_DOMAIN_PYTHON:
        case LTTNG_DOMAIN_UST:
                break;
        default:
                ERR("Unknown UST domain on create session %d", domain->type);
                ret = LTTNG_ERR_UNKNOWN_DOMAIN;
                goto error;
        }

        DBG("Creating UST session");

        lus = trace_ust_create_session(session->id);
        if (lus == NULL) {
                ret = LTTNG_ERR_UST_SESS_FAIL;
                goto error;
        }

        lus->uid = session->uid;
        lus->gid = session->gid;
        lus->output_traces = session->output_traces;
        lus->snapshot_mode = session->snapshot_mode;
        lus->live_timer_interval = session->live_timer;
        session->ust_session = lus;
        if (session->shm_path[0]) {
                strncpy(lus->root_shm_path, session->shm_path,
                        sizeof(lus->root_shm_path));
                lus->root_shm_path[sizeof(lus->root_shm_path) - 1] = '\0';
                strncpy(lus->shm_path, session->shm_path,
                        sizeof(lus->shm_path));
                lus->shm_path[sizeof(lus->shm_path) - 1] = '\0';
                strncat(lus->shm_path, "/ust",
                        sizeof(lus->shm_path) - strlen(lus->shm_path) - 1);
        }
        /* Copy session output to the newly created UST session */
        ret = copy_session_consumer(domain->type, session);
        if (ret != LTTNG_OK) {
                goto error;
        }

        return LTTNG_OK;

error:
        free(lus);
        session->ust_session = NULL;
        return ret;
}

/*
 * Create a kernel tracer session then create the default channel.
 */
static int create_kernel_session(struct ltt_session *session)
{
        int ret;

        DBG("Creating kernel session");

        ret = kernel_create_session(session, kernel_tracer_fd);
        if (ret < 0) {
                ret = LTTNG_ERR_KERN_SESS_FAIL;
                goto error;
        }

        /* Code flow safety */
        assert(session->kernel_session);

        /* Copy session output to the newly created Kernel session */
        ret = copy_session_consumer(LTTNG_DOMAIN_KERNEL, session);
        if (ret != LTTNG_OK) {
                goto error;
        }

        session->kernel_session->uid = session->uid;
        session->kernel_session->gid = session->gid;
        session->kernel_session->output_traces = session->output_traces;
        session->kernel_session->snapshot_mode = session->snapshot_mode;

        return LTTNG_OK;

error:
        trace_kernel_destroy_session(session->kernel_session);
        session->kernel_session = NULL;
        return ret;
}

/*
 * Count number of session permitted by uid/gid.
 */
static unsigned int lttng_sessions_count(uid_t uid, gid_t gid)
{
        unsigned int i = 0;
        struct ltt_session *session;

        DBG("Counting number of available session for UID %d GID %d",
                        uid, gid);
        cds_list_for_each_entry(session, &session_list_ptr->head, list) {
                /*
                 * Only list the sessions the user can control.
                 */
                if (!session_access_ok(session, uid, gid)) {
                        continue;
                }
                i++;
        }
        return i;
}

static int receive_userspace_probe(struct command_ctx *cmd_ctx, int sock,
                int *sock_error, struct lttng_event *event)
{
        int fd, ret;
        struct lttng_userspace_probe_location *probe_location;
        const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
        struct lttng_dynamic_buffer probe_location_buffer;
        struct lttng_buffer_view buffer_view;

        /*
         * Create a buffer to store the serialized version of the probe
         * location.
         */
        lttng_dynamic_buffer_init(&probe_location_buffer);
        ret = lttng_dynamic_buffer_set_size(&probe_location_buffer,
                        cmd_ctx->lsm->u.enable.userspace_probe_location_len);
        if (ret) {
                ret = LTTNG_ERR_NOMEM;
                goto error;
        }

        /*
         * Receive the probe location.
         */
        ret = lttcomm_recv_unix_sock(sock, probe_location_buffer.data,
                        probe_location_buffer.size);
        if (ret <= 0) {
                DBG("Nothing recv() from client var len data... continuing");
                *sock_error = 1;
                lttng_dynamic_buffer_reset(&probe_location_buffer);
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        buffer_view = lttng_buffer_view_from_dynamic_buffer(
                        &probe_location_buffer, 0, probe_location_buffer.size);

        /*
         * Extract the probe location from the serialized version.
         */
        ret = lttng_userspace_probe_location_create_from_buffer(
                                &buffer_view, &probe_location);
        if (ret < 0) {
                WARN("Failed to create a userspace probe location from the received buffer");
                lttng_dynamic_buffer_reset( &probe_location_buffer);
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        /*
         * Receive the file descriptor to the target binary from the client.
         */
        DBG("Receiving userspace probe target FD from client ...");
        ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
        if (ret <= 0) {
                DBG("Nothing recv() from client userspace probe fd... continuing");
                *sock_error = 1;
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        /*
         * Set the file descriptor received from the client through the unix
         * socket in the probe location.
         */
        lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
        if (!lookup) {
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        /*
         * From the kernel tracer's perspective, all userspace probe event types
         * are all the same: a file and an offset.
         */
        switch (lttng_userspace_probe_location_lookup_method_get_type(lookup)) {
        case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
                ret = lttng_userspace_probe_location_function_set_binary_fd(
                                probe_location, fd);
                break;
        case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
                ret = lttng_userspace_probe_location_tracepoint_set_binary_fd(
                                probe_location, fd);
                break;
        default:
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        if (ret) {
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        /* Attach the probe location to the event. */
        ret = lttng_event_set_userspace_probe_location(event, probe_location);
        if (ret) {
                ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
                goto error;
        }

        lttng_dynamic_buffer_reset(&probe_location_buffer);
error:
        return ret;
}

/*
 * Check if the current kernel tracer supports the session rotation feature.
 * Return 1 if it does, 0 otherwise.
 */
static int check_rotate_compatible(void)
{
        int ret = 1;

        if (kernel_tracer_version.major != 2 || kernel_tracer_version.minor < 11) {
                DBG("Kernel tracer version is not compatible with the rotation feature");
                ret = 0;
        }

        return ret;
}

/*
 * Process the command requested by the lttng client within the command
 * context structure. This function make sure that the return structure (llm)
 * is set and ready for transmission before returning.
 *
 * Return any error encountered or 0 for success.
 *
 * "sock" is only used for special-case var. len data.
 *
 * Should *NOT* be called with RCU read-side lock held.
 */
static int process_client_msg(struct command_ctx *cmd_ctx, int sock,
                int *sock_error)
{
        int ret = LTTNG_OK;
        int need_tracing_session = 1;
        int need_domain;

        DBG("Processing client command %d", cmd_ctx->lsm->cmd_type);

        assert(!rcu_read_ongoing());

        *sock_error = 0;

        switch (cmd_ctx->lsm->cmd_type) {
        case LTTNG_CREATE_SESSION:
        case LTTNG_CREATE_SESSION_SNAPSHOT:
        case LTTNG_CREATE_SESSION_LIVE:
        case LTTNG_DESTROY_SESSION:
        case LTTNG_LIST_SESSIONS:
        case LTTNG_LIST_DOMAINS:
        case LTTNG_START_TRACE:
        case LTTNG_STOP_TRACE:
        case LTTNG_DATA_PENDING:
        case LTTNG_SNAPSHOT_ADD_OUTPUT:
        case LTTNG_SNAPSHOT_DEL_OUTPUT:
        case LTTNG_SNAPSHOT_LIST_OUTPUT:
        case LTTNG_SNAPSHOT_RECORD:
        case LTTNG_SAVE_SESSION:
        case LTTNG_SET_SESSION_SHM_PATH:
        case LTTNG_REGENERATE_METADATA:
        case LTTNG_REGENERATE_STATEDUMP:
        case LTTNG_REGISTER_TRIGGER:
        case LTTNG_UNREGISTER_TRIGGER:
        case LTTNG_ROTATE_SESSION:
        case LTTNG_ROTATION_GET_INFO:
        case LTTNG_ROTATION_SET_SCHEDULE:
        case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
                need_domain = 0;
                break;
        default:
                need_domain = 1;
        }

        if (config.no_kernel && need_domain
                        && cmd_ctx->lsm->domain.type == LTTNG_DOMAIN_KERNEL) {
                if (!is_root) {
                        ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
                } else {
                        ret = LTTNG_ERR_KERN_NA;
                }
                goto error;
        }

        /* Deny register consumer if we already have a spawned consumer. */
        if (cmd_ctx->lsm->cmd_type == LTTNG_REGISTER_CONSUMER) {
                pthread_mutex_lock(&kconsumer_data.pid_mutex);
                if (kconsumer_data.pid > 0) {
                        ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
                        pthread_mutex_unlock(&kconsumer_data.pid_mutex);
                        goto error;
                }
                pthread_mutex_unlock(&kconsumer_data.pid_mutex);
        }

        /*
         * Check for command that don't needs to allocate a returned payload. We do
         * this here so we don't have to make the call for no payload at each
         * command.
         */
        switch(cmd_ctx->lsm->cmd_type) {
        case LTTNG_LIST_SESSIONS:
        case LTTNG_LIST_TRACEPOINTS:
        case LTTNG_LIST_TRACEPOINT_FIELDS:
        case LTTNG_LIST_DOMAINS:
        case LTTNG_LIST_CHANNELS:
        case LTTNG_LIST_EVENTS:
        case LTTNG_LIST_SYSCALLS:
        case LTTNG_LIST_TRACKER_PIDS:
        case LTTNG_DATA_PENDING:
        case LTTNG_ROTATE_SESSION:
        case LTTNG_ROTATION_GET_INFO:
        case LTTNG_SESSION_LIST_ROTATION_SCHEDULES:
                break;
        default:
                /* Setup lttng message with no payload */
                ret = setup_lttng_msg_no_cmd_header(cmd_ctx, NULL, 0);
                if (ret < 0) {
                        /* This label does not try to unlock the session */
                        goto init_setup_error;
                }
        }

        /* Commands that DO NOT need a session. */
        switch (cmd_ctx->lsm->cmd_type) {
        case LTTNG_CREATE_SESSION:
        case LTTNG_CREATE_SESSION_SNAPSHOT:
        case LTTNG_CREATE_SESSION_LIVE:
        case LTTNG_LIST_SESSIONS:
        case LTTNG_LIST_TRACEPOINTS:
        case LTTNG_LIST_SYSCALLS:
        case LTTNG_LIST_TRACEPOINT_FIELDS:
        case LTTNG_SAVE_SESSION:
        case LTTNG_REGISTER_TRIGGER:
        case LTTNG_UNREGISTER_TRIGGER:
                need_tracing_session = 0;
                break;
        default:
                DBG("Getting session %s by name", cmd_ctx->lsm->session.name);
                /*
                 * We keep the session list lock across _all_ commands
                 * for now, because the per-session lock does not
                 * handle teardown properly.
                 */
                session_lock_list();
                cmd_ctx->session = session_find_by_name(cmd_ctx->lsm->session.name);
                if (cmd_ctx->session == NULL) {
                        ret = LTTNG_ERR_SESS_NOT_FOUND;
                        goto error;
                } else {
                        /* Acquire lock for the session */
                        session_lock(cmd_ctx->session);
                }
                break;
        }

        /*
         * Commands that need a valid session but should NOT create one if none
         * exists. Instead of creating one and destroying it when the command is
         * handled, process that right before so we save some round trip in useless
         * code path.
         */
        switch (cmd_ctx->lsm->cmd_type) {
        case LTTNG_DISABLE_CHANNEL:
        case LTTNG_DISABLE_EVENT:
                switch (cmd_ctx->lsm->domain.type) {
                case LTTNG_DOMAIN_KERNEL:
                        if (!cmd_ctx->session->kernel_session) {
                                ret = LTTNG_ERR_NO_CHANNEL;
                                goto error;
                        }
                        break;
                case LTTNG_DOMAIN_JUL:
                case LTTNG_DOMAIN_LOG4J:
                case LTTNG_DOMAIN_PYTHON:
                case LTTNG_DOMAIN_UST:
                        if (!cmd_ctx->session->ust_session) {
                                ret = LTTNG_ERR_NO_CHANNEL;
                                goto error;
                        }
                        break;
                default:
                        ret = LTTNG_ERR_UNKNOWN_DOMAIN;
                        goto error;
                }
        default:
                break;
        }

        if (!need_domain) {
                goto skip_domain;
        }

        /*
         * Check domain type for specific "pre-action".
         */
        switch (cmd_ctx->lsm->domain.type) {
        case LTTNG_DOMAIN_KERNEL:
                if (!is_root) {
                        ret = LTTNG_ERR_NEED_ROOT_SESSIOND;
                        goto error;
                }

                /* Kernel tracer check */
                if (kernel_tracer_fd == -1) {
                        /* Basically, load kernel tracer modules */
                        ret = init_kernel_tracer();
                        if (ret != 0) {
                                goto error;
                        }
                }

                /* Consumer is in an ERROR state. Report back to client */
                if (uatomic_read(&kernel_consumerd_state) == CONSUMER_ERROR) {
                        ret = LTTNG_ERR_NO_KERNCONSUMERD;
                        goto error;
                }

                /* Need a session for kernel command */
                if (need_tracing_session) {
                        if (cmd_ctx->session->kernel_session == NULL) {
                                ret = create_kernel_session(cmd_ctx->session);
                                if (ret < 0) {
                                        ret = LTTNG_ERR_KERN_SESS_FAIL;
                                        goto error;