Fix: code refactoring of viewer streams in relayd

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

/*
 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
 *                      David Goulet <dgoulet@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 _GNU_SOURCE
#include <getopt.h>
#include <grp.h>
#include <limits.h>
#include <pthread.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.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 <inttypes.h>
#include <urcu/futex.h>
#include <urcu/uatomic.h>
#include <unistd.h>
#include <fcntl.h>
#include <config.h>

#include <lttng/lttng.h>
#include <common/common.h>
#include <common/compat/poll.h>
#include <common/compat/socket.h>
#include <common/defaults.h>
#include <common/daemonize.h>
#include <common/futex.h>
#include <common/sessiond-comm/sessiond-comm.h>
#include <common/sessiond-comm/inet.h>
#include <common/sessiond-comm/relayd.h>
#include <common/uri.h>
#include <common/utils.h>
#include <common/config/config.h>

#include "cmd.h"
#include "ctf-trace.h"
#include "index.h"
#include "utils.h"
#include "lttng-relayd.h"
#include "live.h"
#include "health-relayd.h"
#include "testpoint.h"
#include "viewer-stream.h"

/* command line options */
char *opt_output_path;
static int opt_daemon, opt_background;

/*
 * We need to wait for listener and live listener threads, as well as
 * health check thread, before being ready to signal readiness.
 */
#define NR_LTTNG_RELAY_READY    3
static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
static int recv_child_signal;   /* Set to 1 when a SIGUSR1 signal is received. */
static pid_t child_ppid;        /* Internal parent PID use with daemonize. */

static struct lttng_uri *control_uri;
static struct lttng_uri *data_uri;
static struct lttng_uri *live_uri;

const char *progname;

const char *tracing_group_name = DEFAULT_TRACING_GROUP;
static int tracing_group_name_override;

const char * const config_section_name = "relayd";

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

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

/* Shared between threads */
static int dispatch_thread_exit;

static pthread_t listener_thread;
static pthread_t dispatcher_thread;
static pthread_t worker_thread;
static pthread_t health_thread;

static uint64_t last_relay_stream_id;
static uint64_t last_relay_session_id;

/*
 * Relay command queue.
 *
 * The relay_thread_listener and relay_thread_dispatcher communicate with this
 * queue.
 */
static struct relay_cmd_queue relay_cmd_queue;

/* buffer allocated at startup, used to store the trace data */
static char *data_buffer;
static unsigned int data_buffer_size;

/* We need those values for the file/dir creation. */
static uid_t relayd_uid;
static gid_t relayd_gid;

/* Global relay stream hash table. */
struct lttng_ht *relay_streams_ht;

/* Global relay viewer stream hash table. */
struct lttng_ht *viewer_streams_ht;

/* Global hash table that stores relay index object. */
struct lttng_ht *indexes_ht;

/* Relayd health monitoring */
struct health_app *health_relayd;

static struct option long_options[] = {
        { "control-port", 1, 0, 'C', },
        { "data-port", 1, 0, 'D', },
        { "live-port", 1, 0, 'L', },
        { "daemonize", 0, 0, 'd', },
        { "background", 0, 0, 'b', },
        { "group", 1, 0, 'g', },
        { "help", 0, 0, 'h', },
        { "output", 1, 0, 'o', },
        { "verbose", 0, 0, 'v', },
        { "config", 1, 0, 'f' },
        { NULL, 0, 0, 0, },
};

static const char *config_ignore_options[] = { "help", "config" };

/*
 * usage function on stderr
 */
static
void usage(void)
{
        fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
        fprintf(stderr, "  -h, --help                Display this usage.\n");
        fprintf(stderr, "  -d, --daemonize           Start as a daemon.\n");
        fprintf(stderr, "  -b, --background          Start as a daemon, keeping console open.\n");
        fprintf(stderr, "  -C, --control-port URL    Control port listening.\n");
        fprintf(stderr, "  -D, --data-port URL       Data port listening.\n");
        fprintf(stderr, "  -L, --live-port URL       Live view port listening.\n");
        fprintf(stderr, "  -o, --output PATH         Output path for traces. Must use an absolute path.\n");
        fprintf(stderr, "  -v, --verbose             Verbose mode. Activate DBG() macro.\n");
        fprintf(stderr, "  -g, --group NAME          Specify the tracing group name. (default: tracing)\n");
        fprintf(stderr, "  -f  --config              Load daemon configuration file\n");
}

/*
 * Take an option from the getopt output and set it in the right variable to be
 * used later.
 *
 * Return 0 on success else a negative value.
 */
static
int set_option(int opt, const char *arg, const char *optname)
{
        int ret;

        switch (opt) {
        case 0:
                fprintf(stderr, "option %s", optname);
                if (arg) {
                        fprintf(stderr, " with arg %s\n", arg);
                }
                break;
        case 'C':
                ret = uri_parse(arg, &control_uri);
                if (ret < 0) {
                        ERR("Invalid control URI specified");
                        goto end;
                }
                if (control_uri->port == 0) {
                        control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
                }
                break;
        case 'D':
                ret = uri_parse(arg, &data_uri);
                if (ret < 0) {
                        ERR("Invalid data URI specified");
                        goto end;
                }
                if (data_uri->port == 0) {
                        data_uri->port = DEFAULT_NETWORK_DATA_PORT;
                }
                break;
        case 'L':
                ret = uri_parse(arg, &live_uri);
                if (ret < 0) {
                        ERR("Invalid live URI specified");
                        goto end;
                }
                if (live_uri->port == 0) {
                        live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
                }
                break;
        case 'd':
                opt_daemon = 1;
                break;
        case 'b':
                opt_background = 1;
                break;
        case 'g':
                tracing_group_name = strdup(arg);
                tracing_group_name_override = 1;
                break;
        case 'h':
                usage();
                exit(EXIT_FAILURE);
        case 'o':
                ret = asprintf(&opt_output_path, "%s", arg);
                if (ret < 0) {
                        ret = -errno;
                        PERROR("asprintf opt_output_path");
                        goto end;
                }
                break;
        case 'v':
                /* Verbose level can increase using multiple -v */
                if (arg) {
                        lttng_opt_verbose = config_parse_value(arg);
                } else {
                        lttng_opt_verbose += 1;
                }
                break;
        default:
                /* Unknown option or other error.
                 * Error is printed by getopt, just return */
                ret = -1;
                goto end;
        }

        /* All good. */
        ret = 0;

end:
        return ret;
}

/*
 * config_entry_handler_cb used to handle options read from a config file.
 * See config_entry_handler_cb comment in common/config/config.h for the
 * return value conventions.
 */
static
int config_entry_handler(const struct config_entry *entry, void *unused)
{
        int ret = 0, i;

        if (!entry || !entry->name || !entry->value) {
                ret = -EINVAL;
                goto end;
        }

        /* Check if the option is to be ignored */
        for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
                if (!strcmp(entry->name, config_ignore_options[i])) {
                        goto end;
                }
        }

        for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
                /* Ignore if entry name is not fully matched. */
                if (strcmp(entry->name, long_options[i].name)) {
                        continue;
                }

                /*
                 * If the option takes no argument on the command line, we have to
                 * check if the value is "true". We support non-zero numeric values,
                 * true, on and yes.
                 */
                if (!long_options[i].has_arg) {
                        ret = config_parse_value(entry->value);
                        if (ret <= 0) {
                                if (ret) {
                                        WARN("Invalid configuration value \"%s\" for option %s",
                                                        entry->value, entry->name);
                                }
                                /* False, skip boolean config option. */
                                goto end;
                        }
                }

                ret = set_option(long_options[i].val, entry->value, entry->name);
                goto end;
        }

        WARN("Unrecognized option \"%s\" in daemon configuration file.",
                        entry->name);

end:
        return ret;
}

static
int set_options(int argc, char **argv)
{
        int c, ret = 0, option_index = 0;
        int orig_optopt = optopt, orig_optind = optind;
        char *default_address, *optstring;
        const char *config_path = NULL;

        optstring = utils_generate_optstring(long_options,
                        sizeof(long_options) / sizeof(struct option));
        if (!optstring) {
                ret = -ENOMEM;
                goto exit;
        }

        /* Check for the --config option */

        while ((c = getopt_long(argc, argv, optstring, long_options,
                                        &option_index)) != -1) {
                if (c == '?') {
                        ret = -EINVAL;
                        goto exit;
                } else if (c != 'f') {
                        continue;
                }

                config_path = utils_expand_path(optarg);
                if (!config_path) {
                        ERR("Failed to resolve path: %s", optarg);
                }
        }

        ret = config_get_section_entries(config_path, config_section_name,
                        config_entry_handler, NULL);
        if (ret) {
                if (ret > 0) {
                        ERR("Invalid configuration option at line %i", ret);
                        ret = -1;
                }
                goto exit;
        }

        /* Reset getopt's global state */
        optopt = orig_optopt;
        optind = orig_optind;
        while (1) {
                c = getopt_long(argc, argv, optstring, long_options, &option_index);
                if (c == -1) {
                        break;
                }

                ret = set_option(c, optarg, long_options[option_index].name);
                if (ret < 0) {
                        goto exit;
                }
        }

        /* assign default values */
        if (control_uri == NULL) {
                ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
                                DEFAULT_NETWORK_CONTROL_PORT);
                if (ret < 0) {
                        PERROR("asprintf default data address");
                        goto exit;
                }

                ret = uri_parse(default_address, &control_uri);
                free(default_address);
                if (ret < 0) {
                        ERR("Invalid control URI specified");
                        goto exit;
                }
        }
        if (data_uri == NULL) {
                ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
                                DEFAULT_NETWORK_DATA_PORT);
                if (ret < 0) {
                        PERROR("asprintf default data address");
                        goto exit;
                }

                ret = uri_parse(default_address, &data_uri);
                free(default_address);
                if (ret < 0) {
                        ERR("Invalid data URI specified");
                        goto exit;
                }
        }
        if (live_uri == NULL) {
                ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
                                DEFAULT_NETWORK_VIEWER_PORT);
                if (ret < 0) {
                        PERROR("asprintf default viewer control address");
                        goto exit;
                }

                ret = uri_parse(default_address, &live_uri);
                free(default_address);
                if (ret < 0) {
                        ERR("Invalid viewer control URI specified");
                        goto exit;
                }
        }

exit:
        free(optstring);
        return ret;
}

/*
 * Cleanup the daemon
 */
static
void cleanup(void)
{
        DBG("Cleaning up");

        /* free the dynamically allocated opt_output_path */
        free(opt_output_path);

        /* Close thread quit pipes */
        utils_close_pipe(thread_quit_pipe);

        uri_free(control_uri);
        uri_free(data_uri);
        /* Live URI is freed in the live thread. */

        if (tracing_group_name_override) {
                free((void *) tracing_group_name);
        }
}

/*
 * Write to writable pipe used to notify a thread.
 */
static
int notify_thread_pipe(int wpipe)
{
        ssize_t ret;

        ret = lttng_write(wpipe, "!", 1);
        if (ret < 1) {
                PERROR("write poll pipe");
        }

        return ret;
}

static void notify_health_quit_pipe(int *pipe)
{
        ssize_t ret;

        ret = lttng_write(pipe[1], "4", 1);
        if (ret < 1) {
                PERROR("write relay health quit");
        }
}

/*
 * 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");
        }

        notify_health_quit_pipe(health_quit_pipe);

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

/*
 * Signal handler for the daemon
 *
 * Simply stop all worker threads, leaving main() return gracefully after
 * joining all threads and calling cleanup().
 */
static
void sighandler(int sig)
{
        switch (sig) {
        case SIGPIPE:
                DBG("SIGPIPE caught");
                return;
        case SIGINT:
                DBG("SIGINT caught");
                stop_threads();
                break;
        case SIGTERM:
                DBG("SIGTERM caught");
                stop_threads();
                break;
        case SIGUSR1:
                CMM_STORE_SHARED(recv_child_signal, 1);
                break;
        default:
                break;
        }
}

/*
 * Setup signal handler for :
 *              SIGINT, SIGTERM, SIGPIPE
 */
static
int set_signal_handler(void)
{
        int ret = 0;
        struct sigaction sa;
        sigset_t sigset;

        if ((ret = sigemptyset(&sigset)) < 0) {
                PERROR("sigemptyset");
                return ret;
        }

        sa.sa_handler = sighandler;
        sa.sa_mask = sigset;
        sa.sa_flags = 0;
        if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
                PERROR("sigaction");
                return ret;
        }

        if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
                PERROR("sigaction");
                return ret;
        }

        if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
                PERROR("sigaction");
                return ret;
        }

        if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
                PERROR("sigaction");
                return ret;
        }

        DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");

        return ret;
}

void lttng_relay_notify_ready(void)
{
        /* Notify the parent of the fork() process that we are ready. */
        if (opt_daemon || opt_background) {
                if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
                        kill(child_ppid, SIGUSR1);
                }
        }
}

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

        ret = utils_create_pipe_cloexec(thread_quit_pipe);

        return ret;
}

/*
 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
 */
static
int create_thread_poll_set(struct lttng_poll_event *events, int size)
{
        int ret;

        if (events == NULL || size == 0) {
                ret = -1;
                goto error;
        }

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

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

        return 0;

error:
        return ret;
}

/*
 * Check if the thread quit pipe was triggered.
 *
 * Return 1 if it was triggered else 0;
 */
static
int check_thread_quit_pipe(int fd, uint32_t events)
{
        if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
                return 1;
        }

        return 0;
}

/*
 * Create and init socket from uri.
 */
static
struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
{
        int ret;
        struct lttcomm_sock *sock = NULL;

        sock = lttcomm_alloc_sock_from_uri(uri);
        if (sock == NULL) {
                ERR("Allocating socket");
                goto error;
        }

        ret = lttcomm_create_sock(sock);
        if (ret < 0) {
                goto error;
        }
        DBG("Listening on sock %d", sock->fd);

        ret = sock->ops->bind(sock);
        if (ret < 0) {
                goto error;
        }

        ret = sock->ops->listen(sock, -1);
        if (ret < 0) {
                goto error;

        }

        return sock;

error:
        if (sock) {
                lttcomm_destroy_sock(sock);
        }
        return NULL;
}

/*
 * Return nonzero if stream needs to be closed.
 */
static
int close_stream_check(struct relay_stream *stream)
{

        if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
                /*
                 * We are about to close the stream so set the data pending flag to 1
                 * which will make the end data pending command skip the stream which
                 * is now closed and ready. Note that after proceeding to a file close,
                 * the written file is ready for reading.
                 */
                stream->data_pending_check_done = 1;
                return 1;
        }
        return 0;
}

/*
 * This thread manages the listening for new connections on the network
 */
static
void *relay_thread_listener(void *data)
{
        int i, ret, pollfd, err = -1;
        int val = 1;
        uint32_t revents, nb_fd;
        struct lttng_poll_event events;
        struct lttcomm_sock *control_sock, *data_sock;

        DBG("[thread] Relay listener started");

        health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);

        health_code_update();

        control_sock = relay_init_sock(control_uri);
        if (!control_sock) {
                goto error_sock_control;
        }

        data_sock = relay_init_sock(data_uri);
        if (!data_sock) {
                goto error_sock_relay;
        }

        /*
         * Pass 3 as size here for the thread quit pipe, control and data socket.
         */
        ret = create_thread_poll_set(&events, 3);
        if (ret < 0) {
                goto error_create_poll;
        }

        /* Add the control socket */
        ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
        if (ret < 0) {
                goto error_poll_add;
        }

        /* Add the data socket */
        ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
        if (ret < 0) {
                goto error_poll_add;
        }

        lttng_relay_notify_ready();

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

        while (1) {
                health_code_update();

                DBG("Listener accepting connections");

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;

                DBG("Relay new connection received");
                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);

                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
                        if (ret) {
                                err = 0;
                                goto exit;
                        }

                        if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
                                ERR("socket poll error");
                                goto error;
                        } else if (revents & LPOLLIN) {
                                /*
                                 * Get allocated in this thread,
                                 * enqueued to a global queue, dequeued
                                 * and freed in the worker thread.
                                 */
                                struct relay_command *relay_cmd;
                                struct lttcomm_sock *newsock;

                                relay_cmd = zmalloc(sizeof(struct relay_command));
                                if (relay_cmd == NULL) {
                                        PERROR("relay command zmalloc");
                                        goto error;
                                }

                                if (pollfd == data_sock->fd) {
                                        newsock = data_sock->ops->accept(data_sock);
                                        if (!newsock) {
                                                PERROR("accepting data sock");
                                                free(relay_cmd);
                                                goto error;
                                        }
                                        relay_cmd->type = RELAY_DATA;
                                        DBG("Relay data connection accepted, socket %d", newsock->fd);
                                } else {
                                        assert(pollfd == control_sock->fd);
                                        newsock = control_sock->ops->accept(control_sock);
                                        if (!newsock) {
                                                PERROR("accepting control sock");
                                                free(relay_cmd);
                                                goto error;
                                        }
                                        relay_cmd->type = RELAY_CONTROL;
                                        DBG("Relay control connection accepted, socket %d", newsock->fd);
                                }
                                ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
                                                &val, sizeof(int));
                                if (ret < 0) {
                                        PERROR("setsockopt inet");
                                        lttcomm_destroy_sock(newsock);
                                        free(relay_cmd);
                                        goto error;
                                }
                                relay_cmd->sock = newsock;
                                /*
                                 * Lock free enqueue the request.
                                 */
                                cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);

                                /*
                                 * Wake the dispatch queue futex. Implicit memory
                                 * barrier with the exchange in cds_wfq_enqueue.
                                 */
                                futex_nto1_wake(&relay_cmd_queue.futex);
                        }
                }
        }

exit:
error:
error_poll_add:
error_testpoint:
        lttng_poll_clean(&events);
error_create_poll:
        if (data_sock->fd >= 0) {
                ret = data_sock->ops->close(data_sock);
                if (ret) {
                        PERROR("close");
                }
        }
        lttcomm_destroy_sock(data_sock);
error_sock_relay:
        if (control_sock->fd >= 0) {
                ret = control_sock->ops->close(control_sock);
                if (ret) {
                        PERROR("close");
                }
        }
        lttcomm_destroy_sock(control_sock);
error_sock_control:
        if (err) {
                health_error();
                ERR("Health error occurred in %s", __func__);
        }
        health_unregister(health_relayd);
        DBG("Relay listener thread cleanup complete");
        stop_threads();
        return NULL;
}

/*
 * This thread manages the dispatching of the requests to worker threads
 */
static
void *relay_thread_dispatcher(void *data)
{
        int err = -1;
        ssize_t ret;
        struct cds_wfq_node *node;
        struct relay_command *relay_cmd = NULL;

        DBG("[thread] Relay dispatcher started");

        health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);

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

        health_code_update();

        while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
                health_code_update();

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

                do {
                        health_code_update();

                        /* Dequeue commands */
                        node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
                        if (node == NULL) {
                                DBG("Woken up but nothing in the relay command queue");
                                /* Continue thread execution */
                                break;
                        }

                        relay_cmd = caa_container_of(node, struct relay_command, node);
                        DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);

                        /*
                         * Inform worker thread of the new request. This
                         * call is blocking so we can be assured that the data will be read
                         * at some point in time or wait to the end of the world :)
                         */
                        ret = lttng_write(relay_cmd_pipe[1], relay_cmd,
                                        sizeof(struct relay_command));
                        free(relay_cmd);
                        if (ret < sizeof(struct relay_command)) {
                                PERROR("write cmd pipe");
                                goto error;
                        }
                } while (node != NULL);

                /* Futex wait on queue. Blocking call on futex() */
                health_poll_entry();
                futex_nto1_wait(&relay_cmd_queue.futex);
                health_poll_exit();
        }

        /* Normal exit, no error */
        err = 0;

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

/*
 * Get stream from stream id.
 * Need to be called with RCU read-side lock held.
 */
struct relay_stream *relay_stream_find_by_id(uint64_t stream_id)
{
        struct lttng_ht_node_ulong *node;
        struct lttng_ht_iter iter;
        struct relay_stream *ret;

        lttng_ht_lookup(relay_streams_ht,
                        (void *)((unsigned long) stream_id),
                        &iter);
        node = lttng_ht_iter_get_node_ulong(&iter);
        if (node == NULL) {
                DBG("Relay stream %" PRIu64 " not found", stream_id);
                ret = NULL;
                goto end;
        }

        ret = caa_container_of(node, struct relay_stream, stream_n);

end:
        return ret;
}

static
void deferred_free_stream(struct rcu_head *head)
{
        struct relay_stream *stream =
                caa_container_of(head, struct relay_stream, rcu_node);

        free(stream->path_name);
        free(stream->channel_name);
        free(stream);
}

static
void deferred_free_session(struct rcu_head *head)
{
        struct relay_session *session =
                caa_container_of(head, struct relay_session, rcu_node);
        free(session);
}

/*
 * Close a given stream. The stream is freed using a call RCU.
 *
 * RCU read side lock MUST be acquired. If NO close_stream_check() was called
 * BEFORE the stream lock MUST be acquired.
 */
static void destroy_stream(struct relay_stream *stream)
{
        int delret;
        struct relay_viewer_stream *vstream;
        struct lttng_ht_iter iter;

        assert(stream);

        delret = close(stream->fd);
        if (delret < 0) {
                PERROR("close stream");
        }

        if (stream->index_fd >= 0) {
                delret = close(stream->index_fd);
                if (delret < 0) {
                        PERROR("close stream index_fd");
                }
        }

        vstream = viewer_stream_find_by_id(stream->stream_handle);
        if (vstream) {
                /*
                 * Set the last good value into the viewer stream. This is done
                 * right before the stream gets deleted from the hash table. The
                 * lookup failure on the live thread side of a stream indicates
                 * that the viewer stream index received value should be used.
                 */
                pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
                vstream->total_index_received = stream->total_index_received;
                vstream->tracefile_count_last = stream->tracefile_count_current;
                vstream->close_write_flag = 1;
                pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
        }

        /* Cleanup index of that stream. */
        relay_index_destroy_by_stream_id(stream->stream_handle);

        iter.iter.node = &stream->stream_n.node;
        delret = lttng_ht_del(relay_streams_ht, &iter);
        assert(!delret);
        iter.iter.node = &stream->ctf_trace_node.node;
        delret = lttng_ht_del(stream->ctf_traces_ht, &iter);
        assert(!delret);

        if (stream->ctf_trace) {
                ctf_trace_try_destroy(stream->ctf_trace);
        }

        call_rcu(&stream->rcu_node, deferred_free_stream);
        DBG("Closed tracefile %d from close stream", stream->fd);
}

/*
 * relay_delete_session: Free all memory associated with a session and
 * close all the FDs
 */
static
void relay_delete_session(struct relay_command *cmd,
                struct lttng_ht *sessions_ht)
{
        struct lttng_ht_iter iter;
        struct lttng_ht_node_ulong *node;
        struct relay_stream *stream;
        int ret;

        if (!cmd->session) {
                return;
        }

        DBG("Relay deleting session %" PRIu64, cmd->session->id);

        rcu_read_lock();
        cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) {
                node = lttng_ht_iter_get_node_ulong(&iter);
                if (!node) {
                        continue;
                }
                stream = caa_container_of(node, struct relay_stream, stream_n);
                if (stream->session == cmd->session) {
                        destroy_stream(stream);
                        cmd->session->stream_count--;
                        assert(cmd->session->stream_count >= 0);
                }
        }

        /* Make this session not visible anymore. */
        iter.iter.node = &cmd->session->session_n.node;
        ret = lttng_ht_del(sessions_ht, &iter);
        assert(!ret);
        call_rcu(&cmd->session->rcu_node, deferred_free_session);
        rcu_read_unlock();
}

/*
 * Copy index data from the control port to a given index object.
 */
static void copy_index_control_data(struct relay_index *index,
                struct lttcomm_relayd_index *data)
{
        assert(index);
        assert(data);

        /*
         * The index on disk is encoded in big endian, so we don't need to convert
         * the data received on the network. The data_offset value is NEVER
         * modified here and is updated by the data thread.
         */
        index->index_data.packet_size = data->packet_size;
        index->index_data.content_size = data->content_size;
        index->index_data.timestamp_begin = data->timestamp_begin;
        index->index_data.timestamp_end = data->timestamp_end;
        index->index_data.events_discarded = data->events_discarded;
        index->index_data.stream_id = data->stream_id;
}

/*
 * Handle the RELAYD_CREATE_SESSION command.
 *
 * On success, send back the session id or else return a negative value.
 */
static
int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd,
                struct lttng_ht *sessions_ht)
{
        int ret = 0, send_ret;
        struct relay_session *session;
        struct lttcomm_relayd_status_session reply;

        assert(recv_hdr);
        assert(cmd);

        memset(&reply, 0, sizeof(reply));

        session = zmalloc(sizeof(struct relay_session));
        if (session == NULL) {
                PERROR("relay session zmalloc");
                ret = -1;
                goto error;
        }

        session->id = ++last_relay_session_id;
        session->sock = cmd->sock;
        session->minor = cmd->minor;
        session->major = cmd->major;
        pthread_mutex_init(&session->viewer_ready_lock, NULL);
        cmd->session = session;

        reply.session_id = htobe64(session->id);

        switch (cmd->minor) {
                case 1:
                case 2:
                case 3:
                        break;
                case 4: /* LTTng sessiond 2.4 */
                default:
                        ret = cmd_create_session_2_4(cmd, session);
                        break;
        }

        lttng_ht_node_init_ulong(&session->session_n,
                        (unsigned long) session->id);
        lttng_ht_add_unique_ulong(sessions_ht,
                        &session->session_n);

        DBG("Created session %" PRIu64, session->id);

error:
        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_FATAL);
        } else {
                reply.ret_code = htobe32(LTTNG_OK);
        }

        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (send_ret < 0) {
                ERR("Relayd sending session id");
                ret = send_ret;
        }

        return ret;
}

/*
 * When we have received all the streams and the metadata for a channel,
 * we make them visible to the viewer threads.
 */
static
void set_viewer_ready_flag(struct relay_command *cmd)
{
        struct relay_stream_recv_handle *node, *tmp_node;

        pthread_mutex_lock(&cmd->session->viewer_ready_lock);

        cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) {
                struct relay_stream *stream;

                rcu_read_lock();
                stream = relay_stream_find_by_id(node->id);
                if (!stream) {
                        /*
                         * Stream is most probably being cleaned up by the data thread thus
                         * simply continue to the next one.
                         */
                        rcu_read_unlock();
                        continue;
                }

                stream->viewer_ready = 1;
                rcu_read_unlock();

                /* Clean stream handle node. */
                cds_list_del(&node->node);
                free(node);
        }

        pthread_mutex_unlock(&cmd->session->viewer_ready_lock);
        return;
}

/*
 * Add a recv handle node to the connection recv list with the given stream
 * handle. A new node is allocated thus must be freed when the node is deleted
 * from the list.
 */
static void queue_stream_handle(uint64_t handle, struct relay_command *cmd)
{
        struct relay_stream_recv_handle *node;

        assert(cmd);

        node = zmalloc(sizeof(*node));
        if (!node) {
                PERROR("zmalloc queue stream handle");
                return;
        }

        node->id = handle;
        cds_list_add(&node->node, &cmd->recv_head);
}

/*
 * relay_add_stream: allocate a new stream for a session
 */
static
int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd, struct lttng_ht *sessions_ht)
{
        struct relay_session *session = cmd->session;
        struct relay_stream *stream = NULL;
        struct lttcomm_relayd_status_stream reply;
        int ret, send_ret;

        if (!session || cmd->version_check_done == 0) {
                ERR("Trying to add a stream before version check");
                ret = -1;
                goto end_no_session;
        }

        stream = zmalloc(sizeof(struct relay_stream));
        if (stream == NULL) {
                PERROR("relay stream zmalloc");
                ret = -1;
                goto end_no_session;
        }

        switch (cmd->minor) {
        case 1: /* LTTng sessiond 2.1 */
                ret = cmd_recv_stream_2_1(cmd, stream);
                break;
        case 2: /* LTTng sessiond 2.2 */
        default:
                ret = cmd_recv_stream_2_2(cmd, stream);
                break;
        }
        if (ret < 0) {
                goto err_free_stream;
        }

        rcu_read_lock();
        stream->stream_handle = ++last_relay_stream_id;
        stream->prev_seq = -1ULL;
        stream->session = session;
        stream->index_fd = -1;
        stream->read_index_fd = -1;
        stream->ctf_trace = NULL;
        pthread_mutex_init(&stream->lock, NULL);

        ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
        if (ret < 0) {
                ERR("relay creating output directory");
                goto end;
        }

        /*
         * No need to use run_as API here because whatever we receives, the relayd
         * uses its own credentials for the stream files.
         */
        ret = utils_create_stream_file(stream->path_name, stream->channel_name,
                        stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
        if (ret < 0) {
                ERR("Create output file");
                goto end;
        }
        stream->fd = ret;
        if (stream->tracefile_size) {
                DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
        } else {
                DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
        }

        if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
                stream->metadata_flag = 1;
                /*
                 * When we receive a new metadata stream, we create a new
                 * ctf_trace and we assign this ctf_trace to all streams with
                 * the same path.
                 *
                 * If later on we receive a new stream for the same ctf_trace,
                 * we copy the information from the first hit in the HT to the
                 * new stream.
                 */
                stream->ctf_trace = ctf_trace_create();
                if (!stream->ctf_trace) {
                        ret = -1;
                        goto end;
                }
                stream->ctf_trace->refcount++;
                stream->ctf_trace->metadata_stream = stream;
        }
        ctf_trace_assign(cmd->ctf_traces_ht, stream);
        stream->ctf_traces_ht = cmd->ctf_traces_ht;

        /*
         * Add the stream handle in the recv list of the connection. Once the end
         * stream message is received, this list is emptied and streams are set
         * with the viewer ready flag.
         */
        queue_stream_handle(stream->stream_handle, cmd);

        lttng_ht_node_init_ulong(&stream->stream_n,
                        (unsigned long) stream->stream_handle);
        lttng_ht_add_unique_ulong(relay_streams_ht,
                        &stream->stream_n);

        lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name);
        lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node);
        session->stream_count++;

        DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
                        stream->stream_handle);

end:
        reply.handle = htobe64(stream->stream_handle);
        /* send the session id to the client or a negative return code on error */
        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_UNK);
                /* stream was not properly added to the ht, so free it */
                free(stream);
        } else {
                reply.ret_code = htobe32(LTTNG_OK);
        }

        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_status_stream), 0);
        if (send_ret < 0) {
                ERR("Relay sending stream id");
                ret = send_ret;
        }
        rcu_read_unlock();

end_no_session:
        return ret;

err_free_stream:
        free(stream->path_name);
        free(stream->channel_name);
        free(stream);
        return ret;
}

/*
 * relay_close_stream: close a specific stream
 */
static
int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret, send_ret;
        struct relay_session *session = cmd->session;
        struct lttcomm_relayd_close_stream stream_info;
        struct lttcomm_relayd_generic_reply reply;
        struct relay_stream *stream;

        DBG("Close stream received");

        if (!session || cmd->version_check_done == 0) {
                ERR("Trying to close a stream before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
                        sizeof(struct lttcomm_relayd_close_stream), 0);
        if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid add_stream struct size : %d", ret);
                }
                ret = -1;
                goto end_no_session;
        }

        rcu_read_lock();
        stream = relay_stream_find_by_id(be64toh(stream_info.stream_id));
        if (!stream) {
                ret = -1;
                goto end_unlock;
        }

        stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
        stream->close_flag = 1;
        session->stream_count--;
        assert(session->stream_count >= 0);

        if (close_stream_check(stream)) {
                destroy_stream(stream);
        }

end_unlock:
        rcu_read_unlock();

        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_UNK);
        } else {
                reply.ret_code = htobe32(LTTNG_OK);
        }
        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
        if (send_ret < 0) {
                ERR("Relay sending stream id");
                ret = send_ret;
        }

end_no_session:
        return ret;
}

/*
 * relay_unknown_command: send -1 if received unknown command
 */
static
void relay_unknown_command(struct relay_command *cmd)
{
        struct lttcomm_relayd_generic_reply reply;
        int ret;

        reply.ret_code = htobe32(LTTNG_ERR_UNK);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
        if (ret < 0) {
                ERR("Relay sending unknown command");
        }
}

/*
 * relay_start: send an acknowledgment to the client to tell if we are
 * ready to receive data. We are ready if a session is established.
 */
static
int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret = htobe32(LTTNG_OK);
        struct lttcomm_relayd_generic_reply reply;
        struct relay_session *session = cmd->session;

        if (!session) {
                DBG("Trying to start the streaming without a session established");
                ret = htobe32(LTTNG_ERR_UNK);
        }

        reply.ret_code = ret;
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
        if (ret < 0) {
                ERR("Relay sending start ack");
        }

        return ret;
}

/*
 * Append padding to the file pointed by the file descriptor fd.
 */
static int write_padding_to_file(int fd, uint32_t size)
{
        ssize_t ret = 0;
        char *zeros;

        if (size == 0) {
                goto end;
        }

        zeros = zmalloc(size);
        if (zeros == NULL) {
                PERROR("zmalloc zeros for padding");
                ret = -1;
                goto end;
        }

        ret = lttng_write(fd, zeros, size);
        if (ret < size) {
                PERROR("write padding to file");
        }

        free(zeros);

end:
        return ret;
}

/*
 * relay_recv_metadata: receive the metada for the session.
 */
static
int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret = htobe32(LTTNG_OK);
        ssize_t size_ret;
        struct relay_session *session = cmd->session;
        struct lttcomm_relayd_metadata_payload *metadata_struct;
        struct relay_stream *metadata_stream;
        uint64_t data_size, payload_size;

        if (!session) {
                ERR("Metadata sent before version check");
                ret = -1;
                goto end;
        }

        data_size = payload_size = be64toh(recv_hdr->data_size);
        if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
                ERR("Incorrect data size");
                ret = -1;
                goto end;
        }
        payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);

        if (data_buffer_size < data_size) {
                /* In case the realloc fails, we can free the memory */
                char *tmp_data_ptr;

                tmp_data_ptr = realloc(data_buffer, data_size);
                if (!tmp_data_ptr) {
                        ERR("Allocating data buffer");
                        free(data_buffer);
                        ret = -1;
                        goto end;
                }
                data_buffer = tmp_data_ptr;
                data_buffer_size = data_size;
        }
        memset(data_buffer, 0, data_size);
        DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
        ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
        if (ret < 0 || ret != data_size) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive the whole metadata");
                }
                ret = -1;
                goto end;
        }
        metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;

        rcu_read_lock();
        metadata_stream = relay_stream_find_by_id(
                        be64toh(metadata_struct->stream_id));
        if (!metadata_stream) {
                ret = -1;
                goto end_unlock;
        }

        size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
                        payload_size);
        if (size_ret < payload_size) {
                ERR("Relay error writing metadata on file");
                ret = -1;
                goto end_unlock;
        }

        ret = write_padding_to_file(metadata_stream->fd,
                        be32toh(metadata_struct->padding_size));
        if (ret < 0) {
                goto end_unlock;
        }
        metadata_stream->ctf_trace->metadata_received +=
                payload_size + be32toh(metadata_struct->padding_size);

        DBG2("Relay metadata written");

end_unlock:
        rcu_read_unlock();
end:
        return ret;
}

/*
 * relay_send_version: send relayd version number
 */
static
int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd, struct lttng_ht *sessions_ht)
{
        int ret;
        struct lttcomm_relayd_version reply, msg;

        assert(cmd);

        cmd->version_check_done = 1;

        /* Get version from the other side. */
        ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
        if (ret < 0 || ret != sizeof(msg)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay failed to receive the version values.");
                }
                ret = -1;
                goto end;
        }

        reply.major = RELAYD_VERSION_COMM_MAJOR;
        reply.minor = RELAYD_VERSION_COMM_MINOR;

        /* Major versions must be the same */
        if (reply.major != be32toh(msg.major)) {
                DBG("Incompatible major versions (%u vs %u), deleting session",
                                reply.major, be32toh(msg.major));
                relay_delete_session(cmd, sessions_ht);
                ret = 0;
                goto end;
        }

        cmd->major = reply.major;
        /* We adapt to the lowest compatible version */
        if (reply.minor <= be32toh(msg.minor)) {
                cmd->minor = reply.minor;
        } else {
                cmd->minor = be32toh(msg.minor);
        }

        reply.major = htobe32(reply.major);
        reply.minor = htobe32(reply.minor);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
                        sizeof(struct lttcomm_relayd_version), 0);
        if (ret < 0) {
                ERR("Relay sending version");
        }

        DBG("Version check done using protocol %u.%u", cmd->major,
                        cmd->minor);

end:
        return ret;
}

/*
 * Check for data pending for a given stream id from the session daemon.
 */
static
int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        struct relay_session *session = cmd->session;
        struct lttcomm_relayd_data_pending msg;
        struct lttcomm_relayd_generic_reply reply;
        struct relay_stream *stream;
        int ret;
        uint64_t last_net_seq_num, stream_id;

        DBG("Data pending command received");

        if (!session || cmd->version_check_done == 0) {
                ERR("Trying to check for data before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
        if (ret < sizeof(msg)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid data_pending struct size : %d",
                                        ret);
                }
                ret = -1;
                goto end_no_session;
        }

        stream_id = be64toh(msg.stream_id);
        last_net_seq_num = be64toh(msg.last_net_seq_num);

        rcu_read_lock();
        stream = relay_stream_find_by_id(stream_id);
        if (stream == NULL) {
                ret = -1;
                goto end_unlock;
        }

        DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
                        " and last_seq %" PRIu64, stream_id, stream->prev_seq,
                        last_net_seq_num);

        /* Avoid wrapping issue */
        if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
                /* Data has in fact been written and is NOT pending */
                ret = 0;
        } else {
                /* Data still being streamed thus pending */
                ret = 1;
        }

        /* Pending check is now done. */
        stream->data_pending_check_done = 1;

end_unlock:
        rcu_read_unlock();

        reply.ret_code = htobe32(ret);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
                ERR("Relay data pending ret code failed");
        }

end_no_session:
        return ret;
}

/*
 * Wait for the control socket to reach a quiescent state.
 *
 * Note that for now, when receiving this command from the session daemon, this
 * means that every subsequent commands or data received on the control socket
 * has been handled. So, this is why we simply return OK here.
 */
static
int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret;
        uint64_t stream_id;
        struct relay_stream *stream;
        struct lttng_ht_iter iter;
        struct lttcomm_relayd_quiescent_control msg;
        struct lttcomm_relayd_generic_reply reply;

        DBG("Checking quiescent state on control socket");

        if (!cmd->session || cmd->version_check_done == 0) {
                ERR("Trying to check for data before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
        if (ret < sizeof(msg)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid begin data_pending struct size: %d",
                                        ret);
                }
                ret = -1;
                goto end_no_session;
        }

        stream_id = be64toh(msg.stream_id);

        rcu_read_lock();
        cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
                        stream_n.node) {
                if (stream->stream_handle == stream_id) {
                        stream->data_pending_check_done = 1;
                        DBG("Relay quiescent control pending flag set to %" PRIu64,
                                        stream_id);
                        break;
                }
        }
        rcu_read_unlock();

        reply.ret_code = htobe32(LTTNG_OK);
        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
                ERR("Relay data quiescent control ret code failed");
        }

end_no_session:
        return ret;
}

/*
 * Initialize a data pending command. This means that a client is about to ask
 * for data pending for each stream he/she holds. Simply iterate over all
 * streams of a session and set the data_pending_check_done flag.
 *
 * This command returns to the client a LTTNG_OK code.
 */
static
int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret;
        struct lttng_ht_iter iter;
        struct lttcomm_relayd_begin_data_pending msg;
        struct lttcomm_relayd_generic_reply reply;
        struct relay_stream *stream;
        uint64_t session_id;

        assert(recv_hdr);
        assert(cmd);

        DBG("Init streams for data pending");

        if (!cmd->session || cmd->version_check_done == 0) {
                ERR("Trying to check for data before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
        if (ret < sizeof(msg)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid begin data_pending struct size: %d",
                                        ret);
                }
                ret = -1;
                goto end_no_session;
        }

        session_id = be64toh(msg.session_id);

        /*
         * Iterate over all streams to set the begin data pending flag. For now, the
         * streams are indexed by stream handle so we have to iterate over all
         * streams to find the one associated with the right session_id.
         */
        rcu_read_lock();
        cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
                        stream_n.node) {
                if (stream->session->id == session_id) {
                        stream->data_pending_check_done = 0;
                        DBG("Set begin data pending flag to stream %" PRIu64,
                                        stream->stream_handle);
                }
        }
        rcu_read_unlock();

        /* All good, send back reply. */
        reply.ret_code = htobe32(LTTNG_OK);

        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
                ERR("Relay begin data pending send reply failed");
        }

end_no_session:
        return ret;
}

/*
 * End data pending command. This will check, for a given session id, if each
 * stream associated with it has its data_pending_check_done flag set. If not,
 * this means that the client lost track of the stream but the data is still
 * being streamed on our side. In this case, we inform the client that data is
 * inflight.
 *
 * Return to the client if there is data in flight or not with a ret_code.
 */
static
int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret;
        struct lttng_ht_iter iter;
        struct lttcomm_relayd_end_data_pending msg;
        struct lttcomm_relayd_generic_reply reply;
        struct relay_stream *stream;
        uint64_t session_id;
        uint32_t is_data_inflight = 0;

        assert(recv_hdr);
        assert(cmd);

        DBG("End data pending command");

        if (!cmd->session || cmd->version_check_done == 0) {
                ERR("Trying to check for data before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0);
        if (ret < sizeof(msg)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid end data_pending struct size: %d",
                                        ret);
                }
                ret = -1;
                goto end_no_session;
        }

        session_id = be64toh(msg.session_id);

        /* Iterate over all streams to see if the begin data pending flag is set. */
        rcu_read_lock();
        cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
                        stream_n.node) {
                if (stream->session->id == session_id &&
                                !stream->data_pending_check_done) {
                        is_data_inflight = 1;
                        DBG("Data is still in flight for stream %" PRIu64,
                                        stream->stream_handle);
                        break;
                }
        }
        rcu_read_unlock();

        /* All good, send back reply. */
        reply.ret_code = htobe32(is_data_inflight);

        ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
                ERR("Relay end data pending send reply failed");
        }

end_no_session:
        return ret;
}

/*
 * Receive an index for a specific stream.
 *
 * Return 0 on success else a negative value.
 */
static
int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret, send_ret, index_created = 0;
        struct relay_session *session = cmd->session;
        struct lttcomm_relayd_index index_info;
        struct relay_index *index, *wr_index = NULL;
        struct lttcomm_relayd_generic_reply reply;
        struct relay_stream *stream;
        uint64_t net_seq_num;

        assert(cmd);

        DBG("Relay receiving index");

        if (!session || cmd->version_check_done == 0) {
                ERR("Trying to close a stream before version check");
                ret = -1;
                goto end_no_session;
        }

        ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info,
                        sizeof(index_info), 0);
        if (ret < sizeof(index_info)) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Relay didn't receive valid index struct size : %d", ret);
                }
                ret = -1;
                goto end_no_session;
        }

        net_seq_num = be64toh(index_info.net_seq_num);

        rcu_read_lock();
        stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id));
        if (!stream) {
                ret = -1;
                goto end_rcu_unlock;
        }

        /* Live beacon handling */
        if (index_info.packet_size == 0) {
                DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);

                /*
                 * Only flag a stream inactive when it has already received data.
                 */
                if (stream->total_index_received > 0) {
                        stream->beacon_ts_end = be64toh(index_info.timestamp_end);
                }
                ret = 0;
                goto end_rcu_unlock;
        } else {
                stream->beacon_ts_end = -1ULL;
        }

        index = relay_index_find(stream->stream_handle, net_seq_num);
        if (!index) {
                /* A successful creation will add the object to the HT. */
                index = relay_index_create(stream->stream_handle, net_seq_num);
                if (!index) {
                        goto end_rcu_unlock;
                }
                index_created = 1;
        }

        copy_index_control_data(index, &index_info);

        if (index_created) {
                /*
                 * Try to add the relay index object to the hash table. If an object
                 * already exist, destroy back the index created, set the data in this
                 * object and write it on disk.
                 */
                relay_index_add(index, &wr_index);
                if (wr_index) {
                        copy_index_control_data(wr_index, &index_info);
                        free(index);
                }
        } else {
                /* The index already exists so write it on disk. */
                wr_index = index;
        }

        /* Do we have a writable ready index to write on disk. */
        if (wr_index) {
                /* Starting at 2.4, create the index file if none available. */
                if (cmd->minor >= 4 && stream->index_fd < 0) {
                        ret = index_create_file(stream->path_name, stream->channel_name,
                                        relayd_uid, relayd_gid, stream->tracefile_size,
                                        stream->tracefile_count_current);
                        if (ret < 0) {
                                goto end_rcu_unlock;
                        }
                        stream->index_fd = ret;
                }

                ret = relay_index_write(wr_index->fd, wr_index);
                if (ret < 0) {
                        goto end_rcu_unlock;
                }
                stream->total_index_received++;
        }

end_rcu_unlock:
        rcu_read_unlock();

        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_UNK);
        } else {
                reply.ret_code = htobe32(LTTNG_OK);
        }
        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (send_ret < 0) {
                ERR("Relay sending close index id reply");
                ret = send_ret;
        }

end_no_session:
        return ret;
}

/*
 * Receive the streams_sent message.
 *
 * Return 0 on success else a negative value.
 */
static
int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd)
{
        int ret, send_ret;
        struct lttcomm_relayd_generic_reply reply;

        assert(cmd);

        DBG("Relay receiving streams_sent");

        if (!cmd->session || cmd->version_check_done == 0) {
                ERR("Trying to close a stream before version check");
                ret = -1;
                goto end_no_session;
        }

        /*
         * Flag every pending stream in the connection recv list that they are
         * ready to be used by the viewer.
         */
        set_viewer_ready_flag(cmd);

        /*
         * Inform the viewer that there are new streams in the session.
         */
        uatomic_set(&cmd->session->new_streams, 1);

        reply.ret_code = htobe32(LTTNG_OK);
        send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
        if (send_ret < 0) {
                ERR("Relay sending sent_stream reply");
                ret = send_ret;
        } else {
                /* Success. */
                ret = 0;
        }

end_no_session:
        return ret;
}

/*
 * Process the commands received on the control socket
 */
static
int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
                struct relay_command *cmd, struct relay_local_data *ctx)
{
        int ret = 0;

        switch (be32toh(recv_hdr->cmd)) {
        case RELAYD_CREATE_SESSION:
                ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht);
                break;
        case RELAYD_ADD_STREAM:
                ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht);
                break;
        case RELAYD_START_DATA:
                ret = relay_start(recv_hdr, cmd);
                break;
        case RELAYD_SEND_METADATA:
                ret = relay_recv_metadata(recv_hdr, cmd);
                break;
        case RELAYD_VERSION:
                ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht);
                break;
        case RELAYD_CLOSE_STREAM:
                ret = relay_close_stream(recv_hdr, cmd);
                break;
        case RELAYD_DATA_PENDING:
                ret = relay_data_pending(recv_hdr, cmd);
                break;
        case RELAYD_QUIESCENT_CONTROL:
                ret = relay_quiescent_control(recv_hdr, cmd);
                break;
        case RELAYD_BEGIN_DATA_PENDING:
                ret = relay_begin_data_pending(recv_hdr, cmd);
                break;
        case RELAYD_END_DATA_PENDING:
                ret = relay_end_data_pending(recv_hdr, cmd);
                break;
        case RELAYD_SEND_INDEX:
                ret = relay_recv_index(recv_hdr, cmd);
                break;
        case RELAYD_STREAMS_SENT:
                ret = relay_streams_sent(recv_hdr, cmd);
                break;
        case RELAYD_UPDATE_SYNC_INFO:
        default:
                ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
                relay_unknown_command(cmd);
                ret = -1;
                goto end;
        }

end:
        return ret;
}

/*
 * Handle index for a data stream.
 *
 * RCU read side lock MUST be acquired.
 *
 * Return 0 on success else a negative value.
 */
static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
                int rotate_index)
{
        int ret = 0, index_created = 0;
        uint64_t stream_id, data_offset;
        struct relay_index *index, *wr_index = NULL;

        assert(stream);

        stream_id = stream->stream_handle;
        /* Get data offset because we are about to update the index. */
        data_offset = htobe64(stream->tracefile_size_current);

        /*
         * Lookup for an existing index for that stream id/sequence number. If on
         * exists, the control thread already received the data for it thus we need
         * to write it on disk.
         */
        index = relay_index_find(stream_id, net_seq_num);
        if (!index) {
                /* A successful creation will add the object to the HT. */
                index = relay_index_create(stream_id, net_seq_num);
                if (!index) {
                        ret = -1;
                        goto error;
                }
                index_created = 1;
        }

        if (rotate_index || stream->index_fd < 0) {
                index->to_close_fd = stream->index_fd;
                ret = index_create_file(stream->path_name, stream->channel_name,
                                relayd_uid, relayd_gid, stream->tracefile_size,
                                stream->tracefile_count_current);
                if (ret < 0) {
                        /* This will close the stream's index fd if one. */
                        relay_index_free_safe(index);
                        goto error;
                }
                stream->index_fd = ret;
        }
        index->fd = stream->index_fd;
        index->index_data.offset = data_offset;

        if (index_created) {
                /*
                 * Try to add the relay index object to the hash table. If an object
                 * already exist, destroy back the index created and set the data.
                 */
                relay_index_add(index, &wr_index);
                if (wr_index) {
                        /* Copy back data from the created index. */
                        wr_index->fd = index->fd;
                        wr_index->to_close_fd = index->to_close_fd;
                        wr_index->index_data.offset = data_offset;
                        free(index);
                }
        } else {
                /* The index already exists so write it on disk. */
                wr_index = index;
        }

        /* Do we have a writable ready index to write on disk. */
        if (wr_index) {
                ret = relay_index_write(wr_index->fd, wr_index);
                if (ret < 0) {
                        goto error;
                }
                stream->total_index_received++;
        }

error:
        return ret;
}

/*
 * relay_process_data: Process the data received on the data socket
 */
static
int relay_process_data(struct relay_command *cmd)
{
        int ret = 0, rotate_index = 0;
        ssize_t size_ret;
        struct relay_stream *stream;
        struct lttcomm_relayd_data_hdr data_hdr;
        uint64_t stream_id;
        uint64_t net_seq_num;
        uint32_t data_size;

        ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
                        sizeof(struct lttcomm_relayd_data_hdr), 0);
        if (ret <= 0) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                } else {
                        ERR("Unable to receive data header on sock %d", cmd->sock->fd);
                }
                ret = -1;
                goto end;
        }

        stream_id = be64toh(data_hdr.stream_id);

        rcu_read_lock();
        stream = relay_stream_find_by_id(stream_id);
        if (!stream) {
                ret = -1;
                goto end_rcu_unlock;
        }

        data_size = be32toh(data_hdr.data_size);
        if (data_buffer_size < data_size) {
                char *tmp_data_ptr;

                tmp_data_ptr = realloc(data_buffer, data_size);
                if (!tmp_data_ptr) {
                        ERR("Allocating data buffer");
                        free(data_buffer);
                        ret = -1;
                        goto end_rcu_unlock;
                }
                data_buffer = tmp_data_ptr;
                data_buffer_size = data_size;
        }
        memset(data_buffer, 0, data_size);

        net_seq_num = be64toh(data_hdr.net_seq_num);

        DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
                data_size, stream_id, net_seq_num);
        ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0);
        if (ret <= 0) {
                if (ret == 0) {
                        /* Orderly shutdown. Not necessary to print an error. */
                        DBG("Socket %d did an orderly shutdown", cmd->sock->fd);
                }
                ret = -1;
                goto end_rcu_unlock;
        }

        /* Check if a rotation is needed. */
        if (stream->tracefile_size > 0 &&
                        (stream->tracefile_size_current + data_size) >
                        stream->tracefile_size) {
                struct relay_viewer_stream *vstream;
                uint64_t new_id;

                new_id = (stream->tracefile_count_current + 1) %
                        stream->tracefile_count;
                /*
                 * When we wrap-around back to 0, we start overwriting old
                 * trace data.
                 */
                if (!stream->tracefile_overwrite && new_id == 0) {
                        stream->tracefile_overwrite = 1;
                }
                pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
                if (stream->tracefile_overwrite) {
                        stream->oldest_tracefile_id =
                                (stream->oldest_tracefile_id + 1) %
                                stream->tracefile_count;
                }
                vstream = viewer_stream_find_by_id(stream->stream_handle);
                if (vstream) {
                        /*
                         * The viewer is reading a file about to be
                         * overwritten. Close the FDs it is
                         * currently using and let it handle the fault.
                         */
                        if (vstream->tracefile_count_current == new_id) {
                                pthread_mutex_lock(&vstream->overwrite_lock);
                                vstream->abort_flag = 1;
                                pthread_mutex_unlock(&vstream->overwrite_lock);
                                DBG("Streaming side setting abort_flag on stream %s_%lu\n",
                                                stream->channel_name, new_id);
                        } else if (vstream->tracefile_count_current ==
                                        stream->tracefile_count_current) {
                                /*
                                 * The reader and writer were in the
                                 * same trace file, inform the viewer
                                 * that no new index will ever be added
                                 * to this file.
                                 */
                                vstream->close_write_flag = 1;
                        }
                }
                ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
                                stream->tracefile_size, stream->tracefile_count,
                                relayd_uid, relayd_gid, stream->fd,
                                &(stream->tracefile_count_current), &stream->fd);
                stream->total_index_received = 0;
                pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
                if (ret < 0) {
                        ERR("Rotating stream output file");
                        goto end_rcu_unlock;
                }
                /* Reset current size because we just perform a stream rotation. */
                stream->tracefile_size_current = 0;
                rotate_index = 1;
        }

        /*
         * Index are handled in protocol version 2.4 and above. Also, snapshot and
         * index are NOT supported.
         */
        if (stream->session->minor >= 4 && !stream->session->snapshot) {
                ret = handle_index_data(stream, net_seq_num, rotate_index);
                if (ret < 0) {
                        goto end_rcu_unlock;
                }
        }

        /* Write data to stream output fd. */
        size_ret = lttng_write(stream->fd, data_buffer, data_size);
        if (size_ret < data_size) {
                ERR("Relay error writing data to file");
                ret = -1;
                goto end_rcu_unlock;
        }

        DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
                        ret, stream->stream_handle);

        ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
        if (ret < 0) {
                goto end_rcu_unlock;
        }
        stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);

        stream->prev_seq = net_seq_num;

        /* Check if we need to close the FD */
        if (close_stream_check(stream)) {
                destroy_stream(stream);
        }

end_rcu_unlock:
        rcu_read_unlock();
end:
        return ret;
}

static
void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
{
        int ret;

        lttng_poll_del(events, pollfd);

        ret = close(pollfd);
        if (ret < 0) {
                ERR("Closing pollfd %d", pollfd);
        }
}

static
int relay_add_connection(int fd, struct lttng_poll_event *events,
                struct lttng_ht *relay_connections_ht)
{
        struct relay_command *relay_connection;
        ssize_t ret;

        relay_connection = zmalloc(sizeof(struct relay_command));
        if (relay_connection == NULL) {
                PERROR("Relay command zmalloc");
                goto error;
        }
        ret = lttng_read(fd, relay_connection, sizeof(struct relay_command));
        if (ret < sizeof(struct relay_command)) {
                PERROR("read relay cmd pipe");
                goto error_read;
        }
        CDS_INIT_LIST_HEAD(&relay_connection->recv_head);

        /*
         * Only used by the control side and the reference is copied inside each
         * stream from that connection. Thus a destroy HT must be done after every
         * stream has been destroyed.
         */
        if (relay_connection->type == RELAY_CONTROL) {
                relay_connection->ctf_traces_ht = lttng_ht_new(0,
                                LTTNG_HT_TYPE_STRING);
                if (!relay_connection->ctf_traces_ht) {
                        goto error_read;
                }
        }

        lttng_ht_node_init_ulong(&relay_connection->sock_n,
                        (unsigned long) relay_connection->sock->fd);
        rcu_read_lock();
        lttng_ht_add_unique_ulong(relay_connections_ht,
                        &relay_connection->sock_n);
        rcu_read_unlock();
        return lttng_poll_add(events,
                        relay_connection->sock->fd,
                        LPOLLIN | LPOLLRDHUP);

error_read:
        free(relay_connection);
error:
        return -1;
}

static
void deferred_free_connection(struct rcu_head *head)
{
        struct relay_command *relay_connection =
                caa_container_of(head, struct relay_command, rcu_node);

        lttcomm_destroy_sock(relay_connection->sock);
        free(relay_connection);
}

static
void relay_del_connection(struct lttng_ht *relay_connections_ht,
                struct lttng_ht_iter *iter, struct relay_command *relay_connection,
                struct lttng_ht *sessions_ht)
{
        int ret;

        ret = lttng_ht_del(relay_connections_ht, iter);
        assert(!ret);

        if (relay_connection->type == RELAY_CONTROL) {
                struct relay_stream_recv_handle *node, *tmp_node;

                relay_delete_session(relay_connection, sessions_ht);
                lttng_ht_destroy(relay_connection->ctf_traces_ht);

                /* Clean up recv list. */
                cds_list_for_each_entry_safe(node, tmp_node,
                                &relay_connection->recv_head, node) {
                        cds_list_del(&node->node);
                        free(node);
                }
        }

        call_rcu(&relay_connection->rcu_node, deferred_free_connection);
}

/*
 * This thread does the actual work
 */
static
void *relay_thread_worker(void *data)
{
        int ret, err = -1, last_seen_data_fd = -1;
        uint32_t nb_fd;
        struct relay_command *relay_connection;
        struct lttng_poll_event events;
        struct lttng_ht *relay_connections_ht;
        struct lttng_ht_node_ulong *node;
        struct lttng_ht_iter iter;
        struct lttcomm_relayd_hdr recv_hdr;
        struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
        struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;

        DBG("[thread] Relay worker started");

        rcu_register_thread();

        health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);

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

        health_code_update();

        /* table of connections indexed on socket */
        relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        if (!relay_connections_ht) {
                goto relay_connections_ht_error;
        }

        /* Tables of received indexes indexed by index handle and net_seq_num. */
        indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
        if (!indexes_ht) {
                goto indexes_ht_error;
        }

        ret = create_thread_poll_set(&events, 2);
        if (ret < 0) {
                goto error_poll_create;
        }

        ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
        if (ret < 0) {
                goto error;
        }

restart:
        while (1) {
                int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;

                health_code_update();

                /* Infinite blocking call, waiting for transmission */
                DBG3("Relayd worker thread polling...");
                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;

                /*
                 * Process control. The control connection is prioritised so we don't
                 * starve it with high throughout put tracing data on the data
                 * connection.
                 */
                for (i = 0; i < nb_fd; i++) {
                        /* Fetch once the poll data */
                        uint32_t revents = LTTNG_POLL_GETEV(&events, i);
                        int pollfd = LTTNG_POLL_GETFD(&events, i);

                        health_code_update();

                        /* Thread quit pipe has been closed. Killing thread. */
                        ret = check_thread_quit_pipe(pollfd, revents);
                        if (ret) {
                                err = 0;
                                goto exit;
                        }

                        /* Inspect the relay cmd pipe for new connection */
                        if (pollfd == relay_cmd_pipe[0]) {
                                if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
                                        ERR("Relay pipe error");
                                        goto error;
                                } else if (revents & LPOLLIN) {
                                        DBG("Relay command received");
                                        ret = relay_add_connection(relay_cmd_pipe[0],
                                                        &events, relay_connections_ht);
                                        if (ret < 0) {
                                                goto error;
                                        }
                                }
                        } else if (revents) {
                                rcu_read_lock();
                                lttng_ht_lookup(relay_connections_ht,
                                                (void *)((unsigned long) pollfd),
                                                &iter);
                                node = lttng_ht_iter_get_node_ulong(&iter);
                                if (node == NULL) {
                                        DBG2("Relay sock %d not found", pollfd);
                                        rcu_read_unlock();
                                        goto error;
                                }
                                relay_connection = caa_container_of(node,
                                                struct relay_command, sock_n);

                                if (revents & (LPOLLERR)) {
                                        ERR("POLL ERROR");
                                        relay_cleanup_poll_connection(&events, pollfd);
                                        relay_del_connection(relay_connections_ht,
                                                        &iter, relay_connection, sessions_ht);
                                        if (last_seen_data_fd == pollfd) {
                                                last_seen_data_fd = last_notdel_data_fd;
                                        }
                                } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
                                        DBG("Socket %d hung up", pollfd);
                                        relay_cleanup_poll_connection(&events, pollfd);
                                        relay_del_connection(relay_connections_ht,
                                                        &iter, relay_connection, sessions_ht);
                                        if (last_seen_data_fd == pollfd) {
                                                last_seen_data_fd = last_notdel_data_fd;
                                        }
                                } else if (revents & LPOLLIN) {
                                        /* control socket */
                                        if (relay_connection->type == RELAY_CONTROL) {
                                                ret = relay_connection->sock->ops->recvmsg(
                                                                relay_connection->sock, &recv_hdr,
                                                                sizeof(struct lttcomm_relayd_hdr), 0);
                                                /* connection closed */
                                                if (ret <= 0) {
                                                        relay_cleanup_poll_connection(&events, pollfd);
                                                        relay_del_connection(relay_connections_ht,
                                                                        &iter, relay_connection, sessions_ht);
                                                        DBG("Control connection closed with %d", pollfd);
                                                } else {
                                                        if (relay_connection->session) {
                                                                DBG2("Relay worker receiving data for session : %" PRIu64,
                                                                                relay_connection->session->id);
                                                        }
                                                        ret = relay_process_control(&recv_hdr,
                                                                        relay_connection, relay_ctx);
                                                        if (ret < 0) {
                                                                /* Clear the session on error. */
                                                                relay_cleanup_poll_connection(&events, pollfd);
                                                                relay_del_connection(relay_connections_ht,
                                                                                &iter, relay_connection, sessions_ht);
                                                                DBG("Connection closed with %d", pollfd);
                                                        }
                                                        seen_control = 1;
                                                }
                                        } else {
                                                /*
                                                 * Flag the last seen data fd not deleted. It will be
                                                 * used as the last seen fd if any fd gets deleted in
                                                 * this first loop.
                                                 */
                                                last_notdel_data_fd = pollfd;
                                        }
                                }
                                rcu_read_unlock();
                        }
                }

                /*
                 * The last loop handled a control request, go back to poll to make
                 * sure we prioritise the control socket.
                 */
                if (seen_control) {
                        continue;
                }

                if (last_seen_data_fd >= 0) {
                        for (i = 0; i < nb_fd; i++) {
                                int pollfd = LTTNG_POLL_GETFD(&events, i);

                                health_code_update();

                                if (last_seen_data_fd == pollfd) {
                                        idx = i;
                                        break;
                                }
                        }
                }

                /* Process data connection. */
                for (i = idx + 1; i < nb_fd; i++) {
                        /* Fetch the poll data. */
                        uint32_t revents = LTTNG_POLL_GETEV(&events, i);
                        int pollfd = LTTNG_POLL_GETFD(&events, i);

                        health_code_update();

                        /* Skip the command pipe. It's handled in the first loop. */
                        if (pollfd == relay_cmd_pipe[0]) {
                                continue;
                        }

                        if (revents) {
                                rcu_read_lock();
                                lttng_ht_lookup(relay_connections_ht,
                                                (void *)((unsigned long) pollfd),
                                                &iter);
                                node = lttng_ht_iter_get_node_ulong(&iter);
                                if (node == NULL) {
                                        /* Skip it. Might be removed before. */
                                        rcu_read_unlock();
                                        continue;
                                }
                                relay_connection = caa_container_of(node,
                                                struct relay_command, sock_n);

                                if (revents & LPOLLIN) {
                                        if (relay_connection->type != RELAY_DATA) {
                                                continue;
                                        }

                                        ret = relay_process_data(relay_connection);
                                        /* connection closed */
                                        if (ret < 0) {
                                                relay_cleanup_poll_connection(&events, pollfd);
                                                relay_del_connection(relay_connections_ht,
                                                                &iter, relay_connection, sessions_ht);
                                                DBG("Data connection closed with %d", pollfd);
                                                /*
                                                 * Every goto restart call sets the last seen fd where
                                                 * here we don't really care since we gracefully
                                                 * continue the loop after the connection is deleted.
                                                 */
                                        } else {
                                                /* Keep last seen port. */
                                                last_seen_data_fd = pollfd;
                                                rcu_read_unlock();
                                                goto restart;
                                        }
                                }
                                rcu_read_unlock();
                        }
                }
                last_seen_data_fd = -1;
        }

        /* Normal exit, no error */
        ret = 0;

exit:
error:
        lttng_poll_clean(&events);

        /* empty the hash table and free the memory */
        rcu_read_lock();
        cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
                health_code_update();

                node = lttng_ht_iter_get_node_ulong(&iter);
                if (node) {
                        relay_connection = caa_container_of(node,
                                        struct relay_command, sock_n);
                        relay_del_connection(relay_connections_ht,
                                        &iter, relay_connection, sessions_ht);
                }
        }
        rcu_read_unlock();
error_poll_create:
        lttng_ht_destroy(indexes_ht);
indexes_ht_error:
        lttng_ht_destroy(relay_connections_ht);
relay_connections_ht_error:
        /* Close relay cmd pipes */
        utils_close_pipe(relay_cmd_pipe);
        if (err) {
                DBG("Thread exited with error");
        }
        DBG("Worker thread cleanup complete");
        free(data_buffer);
error_testpoint:
        if (err) {
                health_error();
                ERR("Health error occurred in %s", __func__);
        }
        health_unregister(health_relayd);
        rcu_unregister_thread();
        stop_threads();
        return NULL;
}

/*
 * Create the relay command pipe to wake thread_manage_apps.
 * Closed in cleanup().
 */
static int create_relay_cmd_pipe(void)
{
        int ret;

        ret = utils_create_pipe_cloexec(relay_cmd_pipe);

        return ret;
}

/*
 * main
 */
int main(int argc, char **argv)
{
        int ret = 0;
        void *status;
        struct relay_local_data *relay_ctx;

        /* Parse arguments */
        progname = argv[0];
        if ((ret = set_options(argc, argv)) < 0) {
                goto exit;
        }

        if ((ret = set_signal_handler()) < 0) {
                goto exit;
        }

        /* Try to create directory if -o, --output is specified. */
        if (opt_output_path) {
                if (*opt_output_path != '/') {
                        ERR("Please specify an absolute path for -o, --output PATH");
                        goto exit;
                }

                ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
                if (ret < 0) {
                        ERR("Unable to create %s", opt_output_path);
                        goto exit;
                }
        }

        /* Daemonize */
        if (opt_daemon || opt_background) {
                int i;

                ret = lttng_daemonize(&child_ppid, &recv_child_signal,
                        !opt_background);
                if (ret < 0) {
                        goto exit;
                }

                /*
                 * We are in the child. Make sure all other file
                 * descriptors are closed, in case we are called with
                 * more opened file descriptors than the standard ones.
                 */
                for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
                        (void) close(i);
                }
        }

        /* Create thread quit pipe */
        if ((ret = init_thread_quit_pipe()) < 0) {
                goto error;
        }

        /* We need those values for the file/dir creation. */
        relayd_uid = getuid();
        relayd_gid = getgid();

        /* Check if daemon is UID = 0 */
        if (relayd_uid == 0) {
                if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
                        ERR("Need to be root to use ports < 1024");
                        ret = -1;
                        goto exit;
                }
        }

        /* Setup the thread apps communication pipe. */
        if ((ret = create_relay_cmd_pipe()) < 0) {
                goto exit;
        }

        /* Init relay command queue. */
        cds_wfq_init(&relay_cmd_queue.queue);

        /* Set up max poll set size */
        lttng_poll_set_max_size();

        /* Initialize communication library */
        lttcomm_init();
        lttcomm_inet_init();

        relay_ctx = zmalloc(sizeof(struct relay_local_data));
        if (!relay_ctx) {
                PERROR("relay_ctx");
                goto exit;
        }

        /* tables of sessions indexed by session ID */
        relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        if (!relay_ctx->sessions_ht) {
                goto exit_relay_ctx_sessions;
        }

        /* tables of streams indexed by stream ID */
        relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
        if (!relay_streams_ht) {
                goto exit_relay_ctx_streams;
        }

        /* tables of streams indexed by stream ID */
        viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
        if (!viewer_streams_ht) {
                goto exit_relay_ctx_viewer_streams;
        }

        /* Initialize thread health monitoring */
        health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
        if (!health_relayd) {
                PERROR("health_app_create error");
                goto exit_health_app_create;
        }

        ret = utils_create_pipe(health_quit_pipe);
        if (ret < 0) {
                goto error_health_pipe;
        }

        /* Create thread to manage the client socket */
        ret = pthread_create(&health_thread, NULL,
                        thread_manage_health, (void *) NULL);
        if (ret != 0) {
                PERROR("pthread_create health");
                goto health_error;
        }

        /* Setup the dispatcher thread */
        ret = pthread_create(&dispatcher_thread, NULL,
                        relay_thread_dispatcher, (void *) NULL);
        if (ret != 0) {
                PERROR("pthread_create dispatcher");
                goto exit_dispatcher;
        }

        /* Setup the worker thread */
        ret = pthread_create(&worker_thread, NULL,
                        relay_thread_worker, (void *) relay_ctx);
        if (ret != 0) {
                PERROR("pthread_create worker");
                goto exit_worker;
        }

        /* Setup the listener thread */
        ret = pthread_create(&listener_thread, NULL,
                        relay_thread_listener, (void *) NULL);
        if (ret != 0) {
                PERROR("pthread_create listener");
                goto exit_listener;
        }

        ret = live_start_threads(live_uri, relay_ctx);
        if (ret != 0) {
                ERR("Starting live viewer threads");
                goto exit_live;
        }

exit_live:
        ret = pthread_join(listener_thread, &status);
        if (ret != 0) {
                PERROR("pthread_join");
                goto error;     /* join error, exit without cleanup */
        }

exit_listener:
        ret = pthread_join(worker_thread, &status);
        if (ret != 0) {
                PERROR("pthread_join");
                goto error;     /* join error, exit without cleanup */
        }

exit_worker:
        ret = pthread_join(dispatcher_thread, &status);
        if (ret != 0) {
                PERROR("pthread_join");
                goto error;     /* join error, exit without cleanup */
        }

exit_dispatcher:
        ret = pthread_join(health_thread, &status);
        if (ret != 0) {
                PERROR("pthread_join health thread");
                goto error;     /* join error, exit without cleanup */
        }

        /*
         * Stop live threads only after joining other threads.
         */
        live_stop_threads();

health_error:
        utils_close_pipe(health_quit_pipe);

error_health_pipe:
        health_app_destroy(health_relayd);

exit_health_app_create:
        lttng_ht_destroy(viewer_streams_ht);

exit_relay_ctx_viewer_streams:
        lttng_ht_destroy(relay_streams_ht);

exit_relay_ctx_streams:
        lttng_ht_destroy(relay_ctx->sessions_ht);

exit_relay_ctx_sessions:
        free(relay_ctx);

exit:
        cleanup();
        if (!ret) {
                exit(EXIT_SUCCESS);
        }

error:
        exit(EXIT_FAILURE);
}