Add missing listener threads data vs fork() protection

[lttng-ust.git] / libust / tracectl.c

/* Copyright (C) 2009  Pierre-Marc Fournier
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301 USA
 */

/* This file contains the implementation of the UST listener thread, which
 * receives trace control commands. It also coordinates the initialization of
 * libust.
 */

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <pthread.h>
#include <signal.h>
#include <sys/epoll.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <fcntl.h>
#include <poll.h>
#include <regex.h>
#include <urcu/uatomic_arch.h>
#include <urcu/list.h>

#include <ust/marker.h>
#include <ust/tracepoint.h>
#include <ust/tracectl.h>
#include <ust/clock.h>
#include "tracer.h"
#include "usterr.h"
#include "ustcomm.h"
#include "buffers.h"
#include "marker-control.h"

/* This should only be accessed by the constructor, before the creation
 * of the listener, and then only by the listener.
 */
s64 pidunique = -1LL;

/* The process pid is used to detect a non-traceable fork
 * and allow the non-traceable fork to be ignored
 * by destructor sequences in libust
 */
static pid_t processpid = 0;

static struct ustcomm_header _receive_header;
static struct ustcomm_header *receive_header = &_receive_header;
static char receive_buffer[USTCOMM_BUFFER_SIZE];
static char send_buffer[USTCOMM_BUFFER_SIZE];

static int epoll_fd;

/*
 * Listener thread data vs fork() protection mechanism. Ensures that no listener
 * thread mutexes and data structures are being concurrently modified or held by
 * other threads when fork() is executed.
 */
static pthread_mutex_t listener_thread_data_mutex = PTHREAD_MUTEX_INITIALIZER;

/* Mutex protecting listen_sock. Nests inside listener_thread_data_mutex. */
static pthread_mutex_t listen_sock_mutex = PTHREAD_MUTEX_INITIALIZER;
static struct ustcomm_sock *listen_sock;

extern struct chan_info_struct chan_infos[];

static struct cds_list_head open_buffers_list = CDS_LIST_HEAD_INIT(open_buffers_list);

static struct cds_list_head ust_socks = CDS_LIST_HEAD_INIT(ust_socks);

/* volatile because shared between the listener and the main thread */
int buffers_to_export = 0;

int ust_clock_source;

static long long make_pidunique(void)
{
        s64 retval;
        struct timeval tv;

        gettimeofday(&tv, NULL);

        retval = tv.tv_sec;
        retval <<= 32;
        retval |= tv.tv_usec;

        return retval;
}

static void print_markers(FILE *fp)
{
        struct marker_iter iter;

        lock_markers();
        marker_iter_reset(&iter);
        marker_iter_start(&iter);

        while (iter.marker) {
                fprintf(fp, "marker: %s/%s %d \"%s\" %p\n",
                        (*iter.marker)->channel,
                        (*iter.marker)->name,
                        (int)imv_read((*iter.marker)->state),
                        (*iter.marker)->format,
                        (*iter.marker)->location);
                marker_iter_next(&iter);
        }
        unlock_markers();
}

static void print_trace_events(FILE *fp)
{
        struct trace_event_iter iter;

        lock_trace_events();
        trace_event_iter_reset(&iter);
        trace_event_iter_start(&iter);

        while (iter.trace_event) {
                fprintf(fp, "trace_event: %s\n", (*iter.trace_event)->name);
                trace_event_iter_next(&iter);
        }
        unlock_trace_events();
}

static int connect_ustconsumer(void)
{
        int result, fd;
        char default_daemon_path[] = SOCK_DIR "/ustconsumer";
        char *explicit_daemon_path, *daemon_path;

        explicit_daemon_path = getenv("UST_DAEMON_SOCKET");
        if (explicit_daemon_path) {
                daemon_path = explicit_daemon_path;
        } else {
                daemon_path = default_daemon_path;
        }

        DBG("Connecting to daemon_path %s", daemon_path);

        result = ustcomm_connect_path(daemon_path, &fd);
        if (result < 0) {
                WARN("connect_ustconsumer failed, daemon_path: %s",
                     daemon_path);
                return result;
        }

        return fd;
}


static void request_buffer_consumer(int sock,
                                    const char *trace,
                                    const char *channel,
                                    int cpu)
{
        struct ustcomm_header send_header, recv_header;
        struct ustcomm_buffer_info buf_inf;
        int result = 0;

        result = ustcomm_pack_buffer_info(&send_header,
                                          &buf_inf,
                                          trace,
                                          channel,
                                          cpu);

        if (result < 0) {
                ERR("failed to pack buffer info message %s_%d",
                    channel, cpu);
                return;
        }

        buf_inf.pid = getpid();
        send_header.command = CONSUME_BUFFER;

        result = ustcomm_req(sock, &send_header, (char *) &buf_inf,
                             &recv_header, NULL);
        if (result <= 0) {
                PERROR("request for buffer consumer failed, is the daemon online?");
        }

        return;
}

/* Ask the daemon to collect a trace called trace_name and being
 * produced by this pid.
 *
 * The trace must be at least allocated. (It can also be started.)
 * This is because _ltt_trace_find is used.
 */

static void inform_consumer_daemon(const char *trace_name)
{
        int sock, i,j;
        struct ust_trace *trace;
        const char *ch_name;

        sock = connect_ustconsumer();
        if (sock < 0) {
                return;
        }

        DBG("Connected to ustconsumer");

        ltt_lock_traces();

        trace = _ltt_trace_find(trace_name);
        if (trace == NULL) {
                WARN("inform_consumer_daemon: could not find trace \"%s\"; it is probably already destroyed", trace_name);
                goto unlock_traces;
        }

        for (i=0; i < trace->nr_channels; i++) {
                if (trace->channels[i].request_collection) {
                        /* iterate on all cpus */
                        for (j=0; j<trace->channels[i].n_cpus; j++) {
                                ch_name = trace->channels[i].channel_name;
                                request_buffer_consumer(sock, trace_name,
                                                        ch_name, j);
                                CMM_STORE_SHARED(buffers_to_export,
                                             CMM_LOAD_SHARED(buffers_to_export)+1);
                        }
                }
        }

unlock_traces:
        ltt_unlock_traces();

        close(sock);
}

static struct ust_channel *find_channel(const char *ch_name,
                                        struct ust_trace *trace)
{
        int i;

        for (i=0; i<trace->nr_channels; i++) {
                if (!strcmp(trace->channels[i].channel_name, ch_name)) {
                        return &trace->channels[i];
                }
        }

        return NULL;
}

static int get_buffer_shmid_pipe_fd(const char *trace_name, const char *ch_name,
                                    int ch_cpu,
                                    int *buf_shmid,
                                    int *buf_struct_shmid,
                                    int *buf_pipe_fd)
{
        struct ust_trace *trace;
        struct ust_channel *channel;
        struct ust_buffer *buf;

        DBG("get_buffer_shmid_pipe_fd");

        ltt_lock_traces();
        trace = _ltt_trace_find(trace_name);
        ltt_unlock_traces();

        if (trace == NULL) {
                ERR("cannot find trace!");
                return -ENODATA;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                ERR("cannot find channel %s!", ch_name);
                return -ENODATA;
        }

        buf = channel->buf[ch_cpu];

        *buf_shmid = buf->shmid;
        *buf_struct_shmid = channel->buf_struct_shmids[ch_cpu];
        *buf_pipe_fd = buf->data_ready_fd_read;

        return 0;
}

static int get_subbuf_num_size(const char *trace_name, const char *ch_name,
                               int *num, int *size)
{
        struct ust_trace *trace;
        struct ust_channel *channel;

        DBG("get_subbuf_size");

        ltt_lock_traces();
        trace = _ltt_trace_find(trace_name);
        ltt_unlock_traces();

        if (!trace) {
                ERR("cannot find trace!");
                return -ENODATA;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                ERR("unable to find channel");
                return -ENODATA;
        }

        *num = channel->subbuf_cnt;
        *size = channel->subbuf_size;

        return 0;
}

/* Return the power of two which is equal or higher to v */

static unsigned int pow2_higher_or_eq(unsigned int v)
{
        int hb = fls(v);
        int retval = 1<<(hb-1);

        if (v-retval == 0)
                return retval;
        else
                return retval<<1;
}

static int set_subbuf_size(const char *trace_name, const char *ch_name,
                           unsigned int size)
{
        unsigned int power;
        int retval = 0;
        struct ust_trace *trace;
        struct ust_channel *channel;

        DBG("set_subbuf_size");

        power = pow2_higher_or_eq(size);
        power = max_t(unsigned int, 2u, power);
        if (power != size) {
                WARN("using the next power of two for buffer size = %u\n", power);
        }

        ltt_lock_traces();
        trace = _ltt_trace_find_setup(trace_name);
        if (trace == NULL) {
                ERR("cannot find trace!");
                retval = -ENODATA;
                goto unlock_traces;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                ERR("unable to find channel");
                retval = -ENODATA;
                goto unlock_traces;
        }

        channel->subbuf_size = power;
        DBG("the set_subbuf_size for the requested channel is %zu", channel->subbuf_size);

unlock_traces:
        ltt_unlock_traces();

        return retval;
}

static int set_subbuf_num(const char *trace_name, const char *ch_name,
                                 unsigned int num)
{
        struct ust_trace *trace;
        struct ust_channel *channel;
        int retval = 0;

        DBG("set_subbuf_num");

        if (num < 2) {
                ERR("subbuffer count should be greater than 2");
                return -EINVAL;
        }

        ltt_lock_traces();
        trace = _ltt_trace_find_setup(trace_name);
        if (trace == NULL) {
                ERR("cannot find trace!");
                retval = -ENODATA;
                goto unlock_traces;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                ERR("unable to find channel");
                retval = -ENODATA;
                goto unlock_traces;
        }

        channel->subbuf_cnt = num;
        DBG("the set_subbuf_cnt for the requested channel is %u", channel->subbuf_cnt);

unlock_traces:
        ltt_unlock_traces();
        return retval;
}

static int get_subbuffer(const char *trace_name, const char *ch_name,
                         int ch_cpu, long *consumed_old)
{
        int retval = 0;
        struct ust_trace *trace;
        struct ust_channel *channel;
        struct ust_buffer *buf;

        DBG("get_subbuf");

        *consumed_old = 0;

        ltt_lock_traces();
        trace = _ltt_trace_find(trace_name);

        if (!trace) {
                DBG("Cannot find trace. It was likely destroyed by the user.");
                retval = -ENODATA;
                goto unlock_traces;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                ERR("unable to find channel");
                retval = -ENODATA;
                goto unlock_traces;
        }

        buf = channel->buf[ch_cpu];

        retval = ust_buffers_get_subbuf(buf, consumed_old);
        if (retval < 0) {
                WARN("missed buffer?");
        }

unlock_traces:
        ltt_unlock_traces();

        return retval;
}


static int notify_buffer_mapped(const char *trace_name,
                                const char *ch_name,
                                int ch_cpu)
{
        int retval = 0;
        struct ust_trace *trace;
        struct ust_channel *channel;
        struct ust_buffer *buf;

        DBG("get_buffer_fd");

        ltt_lock_traces();
        trace = _ltt_trace_find(trace_name);

        if (!trace) {
                retval = -ENODATA;
                DBG("Cannot find trace. It was likely destroyed by the user.");
                goto unlock_traces;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                retval = -ENODATA;
                ERR("unable to find channel");
                goto unlock_traces;
        }

        buf = channel->buf[ch_cpu];

        /* Being here is the proof the daemon has mapped the buffer in its
         * memory. We may now decrement buffers_to_export.
         */
        if (uatomic_read(&buf->consumed) == 0) {
                DBG("decrementing buffers_to_export");
                CMM_STORE_SHARED(buffers_to_export, CMM_LOAD_SHARED(buffers_to_export)-1);
        }

        /* The buffer has been exported, ergo, we can add it to the
         * list of open buffers
         */
        cds_list_add(&buf->open_buffers_list, &open_buffers_list);

unlock_traces:
        ltt_unlock_traces();

        return retval;
}

static int put_subbuffer(const char *trace_name, const char *ch_name,
                         int ch_cpu, long consumed_old)
{
        int retval = 0;
        struct ust_trace *trace;
        struct ust_channel *channel;
        struct ust_buffer *buf;

        DBG("put_subbuf");

        ltt_lock_traces();
        trace = _ltt_trace_find(trace_name);

        if (!trace) {
                retval = -ENODATA;
                DBG("Cannot find trace. It was likely destroyed by the user.");
                goto unlock_traces;
        }

        channel = find_channel(ch_name, trace);
        if (!channel) {
                retval = -ENODATA;
                ERR("unable to find channel");
                goto unlock_traces;
        }

        buf = channel->buf[ch_cpu];

        retval = ust_buffers_put_subbuf(buf, consumed_old);
        if (retval < 0) {
                WARN("ust_buffers_put_subbuf: error (subbuf=%s_%d)",
                     ch_name, ch_cpu);
        } else {
                DBG("ust_buffers_put_subbuf: success (subbuf=%s_%d)",
                    ch_name, ch_cpu);
        }

unlock_traces:
        ltt_unlock_traces();

        return retval;
}

static void listener_cleanup(void *ptr)
{
        pthread_mutex_lock(&listen_sock_mutex);
        if (listen_sock) {
                ustcomm_del_named_sock(listen_sock, 0);
                listen_sock = NULL;
        }
        pthread_mutex_unlock(&listen_sock_mutex);
}

static void force_subbuf_switch()
{
        struct ust_buffer *buf;

        cds_list_for_each_entry(buf, &open_buffers_list,
                            open_buffers_list) {
                ltt_force_switch(buf, FORCE_FLUSH);
        }
}

/* Simple commands are those which need only respond with a return value. */
static int process_simple_client_cmd(int command, char *recv_buf)
{
        int result;

        switch(command) {
        case SET_SOCK_PATH:
        {
                struct ustcomm_single_field *sock_msg;
                sock_msg = (struct ustcomm_single_field *)recv_buf;
                result = ustcomm_unpack_single_field(sock_msg);
                if (result < 0) {
                        return result;
                }
                return setenv("UST_DAEMON_SOCKET", sock_msg->field, 1);
        }

        case FORCE_SUBBUF_SWITCH:
                /* FIXME: return codes? */
                force_subbuf_switch();

                break;

        default:
                return -EINVAL;
        }

        return 0;
}


static int process_trace_cmd(int command, char *trace_name)
{
        int result;
        char trace_type[] = "ustrelay";

        switch(command) {
        case START:
                /* start is an operation that setups the trace, allocates it and starts it */
                result = ltt_trace_setup(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_setup failed");
                        return result;
                }

                result = ltt_trace_set_type(trace_name, trace_type);
                if (result < 0) {
                        ERR("ltt_trace_set_type failed");
                        return result;
                }

                result = ltt_trace_alloc(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_alloc failed");
                        return result;
                }

                inform_consumer_daemon(trace_name);

                result = ltt_trace_start(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_start failed");
                        return result;
                }

                return 0;
        case SETUP_TRACE:
                DBG("trace setup");

                result = ltt_trace_setup(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_setup failed");
                        return result;
                }

                result = ltt_trace_set_type(trace_name, trace_type);
                if (result < 0) {
                        ERR("ltt_trace_set_type failed");
                        return result;
                }

                return 0;
        case ALLOC_TRACE:
                DBG("trace alloc");

                result = ltt_trace_alloc(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_alloc failed");
                        return result;
                }
                inform_consumer_daemon(trace_name);

                return 0;

        case CREATE_TRACE:
                DBG("trace create");

                result = ltt_trace_setup(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_setup failed");
                        return result;
                }

                result = ltt_trace_set_type(trace_name, trace_type);
                if (result < 0) {
                        ERR("ltt_trace_set_type failed");
                        return result;
                }

                return 0;
        case START_TRACE:
                DBG("trace start");

                result = ltt_trace_alloc(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_alloc failed");
                        return result;
                }
                if (!result) {
                        inform_consumer_daemon(trace_name);
                }

                result = ltt_trace_start(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_start failed");
                        return result;
                }

                return 0;
        case STOP_TRACE:
                DBG("trace stop");

                result = ltt_trace_stop(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_stop failed");
                        return result;
                }

                return 0;
        case DESTROY_TRACE:
                DBG("trace destroy");

                result = ltt_trace_destroy(trace_name, 0);
                if (result < 0) {
                        ERR("ltt_trace_destroy failed");
                        return result;
                }
                return 0;
        }

        return 0;
}


static void process_channel_cmd(int sock, int command,
                                struct ustcomm_channel_info *ch_inf)
{
        struct ustcomm_header _reply_header;
        struct ustcomm_header *reply_header = &_reply_header;
        struct ustcomm_channel_info *reply_msg =
                (struct ustcomm_channel_info *)send_buffer;
        int result, offset = 0, num, size;

        memset(reply_header, 0, sizeof(*reply_header));

        switch (command) {
        case GET_SUBBUF_NUM_SIZE:
                result = get_subbuf_num_size(ch_inf->trace,
                                             ch_inf->channel,
                                             &num, &size);
                if (result < 0) {
                        reply_header->result = result;
                        break;
                }

                reply_msg->channel = USTCOMM_POISON_PTR;
                reply_msg->subbuf_num = num;
                reply_msg->subbuf_size = size;


                reply_header->size = COMPUTE_MSG_SIZE(reply_msg, offset);

                break;
        case SET_SUBBUF_NUM:
                reply_header->result = set_subbuf_num(ch_inf->trace,
                                                      ch_inf->channel,
                                                      ch_inf->subbuf_num);

                break;
        case SET_SUBBUF_SIZE:
                reply_header->result = set_subbuf_size(ch_inf->trace,
                                                       ch_inf->channel,
                                                       ch_inf->subbuf_size);


                break;
        }
        if (ustcomm_send(sock, reply_header, (char *)reply_msg) < 0) {
                ERR("ustcomm_send failed");
        }
}

static void process_buffer_cmd(int sock, int command,
                               struct ustcomm_buffer_info *buf_inf)
{
        struct ustcomm_header _reply_header;
        struct ustcomm_header *reply_header = &_reply_header;
        struct ustcomm_buffer_info *reply_msg =
                (struct ustcomm_buffer_info *)send_buffer;
        int result, offset = 0, buf_shmid, buf_struct_shmid, buf_pipe_fd;
        long consumed_old;

        memset(reply_header, 0, sizeof(*reply_header));

        switch (command) {
        case GET_BUF_SHMID_PIPE_FD:
                result = get_buffer_shmid_pipe_fd(buf_inf->trace,
                                                  buf_inf->channel,
                                                  buf_inf->ch_cpu,
                                                  &buf_shmid,
                                                  &buf_struct_shmid,
                                                  &buf_pipe_fd);
                if (result < 0) {
                        reply_header->result = result;
                        break;
                }

                reply_msg->channel = USTCOMM_POISON_PTR;
                reply_msg->buf_shmid = buf_shmid;
                reply_msg->buf_struct_shmid = buf_struct_shmid;

                reply_header->size = COMPUTE_MSG_SIZE(reply_msg, offset);
                reply_header->fd_included = 1;

                if (ustcomm_send_fd(sock, reply_header, (char *)reply_msg,
                                    &buf_pipe_fd) < 0) {
                        ERR("ustcomm_send failed");
                }
                return;

        case NOTIFY_BUF_MAPPED:
                reply_header->result =
                        notify_buffer_mapped(buf_inf->trace,
                                             buf_inf->channel,
                                             buf_inf->ch_cpu);
                break;
        case GET_SUBBUFFER:
                result = get_subbuffer(buf_inf->trace, buf_inf->channel,
                                       buf_inf->ch_cpu, &consumed_old);
                if (result < 0) {
                        reply_header->result = result;
                        break;
                }

                reply_msg->channel = USTCOMM_POISON_PTR;
                reply_msg->consumed_old = consumed_old;

                reply_header->size = COMPUTE_MSG_SIZE(reply_msg, offset);

                break;
        case PUT_SUBBUFFER:
                result = put_subbuffer(buf_inf->trace, buf_inf->channel,
                                       buf_inf->ch_cpu,
                                       buf_inf->consumed_old);
                reply_header->result = result;

                break;
        }

        if (ustcomm_send(sock, reply_header, (char *)reply_msg) < 0) {
                ERR("ustcomm_send failed");
        }

}

static void process_marker_cmd(int sock, int command,
                               struct ustcomm_marker_info *marker_inf)
{
        struct ustcomm_header _reply_header;
        struct ustcomm_header *reply_header = &_reply_header;
        int result = 0;

        memset(reply_header, 0, sizeof(*reply_header));

        switch(command) {
        case ENABLE_MARKER:

                result = ltt_marker_connect(marker_inf->channel,
                                            marker_inf->marker,
                                            "default");
                if (result < 0) {
                        WARN("could not enable marker; channel=%s,"
                             " name=%s",
                             marker_inf->channel,
                             marker_inf->marker);

                }
                break;
        case DISABLE_MARKER:
                result = ltt_marker_disconnect(marker_inf->channel,
                                               marker_inf->marker,
                                               "default");
                if (result < 0) {
                        WARN("could not disable marker; channel=%s,"
                             " name=%s",
                             marker_inf->channel,
                             marker_inf->marker);
                }
                break;
        }

        reply_header->result = result;

        if (ustcomm_send(sock, reply_header, NULL) < 0) {
                ERR("ustcomm_send failed");
        }

}
static void process_client_cmd(struct ustcomm_header *recv_header,
                               char *recv_buf, int sock)
{
        int result;
        struct ustcomm_header _reply_header;
        struct ustcomm_header *reply_header = &_reply_header;
        char *send_buf = send_buffer;

        memset(reply_header, 0, sizeof(*reply_header));
        memset(send_buf, 0, sizeof(send_buffer));

        switch(recv_header->command) {
        case GET_SUBBUF_NUM_SIZE:
        case SET_SUBBUF_NUM:
        case SET_SUBBUF_SIZE:
        {
                struct ustcomm_channel_info *ch_inf;
                ch_inf = (struct ustcomm_channel_info *)recv_buf;
                result = ustcomm_unpack_channel_info(ch_inf);
                if (result < 0) {
                        ERR("couldn't unpack channel info");
                        reply_header->result = -EINVAL;
                        goto send_response;
                }
                process_channel_cmd(sock, recv_header->command, ch_inf);
                return;
        }
        case GET_BUF_SHMID_PIPE_FD:
        case NOTIFY_BUF_MAPPED:
        case GET_SUBBUFFER:
        case PUT_SUBBUFFER:
        {
                struct ustcomm_buffer_info *buf_inf;
                buf_inf = (struct ustcomm_buffer_info *)recv_buf;
                result = ustcomm_unpack_buffer_info(buf_inf);
                if (result < 0) {
                        ERR("couldn't unpack buffer info");
                        reply_header->result = -EINVAL;
                        goto send_response;
                }
                process_buffer_cmd(sock, recv_header->command, buf_inf);
                return;
        }
        case ENABLE_MARKER:
        case DISABLE_MARKER:
        {
                struct ustcomm_marker_info *marker_inf;
                marker_inf = (struct ustcomm_marker_info *)recv_buf;
                result = ustcomm_unpack_marker_info(marker_inf);
                if (result < 0) {
                        ERR("couldn't unpack marker info");
                        reply_header->result = -EINVAL;
                        goto send_response;
                }
                process_marker_cmd(sock, recv_header->command, marker_inf);
                return;
        }
        case LIST_MARKERS:
        {
                char *ptr;
                size_t size;
                FILE *fp;

                fp = open_memstream(&ptr, &size);
                if (fp == NULL) {
                        ERR("opening memstream failed");
                        return;
                }
                print_markers(fp);
                fclose(fp);

                reply_header->size = size;

                result = ustcomm_send(sock, reply_header, ptr);

                free(ptr);

                if (result < 0) {
                        PERROR("failed to send markers list");
                }

                break;
        }
        case LIST_TRACE_EVENTS:
        {
                char *ptr;
                size_t size;
                FILE *fp;

                fp = open_memstream(&ptr, &size);
                if (fp == NULL) {
                        ERR("opening memstream failed");
                        return;
                }
                print_trace_events(fp);
                fclose(fp);

                reply_header->size = size;

                result = ustcomm_send(sock, reply_header, ptr);

                free(ptr);

                if (result < 0) {
                        ERR("list_trace_events failed");
                        return;
                }

                break;
        }
        case LOAD_PROBE_LIB:
        {
                char *libfile;

                /* FIXME: No functionality at all... */
                libfile = recv_buf;

                DBG("load_probe_lib loading %s", libfile);

                break;
        }
        case GET_PIDUNIQUE:
        {
                struct ustcomm_pidunique *pid_msg;
                pid_msg = (struct ustcomm_pidunique *)send_buf;

                pid_msg->pidunique = pidunique;
                reply_header->size = sizeof(pid_msg);

                goto send_response;

        }
        case GET_SOCK_PATH:
        {
                struct ustcomm_single_field *sock_msg;
                char *sock_path_env;

                sock_msg = (struct ustcomm_single_field *)send_buf;

                sock_path_env = getenv("UST_DAEMON_SOCKET");

                if (!sock_path_env) {
                        result = ustcomm_pack_single_field(reply_header,
                                                           sock_msg,
                                                           SOCK_DIR "/ustconsumer");

                } else {
                        result = ustcomm_pack_single_field(reply_header,
                                                           sock_msg,
                                                           sock_path_env);
                }
                reply_header->result = result;

                goto send_response;
        }
        case START:
        case SETUP_TRACE:
        case ALLOC_TRACE:
        case CREATE_TRACE:
        case START_TRACE:
        case STOP_TRACE:
        case DESTROY_TRACE:
        {
                struct ustcomm_single_field *trace_inf =
                        (struct ustcomm_single_field *)recv_buf;

                result = ustcomm_unpack_single_field(trace_inf);
                if (result < 0) {
                        ERR("couldn't unpack trace info");
                        reply_header->result = -EINVAL;
                        goto send_response;
                }

                reply_header->result =
                        process_trace_cmd(recv_header->command,
                                          trace_inf->field);
                goto send_response;

        }
        default:
                reply_header->result =
                        process_simple_client_cmd(recv_header->command,
                                                  recv_buf);
                goto send_response;

        }

        return;

send_response:
        ustcomm_send(sock, reply_header, send_buf);
}

#define MAX_EVENTS 10

void *listener_main(void *p)
{
        struct ustcomm_sock *epoll_sock;
        struct epoll_event events[MAX_EVENTS];
        struct sockaddr addr;
        int accept_fd, nfds, result, i, addr_size;

        DBG("LISTENER");

        pthread_cleanup_push(listener_cleanup, NULL);

        for(;;) {
                nfds = epoll_wait(epoll_fd, events, MAX_EVENTS, -1);
                if (nfds == -1) {
                        PERROR("listener_main: epoll_wait failed");
                        continue;
                }

                for (i = 0; i < nfds; i++) {
                        pthread_mutex_lock(&listener_thread_data_mutex);
                        epoll_sock = (struct ustcomm_sock *)events[i].data.ptr;
                        if (epoll_sock == listen_sock) {
                                addr_size = sizeof(struct sockaddr);
                                accept_fd = accept(epoll_sock->fd,
                                                   &addr,
                                                   (socklen_t *)&addr_size);
                                if (accept_fd == -1) {
                                        PERROR("listener_main: accept failed");
                                        continue;
                                }
                                ustcomm_init_sock(accept_fd, epoll_fd,
                                                 &ust_socks);
                        } else {
                                memset(receive_header, 0,
                                       sizeof(*receive_header));
                                memset(receive_buffer, 0,
                                       sizeof(receive_buffer));
                                result = ustcomm_recv(epoll_sock->fd,
                                                      receive_header,
                                                      receive_buffer);
                                if (result == 0) {
                                        ustcomm_del_sock(epoll_sock, 0);
                                } else {
                                        process_client_cmd(receive_header,
                                                           receive_buffer,
                                                           epoll_sock->fd);
                                }
                        }
                        pthread_mutex_unlock(&listener_thread_data_mutex);
                }
        }

        pthread_cleanup_pop(1);
}

/* These should only be accessed in the parent thread,
 * not the listener.
 */
static volatile sig_atomic_t have_listener = 0;
static pthread_t listener_thread;

void create_listener(void)
{
        int result;
        sigset_t sig_all_blocked;
        sigset_t orig_parent_mask;

        if (have_listener) {
                WARN("not creating listener because we already had one");
                return;
        }

        /* A new thread created by pthread_create inherits the signal mask
         * from the parent. To avoid any signal being received by the
         * listener thread, we block all signals temporarily in the parent,
         * while we create the listener thread.
         */

        sigfillset(&sig_all_blocked);

        result = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
        if (result) {
                PERROR("pthread_sigmask: %s", strerror(result));
        }

        result = pthread_create(&listener_thread, NULL, listener_main, NULL);
        if (result == -1) {
                PERROR("pthread_create");
        }

        /* Restore original signal mask in parent */
        result = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
        if (result) {
                PERROR("pthread_sigmask: %s", strerror(result));
        } else {
                have_listener = 1;
        }
}

#define AUTOPROBE_DISABLED      0
#define AUTOPROBE_ENABLE_ALL    1
#define AUTOPROBE_ENABLE_REGEX  2
static int autoprobe_method = AUTOPROBE_DISABLED;
static regex_t autoprobe_regex;

static void auto_probe_connect(struct marker *m)
{
        int result;

        char* concat_name = NULL;
        const char *probe_name = "default";

        if (autoprobe_method == AUTOPROBE_DISABLED) {
                return;
        } else if (autoprobe_method == AUTOPROBE_ENABLE_REGEX) {
                result = asprintf(&concat_name, "%s/%s", m->channel, m->name);
                if (result == -1) {
                        ERR("auto_probe_connect: asprintf failed (marker %s/%s)",
                                m->channel, m->name);
                        return;
                }
                if (regexec(&autoprobe_regex, concat_name, 0, NULL, 0)) {
                        free(concat_name);
                        return;
                }
                free(concat_name);
        }

        result = ltt_marker_connect(m->channel, m->name, probe_name);
        if (result && result != -EEXIST)
                ERR("ltt_marker_connect (marker = %s/%s, errno = %d)", m->channel, m->name, -result);

        DBG("auto connected marker %s (addr: %p) %s to probe default", m->channel, m, m->name);

}

static struct ustcomm_sock * init_app_socket(int epoll_fd)
{
        char *name;
        int result;
        struct ustcomm_sock *sock;

        result = asprintf(&name, "%s/%d", SOCK_DIR, (int)getpid());
        if (result < 0) {
                ERR("string overflow allocating socket name, "
                    "UST thread bailing");
                return NULL;
        }

        result = ensure_dir_exists(SOCK_DIR);
        if (result == -1) {
                ERR("Unable to create socket directory %s, UST thread bailing",
                    SOCK_DIR);
                goto free_name;
        }

        sock = ustcomm_init_named_socket(name, epoll_fd);
        if (!sock) {
                ERR("Error initializing named socket (%s). Check that directory"
                    "exists and that it is writable. UST thread bailing", name);
                goto free_name;
        }

        free(name);
        return sock;

free_name:
        free(name);
        return NULL;
}

static void __attribute__((constructor)) init()
{
        struct timespec ts;
        int result;
        char* autoprobe_val = NULL;
        char* subbuffer_size_val = NULL;
        char* subbuffer_count_val = NULL;
        unsigned int subbuffer_size;
        unsigned int subbuffer_count;
        unsigned int power;

        /* Assign the pidunique, to be able to differentiate the processes with same
         * pid, (before and after an exec).
         */
        pidunique = make_pidunique();
        processpid = getpid();

        DBG("Tracectl constructor");

        /* Set up epoll */
        epoll_fd = epoll_create(MAX_EVENTS);
        if (epoll_fd == -1) {
                ERR("epoll_create failed, tracing shutting down");
                return;
        }

        /* Create the socket */
        listen_sock = init_app_socket(epoll_fd);
        if (!listen_sock) {
                ERR("failed to create application socket,"
                    " tracing shutting down");
                return;
        }

        create_listener();

        /* Get clock the clock source type */

        /* Default clock source */
        ust_clock_source = CLOCK_TRACE;
        if (clock_gettime(ust_clock_source, &ts) != 0) {
                ust_clock_source = CLOCK_MONOTONIC;
                DBG("UST traces will not be synchronized with LTTng traces");
        }

        autoprobe_val = getenv("UST_AUTOPROBE");
        if (autoprobe_val) {
                struct marker_iter iter;

                DBG("Autoprobe enabled.");

                /* Ensure markers are initialized */
                //init_markers();

                /* Ensure marker control is initialized, for the probe */
                init_marker_control();

                /* first, set the callback that will connect the
                 * probe on new markers
                 */
                if (autoprobe_val[0] == '/') {
                        result = regcomp(&autoprobe_regex, autoprobe_val+1, 0);
                        if (result) {
                                char regexerr[150];

                                regerror(result, &autoprobe_regex, regexerr, sizeof(regexerr));
                                ERR("cannot parse regex %s (%s), will ignore UST_AUTOPROBE", autoprobe_val, regexerr);
                                /* don't crash the application just for this */
                        } else {
                                autoprobe_method = AUTOPROBE_ENABLE_REGEX;
                        }
                } else {
                        /* just enable all instrumentation */
                        autoprobe_method = AUTOPROBE_ENABLE_ALL;
                }

                marker_set_new_marker_cb(auto_probe_connect);

                /* Now, connect the probes that were already registered. */
                marker_iter_reset(&iter);
                marker_iter_start(&iter);

                DBG("now iterating on markers already registered");
                while (iter.marker) {
                        DBG("now iterating on marker %s", (*iter.marker)->name);
                        auto_probe_connect(*iter.marker);
                        marker_iter_next(&iter);
                }
        }

        if (getenv("UST_OVERWRITE")) {
                int val = atoi(getenv("UST_OVERWRITE"));
                if (val == 0 || val == 1) {
                        CMM_STORE_SHARED(ust_channels_overwrite_by_default, val);
                } else {
                        WARN("invalid value for UST_OVERWRITE");
                }
        }

        if (getenv("UST_AUTOCOLLECT")) {
                int val = atoi(getenv("UST_AUTOCOLLECT"));
                if (val == 0 || val == 1) {
                        CMM_STORE_SHARED(ust_channels_request_collection_by_default, val);
                } else {
                        WARN("invalid value for UST_AUTOCOLLECT");
                }
        }

        subbuffer_size_val = getenv("UST_SUBBUF_SIZE");
        if (subbuffer_size_val) {
                sscanf(subbuffer_size_val, "%u", &subbuffer_size);
                power = pow2_higher_or_eq(subbuffer_size);
                if (power != subbuffer_size)
                        WARN("using the next power of two for buffer size = %u\n", power);
                chan_infos[LTT_CHANNEL_UST].def_subbufsize = power;
        }

        subbuffer_count_val = getenv("UST_SUBBUF_NUM");
        if (subbuffer_count_val) {
                sscanf(subbuffer_count_val, "%u", &subbuffer_count);
                if (subbuffer_count < 2)
                        subbuffer_count = 2;
                chan_infos[LTT_CHANNEL_UST].def_subbufcount = subbuffer_count;
        }

        if (getenv("UST_TRACE")) {
                char trace_name[] = "auto";
                char trace_type[] = "ustrelay";

                DBG("starting early tracing");

                /* Ensure marker control is initialized */
                init_marker_control();

                /* Ensure markers are initialized */
                init_markers();

                /* Ensure buffers are initialized, for the transport to be available.
                 * We are about to set a trace type and it will fail without this.
                 */
                init_ustrelay_transport();

                /* FIXME: When starting early tracing (here), depending on the
                 * order of constructors, it is very well possible some marker
                 * sections are not yet registered. Because of this, some
                 * channels may not be registered. Yet, we are about to ask the
                 * daemon to collect the channels. Channels which are not yet
                 * registered will not be collected.
                 *
                 * Currently, in LTTng, there is no way to add a channel after
                 * trace start. The reason for this is that it induces complex
                 * concurrency issues on the trace structures, which can only
                 * be resolved using RCU. This has not been done yet. As a
                 * workaround, we are forcing the registration of the "ust"
                 * channel here. This is the only channel (apart from metadata)
                 * that can be reliably used in early tracing.
                 *
                 * Non-early tracing does not have this problem and can use
                 * arbitrary channel names.
                 */
                ltt_channels_register("ust");

                result = ltt_trace_setup(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_setup failed");
                        return;
                }

                result = ltt_trace_set_type(trace_name, trace_type);
                if (result < 0) {
                        ERR("ltt_trace_set_type failed");
                        return;
                }

                result = ltt_trace_alloc(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_alloc failed");
                        return;
                }

                result = ltt_trace_start(trace_name);
                if (result < 0) {
                        ERR("ltt_trace_start failed");
                        return;
                }

                /* Do this after the trace is started in order to avoid creating confusion
                 * if the trace fails to start. */
                inform_consumer_daemon(trace_name);
        }

        return;

        /* should decrementally destroy stuff if error */

}

/* This is only called if we terminate normally, not with an unhandled signal,
 * so we cannot rely on it. However, for now, LTTV requires that the header of
 * the last sub-buffer contain a valid end time for the trace. This is done
 * automatically only when the trace is properly stopped.
 *
 * If the traced program crashed, it is always possible to manually add the
 * right value in the header, or to open the trace in text mode.
 *
 * FIXME: Fix LTTV so it doesn't need this.
 */

static void destroy_traces(void)
{
        int result;

        /* if trace running, finish it */

        DBG("destructor stopping traces");

        result = ltt_trace_stop("auto");
        if (result == -1) {
                ERR("ltt_trace_stop error");
        }

        result = ltt_trace_destroy("auto", 0);
        if (result == -1) {
                ERR("ltt_trace_destroy error");
        }
}

static int trace_recording(void)
{
        int retval = 0;
        struct ust_trace *trace;

        ltt_lock_traces();

        cds_list_for_each_entry(trace, &ltt_traces.head, list) {
                if (trace->active) {
                        retval = 1;
                        break;
                }
        }

        ltt_unlock_traces();

        return retval;
}

int restarting_usleep(useconds_t usecs)
{
        struct timespec tv;
        int result;

        tv.tv_sec = 0;
        tv.tv_nsec = usecs * 1000;

        do {
                result = nanosleep(&tv, &tv);
        } while (result == -1 && errno == EINTR);

        return result;
}

static void stop_listener(void)
{
        int result;

        if (!have_listener)
                return;

        result = pthread_cancel(listener_thread);
        if (result != 0) {
                ERR("pthread_cancel: %s", strerror(result));
        }
        result = pthread_join(listener_thread, NULL);
        if (result != 0) {
                ERR("pthread_join: %s", strerror(result));
        }
}

/* This destructor keeps the process alive for a few seconds in order
 * to leave time for ustconsumer to connect to its buffers. This is necessary
 * for programs whose execution is very short. It is also useful in all
 * programs when tracing is started close to the end of the program
 * execution.
 *
 * FIXME: For now, this only works for the first trace created in a
 * process.
 */

static void __attribute__((destructor)) keepalive()
{
        if (processpid != getpid()) {
                return;
        }

        if (trace_recording() && CMM_LOAD_SHARED(buffers_to_export)) {
                int total = 0;
                DBG("Keeping process alive for consumer daemon...");
                while (CMM_LOAD_SHARED(buffers_to_export)) {
                        const int interv = 200000;
                        restarting_usleep(interv);
                        total += interv;

                        if (total >= 3000000) {
                                WARN("non-consumed buffers remaining after wait limit; not waiting anymore");
                                break;
                        }
                }
                DBG("Finally dying...");
        }

        destroy_traces();

        /* Ask the listener to stop and clean up. */
        stop_listener();
}

void ust_potential_exec(void)
{
        trace_mark(ust, potential_exec, MARK_NOARGS);

        DBG("test");

        keepalive();
}

/* Notify ust that there was a fork. This needs to be called inside
 * the new process, anytime a process whose memory is not shared with
 * the parent is created. If this function is not called, the events
 * of the new process will not be collected.
 *
 * Signals should be disabled before the fork and reenabled only after
 * this call in order to guarantee tracing is not started before ust_fork()
 * sanitizes the new process.
 */

static void ust_fork(void)
{
        struct ust_buffer *buf, *buf_tmp;
        struct ustcomm_sock *sock, *sock_tmp;
        struct ust_trace *trace, *trace_tmp;
        int result;

        /* FIXME: technically, the locks could have been taken before the fork */
        DBG("ust: forking");

        /* Get the pid of the new process */
        processpid = getpid();

        /*
         * FIXME: This could be prettier, we loop over the list twice and
         * following good locking practice should lock around the loop
         */
        cds_list_for_each_entry_safe(trace, trace_tmp, &ltt_traces.head, list) {
                ltt_trace_stop(trace->trace_name);
        }

        /* Delete all active connections, but leave them in the epoll set */
        cds_list_for_each_entry_safe(sock, sock_tmp, &ust_socks, list) {
                ustcomm_del_sock(sock, 1);
        }

        /* Delete all blocked consumers */
        cds_list_for_each_entry_safe(buf, buf_tmp, &open_buffers_list,
                                 open_buffers_list) {
                cds_list_del(&buf->open_buffers_list);
        }

        /*
         * FIXME: This could be prettier, we loop over the list twice and
         * following good locking practice should lock around the loop
         */
        cds_list_for_each_entry_safe(trace, trace_tmp, &ltt_traces.head, list) {
                ltt_trace_destroy(trace->trace_name, 1);
        }

        /* Clean up the listener socket and epoll, keeping the socket file */
        if (listen_sock) {
                ustcomm_del_named_sock(listen_sock, 1);
                listen_sock = NULL;
        }
        close(epoll_fd);

        /* Re-start the launch sequence */
        CMM_STORE_SHARED(buffers_to_export, 0);
        have_listener = 0;

        /* Set up epoll */
        epoll_fd = epoll_create(MAX_EVENTS);
        if (epoll_fd == -1) {
                ERR("epoll_create failed, tracing shutting down");
                return;
        }

        /* Create the socket */
        listen_sock = init_app_socket(epoll_fd);
        if (!listen_sock) {
                ERR("failed to create application socket,"
                    " tracing shutting down");
                return;
        }
        create_listener();
        ltt_trace_setup("auto");
        result = ltt_trace_set_type("auto", "ustrelay");
        if (result < 0) {
                ERR("ltt_trace_set_type failed");
                return;
        }

        ltt_trace_alloc("auto");
        ltt_trace_start("auto");
        inform_consumer_daemon("auto");
}

void ust_before_fork(ust_fork_info_t *fork_info)
{
        /* Disable signals. This is to avoid that the child
         * intervenes before it is properly setup for tracing. It is
         * safer to disable all signals, because then we know we are not
         * breaking anything by restoring the original mask.
         */
        sigset_t all_sigs;
        int result;

        /* FIXME:
                - only do this if tracing is active
        */

        /* Disable signals */
        sigfillset(&all_sigs);
        result = sigprocmask(SIG_BLOCK, &all_sigs, &fork_info->orig_sigs);
        if (result == -1) {
                PERROR("sigprocmask");
                return;
        }

        /*
         * Take the fork lock to make sure we are not in the middle of
         * something in the listener thread.
         */
        pthread_mutex_lock(&listener_thread_data_mutex);
        /*
         * Hold listen_sock_mutex to protect from listen_sock teardown.
         */
        pthread_mutex_lock(&listen_sock_mutex);
}

/* Don't call this function directly in a traced program */
static void ust_after_fork_common(ust_fork_info_t *fork_info)
{
        int result;

        pthread_mutex_unlock(&listen_sock_mutex);
        pthread_mutex_unlock(&listener_thread_data_mutex);

        /* Restore signals */
        result = sigprocmask(SIG_SETMASK, &fork_info->orig_sigs, NULL);
        if (result == -1) {
                PERROR("sigprocmask");
                return;
        }
}

void ust_after_fork_parent(ust_fork_info_t *fork_info)
{
        /* Reenable signals */
        ust_after_fork_common(fork_info);
}

void ust_after_fork_child(ust_fork_info_t *fork_info)
{
        /* First sanitize the child */
        ust_fork();

        /* Then reenable interrupts */
        ust_after_fork_common(fork_info);
}