Fix: missing RCU read side critical sections

[lttng-tools.git] / src / bin / lttng-sessiond / consumer.cpp

/*
 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
 * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
 *
 * SPDX-License-Identifier: GPL-2.0-only
 *
 */

#define _LGPL_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <inttypes.h>

#include <common/common.h>
#include <common/defaults.h>
#include <common/uri.h>
#include <common/relayd/relayd.h>
#include <common/string-utils/format.h>

#include "consumer.h"
#include "health-sessiond.h"
#include "ust-app.h"
#include "utils.h"
#include "lttng-sessiond.h"

/*
 * Return allocated full pathname of the session using the consumer trace path
 * and subdir if available.
 *
 * The caller can safely free(3) the returned value. On error, NULL is
 * returned.
 */
char *setup_channel_trace_path(struct consumer_output *consumer,
                const char *session_path, size_t *consumer_path_offset)
{
        int ret;
        char *pathname;

        LTTNG_ASSERT(consumer);
        LTTNG_ASSERT(session_path);

        health_code_update();

        /*
         * Allocate the string ourself to make sure we never exceed
         * LTTNG_PATH_MAX.
         */
        pathname = (char *) zmalloc(LTTNG_PATH_MAX);
        if (!pathname) {
                goto error;
        }

        /* Get correct path name destination */
        if (consumer->type == CONSUMER_DST_NET &&
                        consumer->relay_major_version == 2 &&
                        consumer->relay_minor_version < 11) {
                ret = snprintf(pathname, LTTNG_PATH_MAX, "%s%s/%s/%s",
                                consumer->dst.net.base_dir,
                                consumer->chunk_path, consumer->domain_subdir,
                                session_path);
                *consumer_path_offset = 0;
        } else {
                ret = snprintf(pathname, LTTNG_PATH_MAX, "%s/%s",
                                consumer->domain_subdir, session_path);
                *consumer_path_offset = strlen(consumer->domain_subdir) + 1;
        }
        DBG3("Consumer trace path relative to current trace chunk: \"%s\"",
                        pathname);
        if (ret < 0) {
                PERROR("Failed to format channel path");
                goto error;
        } else if (ret >= LTTNG_PATH_MAX) {
                ERR("Truncation occurred while formatting channel path");
                goto error;
        }

        return pathname;
error:
        free(pathname);
        return NULL;
}

/*
 * Send a data payload using a given consumer socket of size len.
 *
 * The consumer socket lock MUST be acquired before calling this since this
 * function can change the fd value.
 *
 * Return 0 on success else a negative value on error.
 */
int consumer_socket_send(
                struct consumer_socket *socket, const void *msg, size_t len)
{
        int fd;
        ssize_t size;

        LTTNG_ASSERT(socket);
        LTTNG_ASSERT(socket->fd_ptr);
        LTTNG_ASSERT(msg);

        /* Consumer socket is invalid. Stopping. */
        fd = *socket->fd_ptr;
        if (fd < 0) {
                goto error;
        }

        size = lttcomm_send_unix_sock(fd, msg, len);
        if (size < 0) {
                /* The above call will print a PERROR on error. */
                DBG("Error when sending data to consumer on sock %d", fd);
                /*
                 * At this point, the socket is not usable anymore thus closing it and
                 * setting the file descriptor to -1 so it is not reused.
                 */

                /* This call will PERROR on error. */
                (void) lttcomm_close_unix_sock(fd);
                *socket->fd_ptr = -1;
                goto error;
        }

        return 0;

error:
        return -1;
}

/*
 * Receive a data payload using a given consumer socket of size len.
 *
 * The consumer socket lock MUST be acquired before calling this since this
 * function can change the fd value.
 *
 * Return 0 on success else a negative value on error.
 */
int consumer_socket_recv(struct consumer_socket *socket, void *msg, size_t len)
{
        int fd;
        ssize_t size;

        LTTNG_ASSERT(socket);
        LTTNG_ASSERT(socket->fd_ptr);
        LTTNG_ASSERT(msg);

        /* Consumer socket is invalid. Stopping. */
        fd = *socket->fd_ptr;
        if (fd < 0) {
                goto error;
        }

        size = lttcomm_recv_unix_sock(fd, msg, len);
        if (size <= 0) {
                /* The above call will print a PERROR on error. */
                DBG("Error when receiving data from the consumer socket %d", fd);
                /*
                 * At this point, the socket is not usable anymore thus closing it and
                 * setting the file descriptor to -1 so it is not reused.
                 */

                /* This call will PERROR on error. */
                (void) lttcomm_close_unix_sock(fd);
                *socket->fd_ptr = -1;
                goto error;
        }

        return 0;

error:
        return -1;
}

/*
 * Receive a reply command status message from the consumer. Consumer socket
 * lock MUST be acquired before calling this function.
 *
 * Return 0 on success, -1 on recv error or a negative lttng error code which
 * was possibly returned by the consumer.
 */
int consumer_recv_status_reply(struct consumer_socket *sock)
{
        int ret;
        struct lttcomm_consumer_status_msg reply;

        LTTNG_ASSERT(sock);

        ret = consumer_socket_recv(sock, &reply, sizeof(reply));
        if (ret < 0) {
                goto end;
        }

        if (reply.ret_code == LTTCOMM_CONSUMERD_SUCCESS) {
                /* All good. */
                ret = 0;
        } else {
                ret = -reply.ret_code;
                DBG("Consumer ret code %d", ret);
        }

end:
        return ret;
}

/*
 * Once the ASK_CHANNEL command is sent to the consumer, the channel
 * information are sent back. This call receives that data and populates key
 * and stream_count.
 *
 * On success return 0 and both key and stream_count are set. On error, a
 * negative value is sent back and both parameters are untouched.
 */
int consumer_recv_status_channel(struct consumer_socket *sock,
                uint64_t *key, unsigned int *stream_count)
{
        int ret;
        struct lttcomm_consumer_status_channel reply;

        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(stream_count);
        LTTNG_ASSERT(key);

        ret = consumer_socket_recv(sock, &reply, sizeof(reply));
        if (ret < 0) {
                goto end;
        }

        /* An error is possible so don't touch the key and stream_count. */
        if (reply.ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
                ret = -1;
                goto end;
        }

        *key = reply.key;
        *stream_count = reply.stream_count;
        ret = 0;

end:
        return ret;
}

/*
 * Send destroy relayd command to consumer.
 *
 * On success return positive value. On error, negative value.
 */
int consumer_send_destroy_relayd(struct consumer_socket *sock,
                struct consumer_output *consumer)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(consumer);
        LTTNG_ASSERT(sock);

        DBG2("Sending destroy relayd command to consumer sock %d", *sock->fd_ptr);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_DESTROY_RELAYD;
        msg.u.destroy_relayd.net_seq_idx = consumer->net_seq_index;

        pthread_mutex_lock(sock->lock);
        ret = consumer_socket_send(sock, &msg, sizeof(msg));
        if (ret < 0) {
                goto error;
        }

        /* Don't check the return value. The caller will do it. */
        ret = consumer_recv_status_reply(sock);

        DBG2("Consumer send destroy relayd command done");

error:
        pthread_mutex_unlock(sock->lock);
        return ret;
}

/*
 * For each consumer socket in the consumer output object, send a destroy
 * relayd command.
 */
void consumer_output_send_destroy_relayd(struct consumer_output *consumer)
{
        struct lttng_ht_iter iter;
        struct consumer_socket *socket;

        LTTNG_ASSERT(consumer);

        /* Destroy any relayd connection */
        if (consumer->type == CONSUMER_DST_NET) {
                rcu_read_lock();
                cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
                                node.node) {
                        int ret;

                        /* Send destroy relayd command */
                        ret = consumer_send_destroy_relayd(socket, consumer);
                        if (ret < 0) {
                                DBG("Unable to send destroy relayd command to consumer");
                                /* Continue since we MUST delete everything at this point. */
                        }
                }
                rcu_read_unlock();
        }
}

/*
 * From a consumer_data structure, allocate and add a consumer socket to the
 * consumer output.
 *
 * Return 0 on success, else negative value on error
 */
int consumer_create_socket(struct consumer_data *data,
                struct consumer_output *output)
{
        int ret = 0;
        struct consumer_socket *socket;

        LTTNG_ASSERT(data);

        if (output == NULL || data->cmd_sock < 0) {
                /*
                 * Not an error. Possible there is simply not spawned consumer or it's
                 * disabled for the tracing session asking the socket.
                 */
                goto error;
        }

        rcu_read_lock();
        socket = consumer_find_socket(data->cmd_sock, output);
        rcu_read_unlock();
        if (socket == NULL) {
                socket = consumer_allocate_socket(&data->cmd_sock);
                if (socket == NULL) {
                        ret = -1;
                        goto error;
                }

                socket->registered = 0;
                socket->lock = &data->lock;
                rcu_read_lock();
                consumer_add_socket(socket, output);
                rcu_read_unlock();
        }

        socket->type = data->type;

        DBG3("Consumer socket created (fd: %d) and added to output",
                        data->cmd_sock);

error:
        return ret;
}

/*
 * Return the consumer socket from the given consumer output with the right
 * bitness. On error, returns NULL.
 *
 * The caller MUST acquire a rcu read side lock and keep it until the socket
 * object reference is not needed anymore.
 */
struct consumer_socket *consumer_find_socket_by_bitness(int bits,
                const struct consumer_output *consumer)
{
        int consumer_fd;
        struct consumer_socket *socket = NULL;

        ASSERT_RCU_READ_LOCKED();

        switch (bits) {
        case 64:
                consumer_fd = uatomic_read(&the_ust_consumerd64_fd);
                break;
        case 32:
                consumer_fd = uatomic_read(&the_ust_consumerd32_fd);
                break;
        default:
                abort();
                goto end;
        }

        socket = consumer_find_socket(consumer_fd, consumer);
        if (!socket) {
                ERR("Consumer socket fd %d not found in consumer obj %p",
                                consumer_fd, consumer);
        }

end:
        return socket;
}

/*
 * Find a consumer_socket in a consumer_output hashtable. Read side lock must
 * be acquired before calling this function and across use of the
 * returned consumer_socket.
 */
struct consumer_socket *consumer_find_socket(int key,
                const struct consumer_output *consumer)
{
        struct lttng_ht_iter iter;
        struct lttng_ht_node_ulong *node;
        struct consumer_socket *socket = NULL;

        ASSERT_RCU_READ_LOCKED();

        /* Negative keys are lookup failures */
        if (key < 0 || consumer == NULL) {
                return NULL;
        }

        lttng_ht_lookup(consumer->socks, (void *)((unsigned long) key),
                        &iter);
        node = lttng_ht_iter_get_node_ulong(&iter);
        if (node != NULL) {
                socket = caa_container_of(node, struct consumer_socket, node);
        }

        return socket;
}

/*
 * Allocate a new consumer_socket and return the pointer.
 */
struct consumer_socket *consumer_allocate_socket(int *fd)
{
        struct consumer_socket *socket = NULL;

        LTTNG_ASSERT(fd);

        socket = (consumer_socket *) zmalloc(sizeof(struct consumer_socket));
        if (socket == NULL) {
                PERROR("zmalloc consumer socket");
                goto error;
        }

        socket->fd_ptr = fd;
        lttng_ht_node_init_ulong(&socket->node, *fd);

error:
        return socket;
}

/*
 * Add consumer socket to consumer output object. Read side lock must be
 * acquired before calling this function.
 */
void consumer_add_socket(struct consumer_socket *sock,
                struct consumer_output *consumer)
{
        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(consumer);
        ASSERT_RCU_READ_LOCKED();

        lttng_ht_add_unique_ulong(consumer->socks, &sock->node);
}

/*
 * Delete consumer socket to consumer output object. Read side lock must be
 * acquired before calling this function.
 */
void consumer_del_socket(struct consumer_socket *sock,
                struct consumer_output *consumer)
{
        int ret;
        struct lttng_ht_iter iter;

        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(consumer);
        ASSERT_RCU_READ_LOCKED();

        iter.iter.node = &sock->node.node;
        ret = lttng_ht_del(consumer->socks, &iter);
        LTTNG_ASSERT(!ret);
}

/*
 * RCU destroy call function.
 */
static void destroy_socket_rcu(struct rcu_head *head)
{
        struct lttng_ht_node_ulong *node =
                caa_container_of(head, struct lttng_ht_node_ulong, head);
        struct consumer_socket *socket =
                caa_container_of(node, struct consumer_socket, node);

        free(socket);
}

/*
 * Destroy and free socket pointer in a call RCU. The call must either:
 *  - have acquired the read side lock before calling this function, or
 *  - guarantee the validity of the `struct consumer_socket` object for the
 *    duration of the call.
 */
void consumer_destroy_socket(struct consumer_socket *sock)
{
        LTTNG_ASSERT(sock);

        /*
         * We DO NOT close the file descriptor here since it is global to the
         * session daemon and is closed only if the consumer dies or a custom
         * consumer was registered,
         */
        if (sock->registered) {
                DBG3("Consumer socket was registered. Closing fd %d", *sock->fd_ptr);
                lttcomm_close_unix_sock(*sock->fd_ptr);
        }

        call_rcu(&sock->node.head, destroy_socket_rcu);
}

/*
 * Allocate and assign data to a consumer_output object.
 *
 * Return pointer to structure.
 */
struct consumer_output *consumer_create_output(enum consumer_dst_type type)
{
        struct consumer_output *output = NULL;

        output = (consumer_output *) zmalloc(sizeof(struct consumer_output));
        if (output == NULL) {
                PERROR("zmalloc consumer_output");
                goto error;
        }

        /* By default, consumer output is enabled */
        output->enabled = 1;
        output->type = type;
        output->net_seq_index = (uint64_t) -1ULL;
        urcu_ref_init(&output->ref);

        output->socks = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);

error:
        return output;
}

/*
 * Iterate over the consumer output socket hash table and destroy them. The
 * socket file descriptor are only closed if the consumer output was
 * registered meaning it's an external consumer.
 */
void consumer_destroy_output_sockets(struct consumer_output *obj)
{
        struct lttng_ht_iter iter;
        struct consumer_socket *socket;

        if (!obj->socks) {
                return;
        }

        rcu_read_lock();
        cds_lfht_for_each_entry(obj->socks->ht, &iter.iter, socket, node.node) {
                consumer_del_socket(socket, obj);
                consumer_destroy_socket(socket);
        }
        rcu_read_unlock();
}

/*
 * Delete the consumer_output object from the list and free the ptr.
 */
static void consumer_release_output(struct urcu_ref *ref)
{
        struct consumer_output *obj =
                caa_container_of(ref, struct consumer_output, ref);

        consumer_destroy_output_sockets(obj);

        if (obj->socks) {
                /* Finally destroy HT */
                lttng_ht_destroy(obj->socks);
        }

        free(obj);
}

/*
 * Get the consumer_output object.
 */
void consumer_output_get(struct consumer_output *obj)
{
        urcu_ref_get(&obj->ref);
}

/*
 * Put the consumer_output object.
 */
void consumer_output_put(struct consumer_output *obj)
{
        if (!obj) {
                return;
        }
        urcu_ref_put(&obj->ref, consumer_release_output);
}

/*
 * Copy consumer output and returned the newly allocated copy.
 */
struct consumer_output *consumer_copy_output(struct consumer_output *src)
{
        int ret;
        struct consumer_output *output;

        LTTNG_ASSERT(src);

        output = consumer_create_output(src->type);
        if (output == NULL) {
                goto end;
        }
        output->enabled = src->enabled;
        output->net_seq_index = src->net_seq_index;
        memcpy(output->domain_subdir, src->domain_subdir,
                        sizeof(output->domain_subdir));
        output->snapshot = src->snapshot;
        output->relay_major_version = src->relay_major_version;
        output->relay_minor_version = src->relay_minor_version;
        output->relay_allows_clear = src->relay_allows_clear;
        memcpy(&output->dst, &src->dst, sizeof(output->dst));
        ret = consumer_copy_sockets(output, src);
        if (ret < 0) {
                goto error_put;
        }
end:
        return output;

error_put:
        consumer_output_put(output);
        return NULL;
}

/*
 * Copy consumer sockets from src to dst.
 *
 * Return 0 on success or else a negative value.
 */
int consumer_copy_sockets(struct consumer_output *dst,
                struct consumer_output *src)
{
        int ret = 0;
        struct lttng_ht_iter iter;
        struct consumer_socket *socket, *copy_sock;

        LTTNG_ASSERT(dst);
        LTTNG_ASSERT(src);

        rcu_read_lock();
        cds_lfht_for_each_entry(src->socks->ht, &iter.iter, socket, node.node) {
                /* Ignore socket that are already there. */
                copy_sock = consumer_find_socket(*socket->fd_ptr, dst);
                if (copy_sock) {
                        continue;
                }

                /* Create new socket object. */
                copy_sock = consumer_allocate_socket(socket->fd_ptr);
                if (copy_sock == NULL) {
                        rcu_read_unlock();
                        ret = -ENOMEM;
                        goto error;
                }

                copy_sock->registered = socket->registered;
                /*
                 * This is valid because this lock is shared accross all consumer
                 * object being the global lock of the consumer data structure of the
                 * session daemon.
                 */
                copy_sock->lock = socket->lock;
                consumer_add_socket(copy_sock, dst);
        }
        rcu_read_unlock();

error:
        return ret;
}

/*
 * Set network URI to the consumer output.
 *
 * Return 0 on success. Return 1 if the URI were equal. Else, negative value on
 * error.
 */
int consumer_set_network_uri(const struct ltt_session *session,
                struct consumer_output *output,
                struct lttng_uri *uri)
{
        int ret;
        struct lttng_uri *dst_uri = NULL;

        /* Code flow error safety net. */
        LTTNG_ASSERT(output);
        LTTNG_ASSERT(uri);

        switch (uri->stype) {
        case LTTNG_STREAM_CONTROL:
                dst_uri = &output->dst.net.control;
                output->dst.net.control_isset = 1;
                if (uri->port == 0) {
                        /* Assign default port. */
                        uri->port = DEFAULT_NETWORK_CONTROL_PORT;
                } else {
                        if (output->dst.net.data_isset && uri->port ==
                                        output->dst.net.data.port) {
                                ret = -LTTNG_ERR_INVALID;
                                goto error;
                        }
                }
                DBG3("Consumer control URI set with port %d", uri->port);
                break;
        case LTTNG_STREAM_DATA:
                dst_uri = &output->dst.net.data;
                output->dst.net.data_isset = 1;
                if (uri->port == 0) {
                        /* Assign default port. */
                        uri->port = DEFAULT_NETWORK_DATA_PORT;
                } else {
                        if (output->dst.net.control_isset && uri->port ==
                                        output->dst.net.control.port) {
                                ret = -LTTNG_ERR_INVALID;
                                goto error;
                        }
                }
                DBG3("Consumer data URI set with port %d", uri->port);
                break;
        default:
                ERR("Set network uri type unknown %d", uri->stype);
                ret = -LTTNG_ERR_INVALID;
                goto error;
        }

        ret = uri_compare(dst_uri, uri);
        if (!ret) {
                /* Same URI, don't touch it and return success. */
                DBG3("URI network compare are the same");
                goto equal;
        }

        /* URIs were not equal, replacing it. */
        memcpy(dst_uri, uri, sizeof(struct lttng_uri));
        output->type = CONSUMER_DST_NET;
        if (dst_uri->stype != LTTNG_STREAM_CONTROL) {
                /* Only the control uri needs to contain the path. */
                goto end;
        }

        /*
         * If the user has specified a subdir as part of the control
         * URL, the session's base output directory is:
         *   /RELAYD_OUTPUT_PATH/HOSTNAME/USER_SPECIFIED_DIR
         *
         * Hence, the "base_dir" from which all stream files and
         * session rotation chunks are created takes the form
         *   /HOSTNAME/USER_SPECIFIED_DIR
         *
         * If the user has not specified an output directory as part of
         * the control URL, the base output directory has the form:
         *   /RELAYD_OUTPUT_PATH/HOSTNAME/SESSION_NAME-CREATION_TIME
         *
         * Hence, the "base_dir" from which all stream files and
         * session rotation chunks are created takes the form
         *   /HOSTNAME/SESSION_NAME-CREATION_TIME
         *
         * Note that automatically generated session names already
         * contain the session's creation time. In that case, the
         * creation time is omitted to prevent it from being duplicated
         * in the final directory hierarchy.
         */
        if (*uri->subdir) {
                if (strstr(uri->subdir, "../")) {
                        ERR("Network URI subdirs are not allowed to walk up the path hierarchy");
                        ret = -LTTNG_ERR_INVALID;
                        goto error;
                }
                ret = snprintf(output->dst.net.base_dir,
                                sizeof(output->dst.net.base_dir),
                                "/%s/%s/", session->hostname, uri->subdir);
        } else {
                if (session->has_auto_generated_name) {
                        ret = snprintf(output->dst.net.base_dir,
                                        sizeof(output->dst.net.base_dir),
                                        "/%s/%s/", session->hostname,
                                        session->name);
                } else {
                        char session_creation_datetime[16];
                        size_t strftime_ret;
                        struct tm *timeinfo;

                        timeinfo = localtime(&session->creation_time);
                        if (!timeinfo) {
                                ret = -LTTNG_ERR_FATAL;
                                goto error;
                        }
                        strftime_ret = strftime(session_creation_datetime,
                                        sizeof(session_creation_datetime),
                                        "%Y%m%d-%H%M%S", timeinfo);
                        if (strftime_ret == 0) {
                                ERR("Failed to format session creation timestamp while setting network URI");
                                ret = -LTTNG_ERR_FATAL;
                                goto error;
                        }
                        ret = snprintf(output->dst.net.base_dir,
                                        sizeof(output->dst.net.base_dir),
                                        "/%s/%s-%s/", session->hostname,
                                        session->name,
                                        session_creation_datetime);
                }
        }
        if (ret >= sizeof(output->dst.net.base_dir)) {
                ret = -LTTNG_ERR_INVALID;
                ERR("Truncation occurred while setting network output base directory");
                goto error;
        } else if (ret == -1) {
                ret = -LTTNG_ERR_INVALID;
                PERROR("Error occurred while setting network output base directory");
                goto error;
        }

        DBG3("Consumer set network uri base_dir path %s",
                        output->dst.net.base_dir);

end:
        return 0;
equal:
        return 1;
error:
        return ret;
}

/*
 * Send file descriptor to consumer via sock.
 *
 * The consumer socket lock must be held by the caller.
 */
int consumer_send_fds(struct consumer_socket *sock, const int *fds,
                size_t nb_fd)
{
        int ret;

        LTTNG_ASSERT(fds);
        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(nb_fd > 0);
        LTTNG_ASSERT(pthread_mutex_trylock(sock->lock) == EBUSY);

        ret = lttcomm_send_fds_unix_sock(*sock->fd_ptr, fds, nb_fd);
        if (ret < 0) {
                /* The above call will print a PERROR on error. */
                DBG("Error when sending consumer fds on sock %d", *sock->fd_ptr);
                goto error;
        }

        ret = consumer_recv_status_reply(sock);
error:
        return ret;
}

/*
 * Consumer send communication message structure to consumer.
 *
 * The consumer socket lock must be held by the caller.
 */
int consumer_send_msg(struct consumer_socket *sock,
                const struct lttcomm_consumer_msg *msg)
{
        int ret;

        LTTNG_ASSERT(msg);
        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(pthread_mutex_trylock(sock->lock) == EBUSY);

        ret = consumer_socket_send(sock, msg, sizeof(struct lttcomm_consumer_msg));
        if (ret < 0) {
                goto error;
        }

        ret = consumer_recv_status_reply(sock);

error:
        return ret;
}

/*
 * Consumer send channel communication message structure to consumer.
 *
 * The consumer socket lock must be held by the caller.
 */
int consumer_send_channel(struct consumer_socket *sock,
                struct lttcomm_consumer_msg *msg)
{
        int ret;

        LTTNG_ASSERT(msg);
        LTTNG_ASSERT(sock);

        ret = consumer_send_msg(sock, msg);
        if (ret < 0) {
                goto error;
        }

error:
        return ret;
}

/*
 * Populate the given consumer msg structure with the ask_channel command
 * information.
 */
void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t subbuf_size,
                uint64_t num_subbuf,
                int overwrite,
                unsigned int switch_timer_interval,
                unsigned int read_timer_interval,
                unsigned int live_timer_interval,
                bool is_in_live_session,
                unsigned int monitor_timer_interval,
                int output,
                int type,
                uint64_t session_id,
                const char *pathname,
                const char *name,
                uint64_t relayd_id,
                uint64_t key,
                unsigned char *uuid,
                uint32_t chan_id,
                uint64_t tracefile_size,
                uint64_t tracefile_count,
                uint64_t session_id_per_pid,
                unsigned int monitor,
                uint32_t ust_app_uid,
                int64_t blocking_timeout,
                const char *root_shm_path,
                const char *shm_path,
                struct lttng_trace_chunk *trace_chunk,
                const struct lttng_credentials *buffer_credentials)
{
        LTTNG_ASSERT(msg);

        /* Zeroed structure */
        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
        msg->u.ask_channel.buffer_credentials.uid = UINT32_MAX;
        msg->u.ask_channel.buffer_credentials.gid = UINT32_MAX;

        if (trace_chunk) {
                uint64_t chunk_id;
                enum lttng_trace_chunk_status chunk_status;

                chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
                LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
                LTTNG_OPTIONAL_SET(&msg->u.ask_channel.chunk_id, chunk_id);
        }
        msg->u.ask_channel.buffer_credentials.uid =
                        lttng_credentials_get_uid(buffer_credentials);
        msg->u.ask_channel.buffer_credentials.gid =
                        lttng_credentials_get_gid(buffer_credentials);

        msg->cmd_type = LTTNG_CONSUMER_ASK_CHANNEL_CREATION;
        msg->u.ask_channel.subbuf_size = subbuf_size;
        msg->u.ask_channel.num_subbuf = num_subbuf ;
        msg->u.ask_channel.overwrite = overwrite;
        msg->u.ask_channel.switch_timer_interval = switch_timer_interval;
        msg->u.ask_channel.read_timer_interval = read_timer_interval;
        msg->u.ask_channel.live_timer_interval = live_timer_interval;
        msg->u.ask_channel.is_live = is_in_live_session;
        msg->u.ask_channel.monitor_timer_interval = monitor_timer_interval;
        msg->u.ask_channel.output = output;
        msg->u.ask_channel.type = type;
        msg->u.ask_channel.session_id = session_id;
        msg->u.ask_channel.session_id_per_pid = session_id_per_pid;
        msg->u.ask_channel.relayd_id = relayd_id;
        msg->u.ask_channel.key = key;
        msg->u.ask_channel.chan_id = chan_id;
        msg->u.ask_channel.tracefile_size = tracefile_size;
        msg->u.ask_channel.tracefile_count = tracefile_count;
        msg->u.ask_channel.monitor = monitor;
        msg->u.ask_channel.ust_app_uid = ust_app_uid;
        msg->u.ask_channel.blocking_timeout = blocking_timeout;

        memcpy(msg->u.ask_channel.uuid, uuid, sizeof(msg->u.ask_channel.uuid));

        if (pathname) {
                strncpy(msg->u.ask_channel.pathname, pathname,
                                sizeof(msg->u.ask_channel.pathname));
                msg->u.ask_channel.pathname[sizeof(msg->u.ask_channel.pathname)-1] = '\0';
        }

        strncpy(msg->u.ask_channel.name, name, sizeof(msg->u.ask_channel.name));
        msg->u.ask_channel.name[sizeof(msg->u.ask_channel.name) - 1] = '\0';

        if (root_shm_path) {
                strncpy(msg->u.ask_channel.root_shm_path, root_shm_path,
                        sizeof(msg->u.ask_channel.root_shm_path));
                msg->u.ask_channel.root_shm_path[sizeof(msg->u.ask_channel.root_shm_path) - 1] = '\0';
        }
        if (shm_path) {
                strncpy(msg->u.ask_channel.shm_path, shm_path,
                        sizeof(msg->u.ask_channel.shm_path));
                msg->u.ask_channel.shm_path[sizeof(msg->u.ask_channel.shm_path) - 1] = '\0';
        }
}

/*
 * Init channel communication message structure.
 */
void consumer_init_add_channel_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t channel_key,
                uint64_t session_id,
                const char *pathname,
                uid_t uid,
                gid_t gid,
                uint64_t relayd_id,
                const char *name,
                unsigned int nb_init_streams,
                enum lttng_event_output output,
                int type,
                uint64_t tracefile_size,
                uint64_t tracefile_count,
                unsigned int monitor,
                unsigned int live_timer_interval,
                bool is_in_live_session,
                unsigned int monitor_timer_interval,
                struct lttng_trace_chunk *trace_chunk)
{
        LTTNG_ASSERT(msg);

        /* Zeroed structure */
        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));

        if (trace_chunk) {
                uint64_t chunk_id;
                enum lttng_trace_chunk_status chunk_status;

                chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
                LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
                LTTNG_OPTIONAL_SET(&msg->u.channel.chunk_id, chunk_id);
        }

        /* Send channel */
        msg->cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
        msg->u.channel.channel_key = channel_key;
        msg->u.channel.session_id = session_id;
        msg->u.channel.relayd_id = relayd_id;
        msg->u.channel.nb_init_streams = nb_init_streams;
        msg->u.channel.output = output;
        msg->u.channel.type = type;
        msg->u.channel.tracefile_size = tracefile_size;
        msg->u.channel.tracefile_count = tracefile_count;
        msg->u.channel.monitor = monitor;
        msg->u.channel.live_timer_interval = live_timer_interval;
        msg->u.channel.is_live = is_in_live_session;
        msg->u.channel.monitor_timer_interval = monitor_timer_interval;

        strncpy(msg->u.channel.pathname, pathname,
                        sizeof(msg->u.channel.pathname));
        msg->u.channel.pathname[sizeof(msg->u.channel.pathname) - 1] = '\0';

        strncpy(msg->u.channel.name, name, sizeof(msg->u.channel.name));
        msg->u.channel.name[sizeof(msg->u.channel.name) - 1] = '\0';
}

/*
 * Init stream communication message structure.
 */
void consumer_init_add_stream_comm_msg(struct lttcomm_consumer_msg *msg,
                uint64_t channel_key,
                uint64_t stream_key,
                int32_t cpu)
{
        LTTNG_ASSERT(msg);

        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));

        msg->cmd_type = LTTNG_CONSUMER_ADD_STREAM;
        msg->u.stream.channel_key = channel_key;
        msg->u.stream.stream_key = stream_key;
        msg->u.stream.cpu = cpu;
}

void consumer_init_streams_sent_comm_msg(struct lttcomm_consumer_msg *msg,
                enum lttng_consumer_command cmd,
                uint64_t channel_key, uint64_t net_seq_idx)
{
        LTTNG_ASSERT(msg);

        memset(msg, 0, sizeof(struct lttcomm_consumer_msg));

        msg->cmd_type = cmd;
        msg->u.sent_streams.channel_key = channel_key;
        msg->u.sent_streams.net_seq_idx = net_seq_idx;
}

/*
 * Send stream communication structure to the consumer.
 */
int consumer_send_stream(struct consumer_socket *sock,
                struct consumer_output *dst, struct lttcomm_consumer_msg *msg,
                const int *fds, size_t nb_fd)
{
        int ret;

        LTTNG_ASSERT(msg);
        LTTNG_ASSERT(dst);
        LTTNG_ASSERT(sock);
        LTTNG_ASSERT(fds);

        ret = consumer_send_msg(sock, msg);
        if (ret < 0) {
                goto error;
        }

        ret = consumer_send_fds(sock, fds, nb_fd);
        if (ret < 0) {
                goto error;
        }

error:
        return ret;
}

/*
 * Send relayd socket to consumer associated with a session name.
 *
 * The consumer socket lock must be held by the caller.
 *
 * On success return positive value. On error, negative value.
 */
int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
                struct lttcomm_relayd_sock *rsock, struct consumer_output *consumer,
                enum lttng_stream_type type, uint64_t session_id,
                const char *session_name, const char *hostname,
                const char *base_path, int session_live_timer,
                const uint64_t *current_chunk_id, time_t session_creation_time,
                bool session_name_contains_creation_time)
{
        int ret;
        int fd;
        struct lttcomm_consumer_msg msg;

        /* Code flow error. Safety net. */
        LTTNG_ASSERT(rsock);
        LTTNG_ASSERT(consumer);
        LTTNG_ASSERT(consumer_sock);

        memset(&msg, 0, sizeof(msg));
        /* Bail out if consumer is disabled */
        if (!consumer->enabled) {
                ret = LTTNG_OK;
                goto error;
        }

        if (type == LTTNG_STREAM_CONTROL) {
                char output_path[LTTNG_PATH_MAX] = {};
                uint64_t relayd_session_id;

                ret = relayd_create_session(rsock, &relayd_session_id,
                                session_name, hostname, base_path,
                                session_live_timer, consumer->snapshot,
                                session_id, the_sessiond_uuid, current_chunk_id,
                                session_creation_time,
                                session_name_contains_creation_time,
                                output_path);
                if (ret < 0) {
                        /* Close the control socket. */
                        (void) relayd_close(rsock);
                        goto error;
                }
                msg.u.relayd_sock.relayd_session_id = relayd_session_id;
                DBG("Created session on relay, output path reply: %s",
                        output_path);
        }

        msg.cmd_type = LTTNG_CONSUMER_ADD_RELAYD_SOCKET;
        /*
         * Assign network consumer output index using the temporary consumer since
         * this call should only be made from within a set_consumer_uri() function
         * call in the session daemon.
         */
        msg.u.relayd_sock.net_index = consumer->net_seq_index;
        msg.u.relayd_sock.type = type;
        msg.u.relayd_sock.session_id = session_id;
        memcpy(&msg.u.relayd_sock.sock, rsock, sizeof(msg.u.relayd_sock.sock));

        DBG3("Sending relayd sock info to consumer on %d", *consumer_sock->fd_ptr);
        ret = consumer_send_msg(consumer_sock, &msg);
        if (ret < 0) {
                goto error;
        }

        DBG3("Sending relayd socket file descriptor to consumer");
        fd = rsock->sock.fd;
        ret = consumer_send_fds(consumer_sock, &fd, 1);
        if (ret < 0) {
                goto error;
        }

        DBG2("Consumer relayd socket sent");

error:
        return ret;
}

static
int consumer_send_pipe(struct consumer_socket *consumer_sock,
                enum lttng_consumer_command cmd, int pipe)
{
        int ret;
        struct lttcomm_consumer_msg msg;
        const char *pipe_name;
        const char *command_name;

        switch (cmd) {
        case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
                pipe_name = "channel monitor";
                command_name = "SET_CHANNEL_MONITOR_PIPE";
                break;
        default:
                ERR("Unexpected command received in %s (cmd = %d)", __func__,
                                (int) cmd);
                abort();
        }

        /* Code flow error. Safety net. */

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = cmd;

        pthread_mutex_lock(consumer_sock->lock);
        DBG3("Sending %s command to consumer", command_name);
        ret = consumer_send_msg(consumer_sock, &msg);
        if (ret < 0) {
                goto error;
        }

        DBG3("Sending %s pipe %d to consumer on socket %d",
                        pipe_name,
                        pipe, *consumer_sock->fd_ptr);
        ret = consumer_send_fds(consumer_sock, &pipe, 1);
        if (ret < 0) {
                goto error;
        }

        DBG2("%s pipe successfully sent", pipe_name);
error:
        pthread_mutex_unlock(consumer_sock->lock);
        return ret;
}

int consumer_send_channel_monitor_pipe(struct consumer_socket *consumer_sock,
                int pipe)
{
        return consumer_send_pipe(consumer_sock,
                        LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE, pipe);
}

/*
 * Ask the consumer if the data is pending for the specific session id.
 * Returns 1 if data is pending, 0 otherwise, or < 0 on error.
 */
int consumer_is_data_pending(uint64_t session_id,
                struct consumer_output *consumer)
{
        int ret;
        int32_t ret_code = 0;  /* Default is that the data is NOT pending */
        struct consumer_socket *socket;
        struct lttng_ht_iter iter;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(consumer);

        DBG3("Consumer data pending for id %" PRIu64, session_id);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_DATA_PENDING;
        msg.u.data_pending.session_id = session_id;

        /* Send command for each consumer */
        rcu_read_lock();
        cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
                        node.node) {
                pthread_mutex_lock(socket->lock);
                ret = consumer_socket_send(socket, &msg, sizeof(msg));
                if (ret < 0) {
                        pthread_mutex_unlock(socket->lock);
                        goto error_unlock;
                }

                /*
                 * No need for a recv reply status because the answer to the command is
                 * the reply status message.
                 */

                ret = consumer_socket_recv(socket, &ret_code, sizeof(ret_code));
                if (ret < 0) {
                        pthread_mutex_unlock(socket->lock);
                        goto error_unlock;
                }
                pthread_mutex_unlock(socket->lock);

                if (ret_code == 1) {
                        break;
                }
        }
        rcu_read_unlock();

        DBG("Consumer data is %s pending for session id %" PRIu64,
                        ret_code == 1 ? "" : "NOT", session_id);
        return ret_code;

error_unlock:
        rcu_read_unlock();
        return -1;
}

/*
 * Send a flush command to consumer using the given channel key.
 *
 * Return 0 on success else a negative value.
 */
int consumer_flush_channel(struct consumer_socket *socket, uint64_t key)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG2("Consumer flush channel key %" PRIu64, key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_FLUSH_CHANNEL;
        msg.u.flush_channel.key = key;

        pthread_mutex_lock(socket->lock);
        health_code_update();

        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto end;
        }

end:
        health_code_update();
        pthread_mutex_unlock(socket->lock);
        return ret;
}

/*
 * Send a clear quiescent command to consumer using the given channel key.
 *
 * Return 0 on success else a negative value.
 */
int consumer_clear_quiescent_channel(struct consumer_socket *socket, uint64_t key)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG2("Consumer clear quiescent channel key %" PRIu64, key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL;
        msg.u.clear_quiescent_channel.key = key;

        pthread_mutex_lock(socket->lock);
        health_code_update();

        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto end;
        }

end:
        health_code_update();
        pthread_mutex_unlock(socket->lock);
        return ret;
}

/*
 * Send a close metadata command to consumer using the given channel key.
 * Called with registry lock held.
 *
 * Return 0 on success else a negative value.
 */
int consumer_close_metadata(struct consumer_socket *socket,
                uint64_t metadata_key)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG2("Consumer close metadata channel key %" PRIu64, metadata_key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_CLOSE_METADATA;
        msg.u.close_metadata.key = metadata_key;

        pthread_mutex_lock(socket->lock);
        health_code_update();

        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto end;
        }

end:
        health_code_update();
        pthread_mutex_unlock(socket->lock);
        return ret;
}

/*
 * Send a setup metdata command to consumer using the given channel key.
 *
 * Return 0 on success else a negative value.
 */
int consumer_setup_metadata(struct consumer_socket *socket,
                uint64_t metadata_key)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG2("Consumer setup metadata channel key %" PRIu64, metadata_key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_SETUP_METADATA;
        msg.u.setup_metadata.key = metadata_key;

        pthread_mutex_lock(socket->lock);
        health_code_update();

        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto end;
        }

end:
        health_code_update();
        pthread_mutex_unlock(socket->lock);
        return ret;
}

/*
 * Send metadata string to consumer.
 * RCU read-side lock must be held to guarantee existence of socket.
 *
 * Return 0 on success else a negative value.
 */
int consumer_push_metadata(struct consumer_socket *socket,
                uint64_t metadata_key, char *metadata_str, size_t len,
                size_t target_offset, uint64_t version)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);
        ASSERT_RCU_READ_LOCKED();

        DBG2("Consumer push metadata to consumer socket %d", *socket->fd_ptr);

        pthread_mutex_lock(socket->lock);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_PUSH_METADATA;
        msg.u.push_metadata.key = metadata_key;
        msg.u.push_metadata.target_offset = target_offset;
        msg.u.push_metadata.len = len;
        msg.u.push_metadata.version = version;

        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        if (ret < 0 || len == 0) {
                goto end;
        }

        DBG3("Consumer pushing metadata on sock %d of len %zu", *socket->fd_ptr,
                        len);

        ret = consumer_socket_send(socket, metadata_str, len);
        if (ret < 0) {
                goto end;
        }

        health_code_update();
        ret = consumer_recv_status_reply(socket);
        if (ret < 0) {
                goto end;
        }

end:
        pthread_mutex_unlock(socket->lock);
        health_code_update();
        return ret;
}

/*
 * Ask the consumer to snapshot a specific channel using the key.
 *
 * Returns LTTNG_OK on success or else an LTTng error code.
 */
enum lttng_error_code consumer_snapshot_channel(struct consumer_socket *socket,
                uint64_t key, const struct consumer_output *output, int metadata,
                uid_t uid, gid_t gid, const char *channel_path, int wait,
                uint64_t nb_packets_per_stream)
{
        int ret;
        enum lttng_error_code status = LTTNG_OK;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);
        LTTNG_ASSERT(output);

        DBG("Consumer snapshot channel key %" PRIu64, key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_SNAPSHOT_CHANNEL;
        msg.u.snapshot_channel.key = key;
        msg.u.snapshot_channel.nb_packets_per_stream = nb_packets_per_stream;
        msg.u.snapshot_channel.metadata = metadata;

        if (output->type == CONSUMER_DST_NET) {
                msg.u.snapshot_channel.relayd_id =
                                output->net_seq_index;
                msg.u.snapshot_channel.use_relayd = 1;
        } else {
                msg.u.snapshot_channel.relayd_id = (uint64_t) -1ULL;
        }
        ret = lttng_strncpy(msg.u.snapshot_channel.pathname,
                        channel_path,
                        sizeof(msg.u.snapshot_channel.pathname));
        if (ret < 0) {
                ERR("Snapshot path exceeds the maximal allowed length of %zu bytes (%zu bytes required) with path \"%s\"",
                                sizeof(msg.u.snapshot_channel.pathname),
                                strlen(channel_path),
                                channel_path);
                status = LTTNG_ERR_SNAPSHOT_FAIL;
                goto error;
        }

        health_code_update();
        pthread_mutex_lock(socket->lock);
        ret = consumer_send_msg(socket, &msg);
        pthread_mutex_unlock(socket->lock);
        if (ret < 0) {
                switch (-ret) {
                case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
                        status = LTTNG_ERR_CHAN_NOT_FOUND;
                        break;
                default:
                        status = LTTNG_ERR_SNAPSHOT_FAIL;
                        break;
                }
                goto error;
        }

error:
        health_code_update();
        return status;
}

/*
 * Ask the consumer the number of discarded events for a channel.
 */
int consumer_get_discarded_events(uint64_t session_id, uint64_t channel_key,
                struct consumer_output *consumer, uint64_t *discarded)
{
        int ret;
        struct consumer_socket *socket;
        struct lttng_ht_iter iter;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(consumer);

        DBG3("Consumer discarded events id %" PRIu64, session_id);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_DISCARDED_EVENTS;
        msg.u.discarded_events.session_id = session_id;
        msg.u.discarded_events.channel_key = channel_key;

        *discarded = 0;

        /* Send command for each consumer */
        rcu_read_lock();
        cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
                        node.node) {
                uint64_t consumer_discarded = 0;
                pthread_mutex_lock(socket->lock);
                ret = consumer_socket_send(socket, &msg, sizeof(msg));
                if (ret < 0) {
                        pthread_mutex_unlock(socket->lock);
                        goto end;
                }

                /*
                 * No need for a recv reply status because the answer to the
                 * command is the reply status message.
                 */
                ret = consumer_socket_recv(socket, &consumer_discarded,
                                sizeof(consumer_discarded));
                if (ret < 0) {
                        ERR("get discarded events");
                        pthread_mutex_unlock(socket->lock);
                        goto end;
                }
                pthread_mutex_unlock(socket->lock);
                *discarded += consumer_discarded;
        }
        ret = 0;
        DBG("Consumer discarded %" PRIu64 " events in session id %" PRIu64,
                        *discarded, session_id);

end:
        rcu_read_unlock();
        return ret;
}

/*
 * Ask the consumer the number of lost packets for a channel.
 */
int consumer_get_lost_packets(uint64_t session_id, uint64_t channel_key,
                struct consumer_output *consumer, uint64_t *lost)
{
        int ret;
        struct consumer_socket *socket;
        struct lttng_ht_iter iter;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(consumer);

        DBG3("Consumer lost packets id %" PRIu64, session_id);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_LOST_PACKETS;
        msg.u.lost_packets.session_id = session_id;
        msg.u.lost_packets.channel_key = channel_key;

        *lost = 0;

        /* Send command for each consumer */
        rcu_read_lock();
        cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
                        node.node) {
                uint64_t consumer_lost = 0;
                pthread_mutex_lock(socket->lock);
                ret = consumer_socket_send(socket, &msg, sizeof(msg));
                if (ret < 0) {
                        pthread_mutex_unlock(socket->lock);
                        goto end;
                }

                /*
                 * No need for a recv reply status because the answer to the
                 * command is the reply status message.
                 */
                ret = consumer_socket_recv(socket, &consumer_lost,
                                sizeof(consumer_lost));
                if (ret < 0) {
                        ERR("get lost packets");
                        pthread_mutex_unlock(socket->lock);
                        goto end;
                }
                pthread_mutex_unlock(socket->lock);
                *lost += consumer_lost;
        }
        ret = 0;
        DBG("Consumer lost %" PRIu64 " packets in session id %" PRIu64,
                        *lost, session_id);

end:
        rcu_read_unlock();
        return ret;
}

/*
 * Ask the consumer to rotate a channel.
 *
 * The new_chunk_id is the session->rotate_count that has been incremented
 * when the rotation started. On the relay, this allows to keep track in which
 * chunk each stream is currently writing to (for the rotate_pending operation).
 */
int consumer_rotate_channel(struct consumer_socket *socket, uint64_t key,
                uid_t uid, gid_t gid, struct consumer_output *output,
                bool is_metadata_channel)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG("Consumer rotate channel key %" PRIu64, key);

        pthread_mutex_lock(socket->lock);
        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_ROTATE_CHANNEL;
        msg.u.rotate_channel.key = key;
        msg.u.rotate_channel.metadata = !!is_metadata_channel;

        if (output->type == CONSUMER_DST_NET) {
                msg.u.rotate_channel.relayd_id = output->net_seq_index;
        } else {
                msg.u.rotate_channel.relayd_id = (uint64_t) -1ULL;
        }

        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                switch (-ret) {
                case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
                        ret = -LTTNG_ERR_CHAN_NOT_FOUND;
                        break;
                default:
                        ret = -LTTNG_ERR_ROTATION_FAIL_CONSUMER;
                        break;
                }
                goto error;
        }
error:
        pthread_mutex_unlock(socket->lock);
        health_code_update();
        return ret;
}

int consumer_open_channel_packets(struct consumer_socket *socket, uint64_t key)
{
        int ret;
        lttcomm_consumer_msg msg = {
                .cmd_type = LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS,
        };
        msg.u.open_channel_packets.key = key;

        LTTNG_ASSERT(socket);

        DBG("Consumer open channel packets: channel key = %" PRIu64, key);

        health_code_update();

        pthread_mutex_lock(socket->lock);
        ret = consumer_send_msg(socket, &msg);
        pthread_mutex_unlock(socket->lock);
        if (ret < 0) {
                goto error_socket;
        }

error_socket:
        health_code_update();
        return ret;
}

int consumer_clear_channel(struct consumer_socket *socket, uint64_t key)
{
        int ret;
        struct lttcomm_consumer_msg msg;

        LTTNG_ASSERT(socket);

        DBG("Consumer clear channel %" PRIu64, key);

        memset(&msg, 0, sizeof(msg));
        msg.cmd_type = LTTNG_CONSUMER_CLEAR_CHANNEL;
        msg.u.clear_channel.key = key;

        health_code_update();

        pthread_mutex_lock(socket->lock);
        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto error_socket;
        }

error_socket:
        pthread_mutex_unlock(socket->lock);

        health_code_update();
        return ret;
}

int consumer_init(struct consumer_socket *socket,
                const lttng_uuid sessiond_uuid)
{
        int ret;
        struct lttcomm_consumer_msg msg = {
                .cmd_type = LTTNG_CONSUMER_INIT,
        };

        LTTNG_ASSERT(socket);

        DBG("Sending consumer initialization command");
        lttng_uuid_copy(msg.u.init.sessiond_uuid, sessiond_uuid);

        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        if (ret < 0) {
                goto error;
        }

error:
        health_code_update();
        return ret;
}

/*
 * Ask the consumer to create a new chunk for a given session.
 *
 * Called with the consumer socket lock held.
 */
int consumer_create_trace_chunk(struct consumer_socket *socket,
                uint64_t relayd_id, uint64_t session_id,
                struct lttng_trace_chunk *chunk,
                const char *domain_subdir)
{
        int ret;
        enum lttng_trace_chunk_status chunk_status;
        struct lttng_credentials chunk_credentials;
        const struct lttng_directory_handle *chunk_directory_handle = NULL;
        struct lttng_directory_handle *domain_handle = NULL;
        int domain_dirfd;
        const char *chunk_name;
        bool chunk_name_overridden;
        uint64_t chunk_id;
        time_t creation_timestamp;
        char creation_timestamp_buffer[ISO8601_STR_LEN];
        const char *creation_timestamp_str = "(none)";
        const bool chunk_has_local_output = relayd_id == -1ULL;
        enum lttng_trace_chunk_status tc_status;
        struct lttcomm_consumer_msg msg = {
                .cmd_type = LTTNG_CONSUMER_CREATE_TRACE_CHUNK,
        };
        msg.u.create_trace_chunk.session_id = session_id;

        LTTNG_ASSERT(socket);
        LTTNG_ASSERT(chunk);

        if (relayd_id != -1ULL) {
                LTTNG_OPTIONAL_SET(&msg.u.create_trace_chunk.relayd_id,
                                relayd_id);
        }

        chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name,
                        &chunk_name_overridden);
        if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK &&
                        chunk_status != LTTNG_TRACE_CHUNK_STATUS_NONE) {
                ERR("Failed to get name of trace chunk");
                ret = -LTTNG_ERR_FATAL;
                goto error;
        }
        if (chunk_name_overridden) {
                ret = lttng_strncpy(msg.u.create_trace_chunk.override_name,
                                chunk_name,
                                sizeof(msg.u.create_trace_chunk.override_name));
                if (ret) {
                        ERR("Trace chunk name \"%s\" exceeds the maximal length allowed by the consumer protocol",
                                        chunk_name);
                        ret = -LTTNG_ERR_FATAL;
                        goto error;
                }
        }

        chunk_status = lttng_trace_chunk_get_creation_timestamp(chunk,
                        &creation_timestamp);
        if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                ret = -LTTNG_ERR_FATAL;
                goto error;
        }
        msg.u.create_trace_chunk.creation_timestamp =
                        (uint64_t) creation_timestamp;
        /* Only used for logging purposes. */
        ret = time_to_iso8601_str(creation_timestamp,
                        creation_timestamp_buffer,
                        sizeof(creation_timestamp_buffer));
        creation_timestamp_str = !ret ? creation_timestamp_buffer :
                        "(formatting error)";

        chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
        if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                /*
                 * Anonymous trace chunks should never be transmitted
                 * to remote peers (consumerd and relayd). They are used
                 * internally for backward-compatibility purposes.
                 */
                ret = -LTTNG_ERR_FATAL;
                goto error;
        }
        msg.u.create_trace_chunk.chunk_id = chunk_id;

        if (chunk_has_local_output) {
                chunk_status = lttng_trace_chunk_borrow_chunk_directory_handle(
                                chunk, &chunk_directory_handle);
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        ret = -LTTNG_ERR_FATAL;
                        goto error;
                }
                chunk_status = lttng_trace_chunk_get_credentials(
                                chunk, &chunk_credentials);
                if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        /*
                         * Not associating credentials to a sessiond chunk is a
                         * fatal internal error.
                         */
                        ret = -LTTNG_ERR_FATAL;
                        goto error;
                }
                tc_status = lttng_trace_chunk_create_subdirectory(
                                chunk, domain_subdir);
                if (tc_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                        PERROR("Failed to create chunk domain output directory \"%s\"",
                                domain_subdir);
                        ret = -LTTNG_ERR_FATAL;
                        goto error;
                }
                domain_handle = lttng_directory_handle_create_from_handle(
                                domain_subdir,
                                chunk_directory_handle);
                if (!domain_handle) {
                        ret = -LTTNG_ERR_FATAL;
                        goto error;
                }

                /*
                 * This will only compile on platforms that support
                 * dirfd (POSIX.2008). This is fine as the session daemon
                 * is only built for such platforms.
                 *
                 * The ownership of the chunk directory handle's is maintained
                 * by the trace chunk.
                 */
                domain_dirfd = lttng_directory_handle_get_dirfd(
                                domain_handle);
                LTTNG_ASSERT(domain_dirfd >= 0);

                msg.u.create_trace_chunk.credentials.value.uid =
                                lttng_credentials_get_uid(&chunk_credentials);
                msg.u.create_trace_chunk.credentials.value.gid =
                                lttng_credentials_get_gid(&chunk_credentials);
                msg.u.create_trace_chunk.credentials.is_set = 1;
        }

        DBG("Sending consumer create trace chunk command: relayd_id = %" PRId64
                        ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
                        ", creation_timestamp = %s",
                        relayd_id, session_id, chunk_id,
                        creation_timestamp_str);
        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        health_code_update();
        if (ret < 0) {
                ERR("Trace chunk creation error on consumer");
                ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
                goto error;
        }

        if (chunk_has_local_output) {
                DBG("Sending trace chunk domain directory fd to consumer");
                health_code_update();
                ret = consumer_send_fds(socket, &domain_dirfd, 1);
                health_code_update();
                if (ret < 0) {
                        ERR("Trace chunk creation error on consumer");
                        ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
                        goto error;
                }
        }
error:
        lttng_directory_handle_put(domain_handle);
        return ret;
}

/*
 * Ask the consumer to close a trace chunk for a given session.
 *
 * Called with the consumer socket lock held.
 */
int consumer_close_trace_chunk(struct consumer_socket *socket,
                uint64_t relayd_id, uint64_t session_id,
                struct lttng_trace_chunk *chunk,
                char *closed_trace_chunk_path)
{
        int ret;
        enum lttng_trace_chunk_status chunk_status;
        lttcomm_consumer_msg msg = {
                .cmd_type = LTTNG_CONSUMER_CLOSE_TRACE_CHUNK,
        };
        msg.u.close_trace_chunk.session_id = session_id;

        struct lttcomm_consumer_close_trace_chunk_reply reply;
        uint64_t chunk_id;
        time_t close_timestamp;
        enum lttng_trace_chunk_command_type close_command;
        const char *close_command_name = "none";
        struct lttng_dynamic_buffer path_reception_buffer;

        LTTNG_ASSERT(socket);
        lttng_dynamic_buffer_init(&path_reception_buffer);

        if (relayd_id != -1ULL) {
                LTTNG_OPTIONAL_SET(
                                &msg.u.close_trace_chunk.relayd_id, relayd_id);
        }

        chunk_status = lttng_trace_chunk_get_close_command(
                        chunk, &close_command);
        switch (chunk_status) {
        case LTTNG_TRACE_CHUNK_STATUS_OK:
                LTTNG_OPTIONAL_SET(&msg.u.close_trace_chunk.close_command,
                                (uint32_t) close_command);
                break;
        case LTTNG_TRACE_CHUNK_STATUS_NONE:
                break;
        default:
                ERR("Failed to get trace chunk close command");
                ret = -1;
                goto error;
        }

        chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
        /*
         * Anonymous trace chunks should never be transmitted to remote peers
         * (consumerd and relayd). They are used internally for
         * backward-compatibility purposes.
         */
        LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
        msg.u.close_trace_chunk.chunk_id = chunk_id;

        chunk_status = lttng_trace_chunk_get_close_timestamp(chunk,
                        &close_timestamp);
        /*
         * A trace chunk should be closed locally before being closed remotely.
         * Otherwise, the close timestamp would never be transmitted to the
         * peers.
         */
        LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
        msg.u.close_trace_chunk.close_timestamp = (uint64_t) close_timestamp;

        if (msg.u.close_trace_chunk.close_command.is_set) {
                close_command_name = lttng_trace_chunk_command_type_get_name(
                                close_command);
        }
        DBG("Sending consumer close trace chunk command: relayd_id = %" PRId64
                        ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
                        ", close command = \"%s\"",
                        relayd_id, session_id, chunk_id, close_command_name);

        health_code_update();
        ret = consumer_socket_send(socket, &msg, sizeof(struct lttcomm_consumer_msg));
        if (ret < 0) {
                ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
                goto error;
        }
        ret = consumer_socket_recv(socket, &reply, sizeof(reply));
        if (ret < 0) {
                ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
                goto error;
        }
        if (reply.path_length >= LTTNG_PATH_MAX) {
                ERR("Invalid path returned by relay daemon: %" PRIu32 "bytes exceeds maximal allowed length of %d bytes",
                                reply.path_length, LTTNG_PATH_MAX);
                ret = -LTTNG_ERR_INVALID_PROTOCOL;
                goto error;
        }
        ret = lttng_dynamic_buffer_set_size(&path_reception_buffer,
                        reply.path_length);
        if (ret) {
                ERR("Failed to allocate reception buffer of path returned by the \"close trace chunk\" command");
                ret = -LTTNG_ERR_NOMEM;
                goto error;
        }
        ret = consumer_socket_recv(socket, path_reception_buffer.data,
                        path_reception_buffer.size);
        if (ret < 0) {
                ERR("Communication error while receiving path of closed trace chunk");
                ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
                goto error;
        }
        if (path_reception_buffer.data[path_reception_buffer.size - 1] != '\0') {
                ERR("Invalid path returned by relay daemon: not null-terminated");
                ret = -LTTNG_ERR_INVALID_PROTOCOL;
                goto error;
        }
        if (closed_trace_chunk_path) {
                /*
                 * closed_trace_chunk_path is assumed to have a length >=
                 * LTTNG_PATH_MAX
                 */
                memcpy(closed_trace_chunk_path, path_reception_buffer.data,
                                path_reception_buffer.size);
        }
error:
        lttng_dynamic_buffer_reset(&path_reception_buffer);
        health_code_update();
        return ret;
}

/*
 * Ask the consumer if a trace chunk exists.
 *
 * Called with the consumer socket lock held.
 * Returns 0 on success, or a negative value on error.
 */
int consumer_trace_chunk_exists(struct consumer_socket *socket,
                uint64_t relayd_id, uint64_t session_id,
                struct lttng_trace_chunk *chunk,
                enum consumer_trace_chunk_exists_status *result)
{
        int ret;
        enum lttng_trace_chunk_status chunk_status;
        lttcomm_consumer_msg msg = {
                .cmd_type = LTTNG_CONSUMER_TRACE_CHUNK_EXISTS,
        };
        msg.u.trace_chunk_exists.session_id = session_id;

        uint64_t chunk_id;
        const char *consumer_reply_str;

        LTTNG_ASSERT(socket);

        if (relayd_id != -1ULL) {
                LTTNG_OPTIONAL_SET(&msg.u.trace_chunk_exists.relayd_id,
                                relayd_id);
        }

        chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
        if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
                /*
                 * Anonymous trace chunks should never be transmitted
                 * to remote peers (consumerd and relayd). They are used
                 * internally for backward-compatibility purposes.
                 */
                ret = -LTTNG_ERR_FATAL;
                goto error;
        }
        msg.u.trace_chunk_exists.chunk_id = chunk_id;

        DBG("Sending consumer trace chunk exists command: relayd_id = %" PRId64
                        ", session_id = %" PRIu64
                        ", chunk_id = %" PRIu64, relayd_id, session_id, chunk_id);

        health_code_update();
        ret = consumer_send_msg(socket, &msg);
        switch (-ret) {
        case LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK:
                consumer_reply_str = "unknown trace chunk";
                *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK;
                break;
        case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL:
                consumer_reply_str = "trace chunk exists locally";
                *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_LOCAL;
                break;
        case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE:
                consumer_reply_str = "trace chunk exists on remote peer";
                *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_REMOTE;
                break;
        default:
                ERR("Consumer returned an error from TRACE_CHUNK_EXISTS command");
                ret = -1;
                goto error;
        }
        DBG("Consumer reply to TRACE_CHUNK_EXISTS command: %s",
                        consumer_reply_str);
        ret = 0;
error:
        health_code_update();
        return ret;
}