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

/*
 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License, version 2 only,
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along
 * with this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
 */

#define _GNU_SOURCE
#define _LGPL_SOURCE
#include <errno.h>
#include <inttypes.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <urcu/compiler.h>
#include <lttng/ust-error.h>
#include <signal.h>

#include <common/common.h>
#include <common/sessiond-comm/sessiond-comm.h>

#include "buffer-registry.h"
#include "fd-limit.h"
#include "health-sessiond.h"
#include "ust-app.h"
#include "ust-consumer.h"
#include "ust-ctl.h"
#include "utils.h"

static
int ust_app_flush_app_session(struct ust_app *app, struct ust_app_session *ua_sess);

/* Next available channel key. Access under next_channel_key_lock. */
static uint64_t _next_channel_key;
static pthread_mutex_t next_channel_key_lock = PTHREAD_MUTEX_INITIALIZER;

/* Next available session ID. Access under next_session_id_lock. */
static uint64_t _next_session_id;
static pthread_mutex_t next_session_id_lock = PTHREAD_MUTEX_INITIALIZER;

/*
 * Return the incremented value of next_channel_key.
 */
static uint64_t get_next_channel_key(void)
{
	uint64_t ret;

	pthread_mutex_lock(&next_channel_key_lock);
	ret = ++_next_channel_key;
	pthread_mutex_unlock(&next_channel_key_lock);
	return ret;
}

/*
 * Return the atomically incremented value of next_session_id.
 */
static uint64_t get_next_session_id(void)
{
	uint64_t ret;

	pthread_mutex_lock(&next_session_id_lock);
	ret = ++_next_session_id;
	pthread_mutex_unlock(&next_session_id_lock);
	return ret;
}

static void copy_channel_attr_to_ustctl(
		struct ustctl_consumer_channel_attr *attr,
		struct lttng_ust_channel_attr *uattr)
{
	/* Copy event attributes since the layout is different. */
	attr->subbuf_size = uattr->subbuf_size;
	attr->num_subbuf = uattr->num_subbuf;
	attr->overwrite = uattr->overwrite;
	attr->switch_timer_interval = uattr->switch_timer_interval;
	attr->read_timer_interval = uattr->read_timer_interval;
	attr->output = uattr->output;
}

/*
 * Match function for the hash table lookup.
 *
 * It matches an ust app event based on three attributes which are the event
 * name, the filter bytecode and the loglevel.
 */
static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key)
{
	struct ust_app_event *event;
	const struct ust_app_ht_key *key;

	assert(node);
	assert(_key);

	event = caa_container_of(node, struct ust_app_event, node.node);
	key = _key;

	/* Match the 4 elements of the key: name, filter, loglevel, exclusions */

	/* Event name */
	if (strncmp(event->attr.name, key->name, sizeof(event->attr.name)) != 0) {
		goto no_match;
	}

	/* Event loglevel. */
	if (event->attr.loglevel != key->loglevel) {
		if (event->attr.loglevel_type == LTTNG_UST_LOGLEVEL_ALL
				&& key->loglevel == 0 && event->attr.loglevel == -1) {
			/*
			 * Match is accepted. This is because on event creation, the
			 * loglevel is set to -1 if the event loglevel type is ALL so 0 and
			 * -1 are accepted for this loglevel type since 0 is the one set by
			 * the API when receiving an enable event.
			 */
		} else {
			goto no_match;
		}
	}

	/* One of the filters is NULL, fail. */
	if ((key->filter && !event->filter) || (!key->filter && event->filter)) {
		goto no_match;
	}

	if (key->filter && event->filter) {
		/* Both filters exists, check length followed by the bytecode. */
		if (event->filter->len != key->filter->len ||
				memcmp(event->filter->data, key->filter->data,
					event->filter->len) != 0) {
			goto no_match;
		}
	}

	/* One of the exclusions is NULL, fail. */
	if ((key->exclusion && !event->exclusion) || (!key->exclusion && event->exclusion)) {
		goto no_match;
	}

	if (key->exclusion && event->exclusion) {
		/* Both exclusions exists, check count followed by the names. */
		if (event->exclusion->count != key->exclusion->count ||
				memcmp(event->exclusion->names, key->exclusion->names,
					event->exclusion->count * LTTNG_UST_SYM_NAME_LEN) != 0) {
			goto no_match;
		}
	}


	/* Match. */
	return 1;

no_match:
	return 0;
}

/*
 * Unique add of an ust app event in the given ht. This uses the custom
 * ht_match_ust_app_event match function and the event name as hash.
 */
static void add_unique_ust_app_event(struct ust_app_channel *ua_chan,
		struct ust_app_event *event)
{
	struct cds_lfht_node *node_ptr;
	struct ust_app_ht_key key;
	struct lttng_ht *ht;

	assert(ua_chan);
	assert(ua_chan->events);
	assert(event);

	ht = ua_chan->events;
	key.name = event->attr.name;
	key.filter = event->filter;
	key.loglevel = event->attr.loglevel;
	key.exclusion = event->exclusion;

	node_ptr = cds_lfht_add_unique(ht->ht,
			ht->hash_fct(event->node.key, lttng_ht_seed),
			ht_match_ust_app_event, &key, &event->node.node);
	assert(node_ptr == &event->node.node);
}

/*
 * Close the notify socket from the given RCU head object. This MUST be called
 * through a call_rcu().
 */
static void close_notify_sock_rcu(struct rcu_head *head)
{
	int ret;
	struct ust_app_notify_sock_obj *obj =
		caa_container_of(head, struct ust_app_notify_sock_obj, head);

	/* Must have a valid fd here. */
	assert(obj->fd >= 0);

	ret = close(obj->fd);
	if (ret) {
		ERR("close notify sock %d RCU", obj->fd);
	}
	lttng_fd_put(LTTNG_FD_APPS, 1);

	free(obj);
}

/*
 * Return the session registry according to the buffer type of the given
 * session.
 *
 * A registry per UID object MUST exists before calling this function or else
 * it assert() if not found. RCU read side lock must be acquired.
 */
static struct ust_registry_session *get_session_registry(
		struct ust_app_session *ua_sess)
{
	struct ust_registry_session *registry = NULL;

	assert(ua_sess);

	switch (ua_sess->buffer_type) {
	case LTTNG_BUFFER_PER_PID:
	{
		struct buffer_reg_pid *reg_pid = buffer_reg_pid_find(ua_sess->id);
		if (!reg_pid) {
			goto error;
		}
		registry = reg_pid->registry->reg.ust;
		break;
	}
	case LTTNG_BUFFER_PER_UID:
	{
		struct buffer_reg_uid *reg_uid = buffer_reg_uid_find(
				ua_sess->tracing_id, ua_sess->bits_per_long, ua_sess->uid);
		if (!reg_uid) {
			goto error;
		}
		registry = reg_uid->registry->reg.ust;
		break;
	}
	default:
		assert(0);
	};

error:
	return registry;
}

/*
 * Delete ust context safely. RCU read lock must be held before calling
 * this function.
 */
static
void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx)
{
	int ret;

	assert(ua_ctx);

	if (ua_ctx->obj) {
		ret = ustctl_release_object(sock, ua_ctx->obj);
		if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app sock %d release ctx obj handle %d failed with ret %d",
					sock, ua_ctx->obj->handle, ret);
		}
		free(ua_ctx->obj);
	}
	free(ua_ctx);
}

/*
 * Delete ust app event safely. RCU read lock must be held before calling
 * this function.
 */
static
void delete_ust_app_event(int sock, struct ust_app_event *ua_event)
{
	int ret;

	assert(ua_event);

	free(ua_event->filter);
	if (ua_event->exclusion != NULL)
		free(ua_event->exclusion);
	if (ua_event->obj != NULL) {
		ret = ustctl_release_object(sock, ua_event->obj);
		if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app sock %d release event obj failed with ret %d",
					sock, ret);
		}
		free(ua_event->obj);
	}
	free(ua_event);
}

/*
 * Release ust data object of the given stream.
 *
 * Return 0 on success or else a negative value.
 */
static int release_ust_app_stream(int sock, struct ust_app_stream *stream)
{
	int ret = 0;

	assert(stream);

	if (stream->obj) {
		ret = ustctl_release_object(sock, stream->obj);
		if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app sock %d release stream obj failed with ret %d",
					sock, ret);
		}
		lttng_fd_put(LTTNG_FD_APPS, 2);
		free(stream->obj);
	}

	return ret;
}

/*
 * Delete ust app stream safely. RCU read lock must be held before calling
 * this function.
 */
static
void delete_ust_app_stream(int sock, struct ust_app_stream *stream)
{
	assert(stream);

	(void) release_ust_app_stream(sock, stream);
	free(stream);
}

/*
 * We need to execute ht_destroy outside of RCU read-side critical
 * section and outside of call_rcu thread, so we postpone its execution
 * using ht_cleanup_push. It is simpler than to change the semantic of
 * the many callers of delete_ust_app_session().
 */
static
void delete_ust_app_channel_rcu(struct rcu_head *head)
{
	struct ust_app_channel *ua_chan =
		caa_container_of(head, struct ust_app_channel, rcu_head);

	ht_cleanup_push(ua_chan->ctx);
	ht_cleanup_push(ua_chan->events);
	free(ua_chan);
}

/*
 * Delete ust app channel safely. RCU read lock must be held before calling
 * this function.
 */
static
void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan,
		struct ust_app *app)
{
	int ret;
	struct lttng_ht_iter iter;
	struct ust_app_event *ua_event;
	struct ust_app_ctx *ua_ctx;
	struct ust_app_stream *stream, *stmp;
	struct ust_registry_session *registry;

	assert(ua_chan);

	DBG3("UST app deleting channel %s", ua_chan->name);

	/* Wipe stream */
	cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) {
		cds_list_del(&stream->list);
		delete_ust_app_stream(sock, stream);
	}

	/* Wipe context */
	cds_lfht_for_each_entry(ua_chan->ctx->ht, &iter.iter, ua_ctx, node.node) {
		cds_list_del(&ua_ctx->list);
		ret = lttng_ht_del(ua_chan->ctx, &iter);
		assert(!ret);
		delete_ust_app_ctx(sock, ua_ctx);
	}

	/* Wipe events */
	cds_lfht_for_each_entry(ua_chan->events->ht, &iter.iter, ua_event,
			node.node) {
		ret = lttng_ht_del(ua_chan->events, &iter);
		assert(!ret);
		delete_ust_app_event(sock, ua_event);
	}

	if (ua_chan->session->buffer_type == LTTNG_BUFFER_PER_PID) {
		/* Wipe and free registry from session registry. */
		registry = get_session_registry(ua_chan->session);
		if (registry) {
			ust_registry_channel_del_free(registry, ua_chan->key);
		}
	}

	if (ua_chan->obj != NULL) {
		/* Remove channel from application UST object descriptor. */
		iter.iter.node = &ua_chan->ust_objd_node.node;
		ret = lttng_ht_del(app->ust_objd, &iter);
		assert(!ret);
		ret = ustctl_release_object(sock, ua_chan->obj);
		if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app sock %d release channel obj failed with ret %d",
					sock, ret);
		}
		lttng_fd_put(LTTNG_FD_APPS, 1);
		free(ua_chan->obj);
	}
	call_rcu(&ua_chan->rcu_head, delete_ust_app_channel_rcu);
}

/*
 * Push metadata to consumer socket.
 *
 * RCU read-side lock must be held to guarantee existance of socket.
 * Must be called with the ust app session lock held.
 * Must be called with the registry lock held.
 *
 * On success, return the len of metadata pushed or else a negative value.
 * Returning a -EPIPE return value means we could not send the metadata,
 * but it can be caused by recoverable errors (e.g. the application has
 * terminated concurrently).
 */
ssize_t ust_app_push_metadata(struct ust_registry_session *registry,
		struct consumer_socket *socket, int send_zero_data)
{
	int ret;
	char *metadata_str = NULL;
	size_t len, offset, new_metadata_len_sent;
	ssize_t ret_val;
	uint64_t metadata_key;

	assert(registry);
	assert(socket);

	metadata_key = registry->metadata_key;

	/*
	 * Means that no metadata was assigned to the session. This can
	 * happens if no start has been done previously.
	 */
	if (!metadata_key) {
		return 0;
	}

	/*
	 * On a push metadata error either the consumer is dead or the
	 * metadata channel has been destroyed because its endpoint
	 * might have died (e.g: relayd), or because the application has
	 * exited. If so, the metadata closed flag is set to 1 so we
	 * deny pushing metadata again which is not valid anymore on the
	 * consumer side.
	 */
	if (registry->metadata_closed) {
		return -EPIPE;
	}

	offset = registry->metadata_len_sent;
	len = registry->metadata_len - registry->metadata_len_sent;
	new_metadata_len_sent = registry->metadata_len;
	if (len == 0) {
		DBG3("No metadata to push for metadata key %" PRIu64,
				registry->metadata_key);
		ret_val = len;
		if (send_zero_data) {
			DBG("No metadata to push");
			goto push_data;
		}
		goto end;
	}

	/* Allocate only what we have to send. */
	metadata_str = zmalloc(len);
	if (!metadata_str) {
		PERROR("zmalloc ust app metadata string");
		ret_val = -ENOMEM;
		goto error;
	}
	/* Copy what we haven't sent out. */
	memcpy(metadata_str, registry->metadata + offset, len);

push_data:
	pthread_mutex_unlock(&registry->lock);
	/*
	 * We need to unlock the registry while we push metadata to
	 * break a circular dependency between the consumerd metadata
	 * lock and the sessiond registry lock. Indeed, pushing metadata
	 * to the consumerd awaits that it gets pushed all the way to
	 * relayd, but doing so requires grabbing the metadata lock. If
	 * a concurrent metadata request is being performed by
	 * consumerd, this can try to grab the registry lock on the
	 * sessiond while holding the metadata lock on the consumer
	 * daemon. Those push and pull schemes are performed on two
	 * different bidirectionnal communication sockets.
	 */
	ret = consumer_push_metadata(socket, metadata_key,
			metadata_str, len, offset);
	pthread_mutex_lock(&registry->lock);
	if (ret < 0) {
		/*
		 * There is an acceptable race here between the registry
		 * metadata key assignment and the creation on the
		 * consumer. The session daemon can concurrently push
		 * metadata for this registry while being created on the
		 * consumer since the metadata key of the registry is
		 * assigned *before* it is setup to avoid the consumer
		 * to ask for metadata that could possibly be not found
		 * in the session daemon.
		 *
		 * The metadata will get pushed either by the session
		 * being stopped or the consumer requesting metadata if
		 * that race is triggered.
		 */
		if (ret == -LTTCOMM_CONSUMERD_CHANNEL_FAIL) {
			ret = 0;
		} else {
			ERR("Error pushing metadata to consumer");
		}
		ret_val = ret;
		goto error_push;
	} else {
		/*
		 * Metadata may have been concurrently pushed, since
		 * we're not holding the registry lock while pushing to
		 * consumer.  This is handled by the fact that we send
		 * the metadata content, size, and the offset at which
		 * that metadata belongs. This may arrive out of order
		 * on the consumer side, and the consumer is able to
		 * deal with overlapping fragments. The consumer
		 * supports overlapping fragments, which must be
		 * contiguous starting from offset 0. We keep the
		 * largest metadata_len_sent value of the concurrent
		 * send.
		 */
		registry->metadata_len_sent =
			max_t(size_t, registry->metadata_len_sent,
				new_metadata_len_sent);
	}
	free(metadata_str);
	return len;

end:
error:
	if (ret_val) {
		/*
		 * On error, flag the registry that the metadata is
		 * closed. We were unable to push anything and this
		 * means that either the consumer is not responding or
		 * the metadata cache has been destroyed on the
		 * consumer.
		 */
		registry->metadata_closed = 1;
	}
error_push:
	free(metadata_str);
	return ret_val;
}

/*
 * For a given application and session, push metadata to consumer.
 * Either sock or consumer is required : if sock is NULL, the default
 * socket to send the metadata is retrieved from consumer, if sock
 * is not NULL we use it to send the metadata.
 * RCU read-side lock must be held while calling this function,
 * therefore ensuring existance of registry. It also ensures existance
 * of socket throughout this function.
 *
 * Return 0 on success else a negative error.
 * Returning a -EPIPE return value means we could not send the metadata,
 * but it can be caused by recoverable errors (e.g. the application has
 * terminated concurrently).
 */
static int push_metadata(struct ust_registry_session *registry,
		struct consumer_output *consumer)
{
	int ret_val;
	ssize_t ret;
	struct consumer_socket *socket;

	assert(registry);
	assert(consumer);

	pthread_mutex_lock(&registry->lock);
	if (registry->metadata_closed) {
		ret_val = -EPIPE;
		goto error;
	}

	/* Get consumer socket to use to push the metadata.*/
	socket = consumer_find_socket_by_bitness(registry->bits_per_long,
			consumer);
	if (!socket) {
		ret_val = -1;
		goto error;
	}

	ret = ust_app_push_metadata(registry, socket, 0);
	if (ret < 0) {
		ret_val = ret;
		goto error;
	}
	pthread_mutex_unlock(&registry->lock);
	return 0;

error:
	pthread_mutex_unlock(&registry->lock);
	return ret_val;
}

/*
 * Send to the consumer a close metadata command for the given session. Once
 * done, the metadata channel is deleted and the session metadata pointer is
 * nullified. The session lock MUST be held unless the application is
 * in the destroy path.
 *
 * Return 0 on success else a negative value.
 */
static int close_metadata(struct ust_registry_session *registry,
		struct consumer_output *consumer)
{
	int ret;
	struct consumer_socket *socket;

	assert(registry);
	assert(consumer);

	rcu_read_lock();

	pthread_mutex_lock(&registry->lock);

	if (!registry->metadata_key || registry->metadata_closed) {
		ret = 0;
		goto end;
	}

	/* Get consumer socket to use to push the metadata.*/
	socket = consumer_find_socket_by_bitness(registry->bits_per_long,
			consumer);
	if (!socket) {
		ret = -1;
		goto error;
	}

	ret = consumer_close_metadata(socket, registry->metadata_key);
	if (ret < 0) {
		goto error;
	}

error:
	/*
	 * Metadata closed. Even on error this means that the consumer is not
	 * responding or not found so either way a second close should NOT be emit
	 * for this registry.
	 */
	registry->metadata_closed = 1;
end:
	pthread_mutex_unlock(&registry->lock);
	rcu_read_unlock();
	return ret;
}

/*
 * We need to execute ht_destroy outside of RCU read-side critical
 * section and outside of call_rcu thread, so we postpone its execution
 * using ht_cleanup_push. It is simpler than to change the semantic of
 * the many callers of delete_ust_app_session().
 */
static
void delete_ust_app_session_rcu(struct rcu_head *head)
{
	struct ust_app_session *ua_sess =
		caa_container_of(head, struct ust_app_session, rcu_head);

	ht_cleanup_push(ua_sess->channels);
	free(ua_sess);
}

/*
 * Delete ust app session safely. RCU read lock must be held before calling
 * this function.
 */
static
void delete_ust_app_session(int sock, struct ust_app_session *ua_sess,
		struct ust_app *app)
{
	int ret;
	struct lttng_ht_iter iter;
	struct ust_app_channel *ua_chan;
	struct ust_registry_session *registry;

	assert(ua_sess);

	pthread_mutex_lock(&ua_sess->lock);

	assert(!ua_sess->deleted);
	ua_sess->deleted = true;

	registry = get_session_registry(ua_sess);
	if (registry) {
		/* Push metadata for application before freeing the application. */
		(void) push_metadata(registry, ua_sess->consumer);

		/*
		 * Don't ask to close metadata for global per UID buffers. Close
		 * metadata only on destroy trace session in this case. Also, the
		 * previous push metadata could have flag the metadata registry to
		 * close so don't send a close command if closed.
		 */
		if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) {
			/* And ask to close it for this session registry. */
			(void) close_metadata(registry, ua_sess->consumer);
		}
	}

	cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan,
			node.node) {
		ret = lttng_ht_del(ua_sess->channels, &iter);
		assert(!ret);
		delete_ust_app_channel(sock, ua_chan, app);
	}

	/* In case of per PID, the registry is kept in the session. */
	if (ua_sess->buffer_type == LTTNG_BUFFER_PER_PID) {
		struct buffer_reg_pid *reg_pid = buffer_reg_pid_find(ua_sess->id);
		if (reg_pid) {
			buffer_reg_pid_remove(reg_pid);
			buffer_reg_pid_destroy(reg_pid);
		}
	}

	if (ua_sess->handle != -1) {
		ret = ustctl_release_handle(sock, ua_sess->handle);
		if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app sock %d release session handle failed with ret %d",
					sock, ret);
		}
	}
	pthread_mutex_unlock(&ua_sess->lock);

	consumer_output_put(ua_sess->consumer);

	call_rcu(&ua_sess->rcu_head, delete_ust_app_session_rcu);
}

/*
 * Delete a traceable application structure from the global list. Never call
 * this function outside of a call_rcu call.
 *
 * RCU read side lock should _NOT_ be held when calling this function.
 */
static
void delete_ust_app(struct ust_app *app)
{
	int ret, sock;
	struct ust_app_session *ua_sess, *tmp_ua_sess;

	/* Delete ust app sessions info */
	sock = app->sock;
	app->sock = -1;

	/* Wipe sessions */
	cds_list_for_each_entry_safe(ua_sess, tmp_ua_sess, &app->teardown_head,
			teardown_node) {
		/* Free every object in the session and the session. */
		rcu_read_lock();
		delete_ust_app_session(sock, ua_sess, app);
		rcu_read_unlock();
	}

	ht_cleanup_push(app->sessions);
	ht_cleanup_push(app->ust_objd);

	/*
	 * Wait until we have deleted the application from the sock hash table
	 * before closing this socket, otherwise an application could re-use the
	 * socket ID and race with the teardown, using the same hash table entry.
	 *
	 * It's OK to leave the close in call_rcu. We want it to stay unique for
	 * all RCU readers that could run concurrently with unregister app,
	 * therefore we _need_ to only close that socket after a grace period. So
	 * it should stay in this RCU callback.
	 *
	 * This close() is a very important step of the synchronization model so
	 * every modification to this function must be carefully reviewed.
	 */
	ret = close(sock);
	if (ret) {
		PERROR("close");
	}
	lttng_fd_put(LTTNG_FD_APPS, 1);

	DBG2("UST app pid %d deleted", app->pid);
	free(app);
}

/*
 * URCU intermediate call to delete an UST app.
 */
static
void delete_ust_app_rcu(struct rcu_head *head)
{
	struct lttng_ht_node_ulong *node =
		caa_container_of(head, struct lttng_ht_node_ulong, head);
	struct ust_app *app =
		caa_container_of(node, struct ust_app, pid_n);

	DBG3("Call RCU deleting app PID %d", app->pid);
	delete_ust_app(app);
}

/*
 * Delete the session from the application ht and delete the data structure by
 * freeing every object inside and releasing them.
 */
static void destroy_app_session(struct ust_app *app,
		struct ust_app_session *ua_sess)
{
	int ret;
	struct lttng_ht_iter iter;

	assert(app);
	assert(ua_sess);

	iter.iter.node = &ua_sess->node.node;
	ret = lttng_ht_del(app->sessions, &iter);
	if (ret) {
		/* Already scheduled for teardown. */
		goto end;
	}

	/* Once deleted, free the data structure. */
	delete_ust_app_session(app->sock, ua_sess, app);

end:
	return;
}

/*
 * Alloc new UST app session.
 */
static
struct ust_app_session *alloc_ust_app_session(struct ust_app *app)
{
	struct ust_app_session *ua_sess;

	/* Init most of the default value by allocating and zeroing */
	ua_sess = zmalloc(sizeof(struct ust_app_session));
	if (ua_sess == NULL) {
		PERROR("malloc");
		goto error_free;
	}

	ua_sess->handle = -1;
	ua_sess->channels = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
	ua_sess->metadata_attr.type = LTTNG_UST_CHAN_METADATA;
	pthread_mutex_init(&ua_sess->lock, NULL);

	return ua_sess;

error_free:
	return NULL;
}

/*
 * Alloc new UST app channel.
 */
static
struct ust_app_channel *alloc_ust_app_channel(char *name,
		struct ust_app_session *ua_sess,
		struct lttng_ust_channel_attr *attr)
{
	struct ust_app_channel *ua_chan;

	/* Init most of the default value by allocating and zeroing */
	ua_chan = zmalloc(sizeof(struct ust_app_channel));
	if (ua_chan == NULL) {
		PERROR("malloc");
		goto error;
	}

	/* Setup channel name */
	strncpy(ua_chan->name, name, sizeof(ua_chan->name));
	ua_chan->name[sizeof(ua_chan->name) - 1] = '\0';

	ua_chan->enabled = 1;
	ua_chan->handle = -1;
	ua_chan->session = ua_sess;
	ua_chan->key = get_next_channel_key();
	ua_chan->ctx = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
	ua_chan->events = lttng_ht_new(0, LTTNG_HT_TYPE_STRING);
	lttng_ht_node_init_str(&ua_chan->node, ua_chan->name);

	CDS_INIT_LIST_HEAD(&ua_chan->streams.head);
	CDS_INIT_LIST_HEAD(&ua_chan->ctx_list);

	/* Copy attributes */
	if (attr) {
		/* Translate from lttng_ust_channel to ustctl_consumer_channel_attr. */
		ua_chan->attr.subbuf_size = attr->subbuf_size;
		ua_chan->attr.num_subbuf = attr->num_subbuf;
		ua_chan->attr.overwrite = attr->overwrite;
		ua_chan->attr.switch_timer_interval = attr->switch_timer_interval;
		ua_chan->attr.read_timer_interval = attr->read_timer_interval;
		ua_chan->attr.output = attr->output;
	}
	/* By default, the channel is a per cpu channel. */
	ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU;

	DBG3("UST app channel %s allocated", ua_chan->name);

	return ua_chan;

error:
	return NULL;
}

/*
 * Allocate and initialize a UST app stream.
 *
 * Return newly allocated stream pointer or NULL on error.
 */
struct ust_app_stream *ust_app_alloc_stream(void)
{
	struct ust_app_stream *stream = NULL;

	stream = zmalloc(sizeof(*stream));
	if (stream == NULL) {
		PERROR("zmalloc ust app stream");
		goto error;
	}

	/* Zero could be a valid value for a handle so flag it to -1. */
	stream->handle = -1;

error:
	return stream;
}

/*
 * Alloc new UST app event.
 */
static
struct ust_app_event *alloc_ust_app_event(char *name,
		struct lttng_ust_event *attr)
{
	struct ust_app_event *ua_event;

	/* Init most of the default value by allocating and zeroing */
	ua_event = zmalloc(sizeof(struct ust_app_event));
	if (ua_event == NULL) {
		PERROR("malloc");
		goto error;
	}

	ua_event->enabled = 1;
	strncpy(ua_event->name, name, sizeof(ua_event->name));
	ua_event->name[sizeof(ua_event->name) - 1] = '\0';
	lttng_ht_node_init_str(&ua_event->node, ua_event->name);

	/* Copy attributes */
	if (attr) {
		memcpy(&ua_event->attr, attr, sizeof(ua_event->attr));
	}

	DBG3("UST app event %s allocated", ua_event->name);

	return ua_event;

error:
	return NULL;
}

/*
 * Alloc new UST app context.
 */
static
struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context *uctx)
{
	struct ust_app_ctx *ua_ctx;

	ua_ctx = zmalloc(sizeof(struct ust_app_ctx));
	if (ua_ctx == NULL) {
		goto error;
	}

	CDS_INIT_LIST_HEAD(&ua_ctx->list);

	if (uctx) {
		memcpy(&ua_ctx->ctx, uctx, sizeof(ua_ctx->ctx));
	}

	DBG3("UST app context %d allocated", ua_ctx->ctx.ctx);

error:
	return ua_ctx;
}

/*
 * Allocate a filter and copy the given original filter.
 *
 * Return allocated filter or NULL on error.
 */
static struct lttng_filter_bytecode *copy_filter_bytecode(
		struct lttng_filter_bytecode *orig_f)
{
	struct lttng_filter_bytecode *filter = NULL;

	/* Copy filter bytecode */
	filter = zmalloc(sizeof(*filter) + orig_f->len);
	if (!filter) {
		PERROR("zmalloc alloc filter bytecode");
		goto error;
	}

	memcpy(filter, orig_f, sizeof(*filter) + orig_f->len);

error:
	return filter;
}

/*
 * Create a liblttng-ust filter bytecode from given bytecode.
 *
 * Return allocated filter or NULL on error.
 */
static struct lttng_ust_filter_bytecode *create_ust_bytecode_from_bytecode(
		struct lttng_filter_bytecode *orig_f)
{
	struct lttng_ust_filter_bytecode *filter = NULL;

	/* Copy filter bytecode */
	filter = zmalloc(sizeof(*filter) + orig_f->len);
	if (!filter) {
		PERROR("zmalloc alloc ust filter bytecode");
		goto error;
	}

	assert(sizeof(struct lttng_filter_bytecode) ==
			sizeof(struct lttng_ust_filter_bytecode));
	memcpy(filter, orig_f, sizeof(*filter) + orig_f->len);
error:
	return filter;
}

/*
 * Find an ust_app using the sock and return it. RCU read side lock must be
 * held before calling this helper function.
 */
struct ust_app *ust_app_find_by_sock(int sock)
{
	struct lttng_ht_node_ulong *node;
	struct lttng_ht_iter iter;

	lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &iter);
	node = lttng_ht_iter_get_node_ulong(&iter);
	if (node == NULL) {
		DBG2("UST app find by sock %d not found", sock);
		goto error;
	}

	return caa_container_of(node, struct ust_app, sock_n);

error:
	return NULL;
}

/*
 * Find an ust_app using the notify sock and return it. RCU read side lock must
 * be held before calling this helper function.
 */
static struct ust_app *find_app_by_notify_sock(int sock)
{
	struct lttng_ht_node_ulong *node;
	struct lttng_ht_iter iter;

	lttng_ht_lookup(ust_app_ht_by_notify_sock, (void *)((unsigned long) sock),
			&iter);
	node = lttng_ht_iter_get_node_ulong(&iter);
	if (node == NULL) {
		DBG2("UST app find by notify sock %d not found", sock);
		goto error;
	}

	return caa_container_of(node, struct ust_app, notify_sock_n);

error:
	return NULL;
}

/*
 * Lookup for an ust app event based on event name, filter bytecode and the
 * event loglevel.
 *
 * Return an ust_app_event object or NULL on error.
 */
static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht,
		char *name, struct lttng_filter_bytecode *filter, int loglevel,
		const struct lttng_event_exclusion *exclusion)
{
	struct lttng_ht_iter iter;
	struct lttng_ht_node_str *node;
	struct ust_app_event *event = NULL;
	struct ust_app_ht_key key;

	assert(name);
	assert(ht);

	/* Setup key for event lookup. */
	key.name = name;
	key.filter = filter;
	key.loglevel = loglevel;
	/* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */
	key.exclusion = exclusion;

	/* Lookup using the event name as hash and a custom match fct. */
	cds_lfht_lookup(ht->ht, ht->hash_fct((void *) name, lttng_ht_seed),
			ht_match_ust_app_event, &key, &iter.iter);
	node = lttng_ht_iter_get_node_str(&iter);
	if (node == NULL) {
		goto end;
	}

	event = caa_container_of(node, struct ust_app_event, node);

end:
	return event;
}

/*
 * Create the channel context on the tracer.
 *
 * Called with UST app session lock held.
 */
static
int create_ust_channel_context(struct ust_app_channel *ua_chan,
		struct ust_app_ctx *ua_ctx, struct ust_app *app)
{
	int ret;

	health_code_update();

	ret = ustctl_add_context(app->sock, &ua_ctx->ctx,
			ua_chan->obj, &ua_ctx->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app create channel context failed for app (pid: %d) "
					"with ret %d", app->pid, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app disable event failed. Application is dead.");
		}
		goto error;
	}

	ua_ctx->handle = ua_ctx->obj->handle;

	DBG2("UST app context handle %d created successfully for channel %s",
			ua_ctx->handle, ua_chan->name);

error:
	health_code_update();
	return ret;
}

/*
 * Set the filter on the tracer.
 */
static
int set_ust_event_filter(struct ust_app_event *ua_event,
		struct ust_app *app)
{
	int ret;
	struct lttng_ust_filter_bytecode *ust_bytecode = NULL;

	health_code_update();

	if (!ua_event->filter) {
		ret = 0;
		goto error;
	}

	ust_bytecode = create_ust_bytecode_from_bytecode(ua_event->filter);
	if (!ust_bytecode) {
		ret = -LTTNG_ERR_NOMEM;
		goto error;
	}
	ret = ustctl_set_filter(app->sock, ust_bytecode,
			ua_event->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app event %s filter failed for app (pid: %d) "
					"with ret %d", ua_event->attr.name, app->pid, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app filter event failed. Application is dead.");
		}
		goto error;
	}

	DBG2("UST filter set successfully for event %s", ua_event->name);

error:
	health_code_update();
	free(ust_bytecode);
	return ret;
}

static
struct lttng_ust_event_exclusion *create_ust_exclusion_from_exclusion(
		struct lttng_event_exclusion *exclusion)
{
	struct lttng_ust_event_exclusion *ust_exclusion = NULL;
	size_t exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) +
		LTTNG_UST_SYM_NAME_LEN * exclusion->count;

	ust_exclusion = zmalloc(exclusion_alloc_size);
	if (!ust_exclusion) {
		PERROR("malloc");
		goto end;
	}

	assert(sizeof(struct lttng_event_exclusion) ==
			sizeof(struct lttng_ust_event_exclusion));
	memcpy(ust_exclusion, exclusion, exclusion_alloc_size);
end:
	return ust_exclusion;
}

/*
 * Set event exclusions on the tracer.
 */
static
int set_ust_event_exclusion(struct ust_app_event *ua_event,
		struct ust_app *app)
{
	int ret;
	struct lttng_ust_event_exclusion *ust_exclusion = NULL;

	health_code_update();

	if (!ua_event->exclusion || !ua_event->exclusion->count) {
		ret = 0;
		goto error;
	}

	ust_exclusion = create_ust_exclusion_from_exclusion(
			ua_event->exclusion);
	if (!ust_exclusion) {
		ret = -LTTNG_ERR_NOMEM;
		goto error;
	}
	ret = ustctl_set_exclusion(app->sock, ust_exclusion, ua_event->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app event %s exclusions failed for app (pid: %d) "
					"with ret %d", ua_event->attr.name, app->pid, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app event exclusion failed. Application is dead.");
		}
		goto error;
	}

	DBG2("UST exclusion set successfully for event %s", ua_event->name);

error:
	health_code_update();
	free(ust_exclusion);
	return ret;
}

/*
 * Disable the specified event on to UST tracer for the UST session.
 */
static int disable_ust_event(struct ust_app *app,
		struct ust_app_session *ua_sess, struct ust_app_event *ua_event)
{
	int ret;

	health_code_update();

	ret = ustctl_disable(app->sock, ua_event->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app event %s disable failed for app (pid: %d) "
					"and session handle %d with ret %d",
					ua_event->attr.name, app->pid, ua_sess->handle, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app disable event failed. Application is dead.");
		}
		goto error;
	}

	DBG2("UST app event %s disabled successfully for app (pid: %d)",
			ua_event->attr.name, app->pid);

error:
	health_code_update();
	return ret;
}

/*
 * Disable the specified channel on to UST tracer for the UST session.
 */
static int disable_ust_channel(struct ust_app *app,
		struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan)
{
	int ret;

	health_code_update();

	ret = ustctl_disable(app->sock, ua_chan->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app channel %s disable failed for app (pid: %d) "
					"and session handle %d with ret %d",
					ua_chan->name, app->pid, ua_sess->handle, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app disable channel failed. Application is dead.");
		}
		goto error;
	}

	DBG2("UST app channel %s disabled successfully for app (pid: %d)",
			ua_chan->name, app->pid);

error:
	health_code_update();
	return ret;
}

/*
 * Enable the specified channel on to UST tracer for the UST session.
 */
static int enable_ust_channel(struct ust_app *app,
		struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan)
{
	int ret;

	health_code_update();

	ret = ustctl_enable(app->sock, ua_chan->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app channel %s enable failed for app (pid: %d) "
					"and session handle %d with ret %d",
					ua_chan->name, app->pid, ua_sess->handle, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app enable channel failed. Application is dead.");
		}
		goto error;
	}

	ua_chan->enabled = 1;

	DBG2("UST app channel %s enabled successfully for app (pid: %d)",
			ua_chan->name, app->pid);

error:
	health_code_update();
	return ret;
}

/*
 * Enable the specified event on to UST tracer for the UST session.
 */
static int enable_ust_event(struct ust_app *app,
		struct ust_app_session *ua_sess, struct ust_app_event *ua_event)
{
	int ret;

	health_code_update();

	ret = ustctl_enable(app->sock, ua_event->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("UST app event %s enable failed for app (pid: %d) "
					"and session handle %d with ret %d",
					ua_event->attr.name, app->pid, ua_sess->handle, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app enable event failed. Application is dead.");
		}
		goto error;
	}

	DBG2("UST app event %s enabled successfully for app (pid: %d)",
			ua_event->attr.name, app->pid);

error:
	health_code_update();
	return ret;
}

/*
 * Send channel and stream buffer to application.
 *
 * Return 0 on success. On error, a negative value is returned.
 */
static int send_channel_pid_to_ust(struct ust_app *app,
		struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan)
{
	int ret;
	struct ust_app_stream *stream, *stmp;

	assert(app);
	assert(ua_sess);
	assert(ua_chan);

	health_code_update();

	DBG("UST app sending channel %s to UST app sock %d", ua_chan->name,
			app->sock);

	/* Send channel to the application. */
	ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan);
	if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
		ret = -ENOTCONN;	/* Caused by app exiting. */
		goto error;
	} else if (ret < 0) {
		goto error;
	}

	health_code_update();

	/* Send all streams to application. */
	cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) {
		ret = ust_consumer_send_stream_to_ust(app, ua_chan, stream);
		if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
			ret = -ENOTCONN;	/* Caused by app exiting. */
			goto error;
		} else if (ret < 0) {
			goto error;
		}
		/* We don't need the stream anymore once sent to the tracer. */
		cds_list_del(&stream->list);
		delete_ust_app_stream(-1, stream);
	}
	/* Flag the channel that it is sent to the application. */
	ua_chan->is_sent = 1;

error:
	health_code_update();
	return ret;
}

/*
 * Create the specified event onto the UST tracer for a UST session.
 *
 * Should be called with session mutex held.
 */
static
int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan, struct ust_app_event *ua_event)
{
	int ret = 0;

	health_code_update();

	/* Create UST event on tracer */
	ret = ustctl_create_event(app->sock, &ua_event->attr, ua_chan->obj,
			&ua_event->obj);
	if (ret < 0) {
		if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
			ERR("Error ustctl create event %s for app pid: %d with ret %d",
					ua_event->attr.name, app->pid, ret);
		} else {
			/*
			 * This is normal behavior, an application can die during the
			 * creation process. Don't report an error so the execution can
			 * continue normally.
			 */
			ret = 0;
			DBG3("UST app create event failed. Application is dead.");
		}
		goto error;
	}

	ua_event->handle = ua_event->obj->handle;

	DBG2("UST app event %s created successfully for pid:%d",
			ua_event->attr.name, app->pid);

	health_code_update();

	/* Set filter if one is present. */
	if (ua_event->filter) {
		ret = set_ust_event_filter(ua_event, app);
		if (ret < 0) {
			goto error;
		}
	}

	/* Set exclusions for the event */
	if (ua_event->exclusion) {
		ret = set_ust_event_exclusion(ua_event, app);
		if (ret < 0) {
			goto error;
		}
	}

	/* If event not enabled, disable it on the tracer */
	if (ua_event->enabled) {
		/*
		 * We now need to explicitly enable the event, since it
		 * is now disabled at creation.
		 */
		ret = enable_ust_event(app, ua_sess, ua_event);
		if (ret < 0) {
			/*
			 * If we hit an EPERM, something is wrong with our enable call. If
			 * we get an EEXIST, there is a problem on the tracer side since we
			 * just created it.
			 */
			switch (ret) {
			case -LTTNG_UST_ERR_PERM:
				/* Code flow problem */
				assert(0);
			case -LTTNG_UST_ERR_EXIST:
				/* It's OK for our use case. */
				ret = 0;
				break;
			default:
				break;
			}
			goto error;
		}
	}

error:
	health_code_update();
	return ret;
}

/*
 * Copy data between an UST app event and a LTT event.
 */
static void shadow_copy_event(struct ust_app_event *ua_event,
		struct ltt_ust_event *uevent)
{
	size_t exclusion_alloc_size;

	strncpy(ua_event->name, uevent->attr.name, sizeof(ua_event->name));
	ua_event->name[sizeof(ua_event->name) - 1] = '\0';

	ua_event->enabled = uevent->enabled;

	/* Copy event attributes */
	memcpy(&ua_event->attr, &uevent->attr, sizeof(ua_event->attr));

	/* Copy filter bytecode */
	if (uevent->filter) {
		ua_event->filter = copy_filter_bytecode(uevent->filter);
		/* Filter might be NULL here in case of ENONEM. */
	}

	/* Copy exclusion data */
	if (uevent->exclusion) {
		exclusion_alloc_size = sizeof(struct lttng_event_exclusion) +
				LTTNG_UST_SYM_NAME_LEN * uevent->exclusion->count;
		ua_event->exclusion = zmalloc(exclusion_alloc_size);
		if (ua_event->exclusion == NULL) {
			PERROR("malloc");
		} else {
			memcpy(ua_event->exclusion, uevent->exclusion,
					exclusion_alloc_size);
		}
	}
}

/*
 * Copy data between an UST app channel and a LTT channel.
 */
static void shadow_copy_channel(struct ust_app_channel *ua_chan,
		struct ltt_ust_channel *uchan)
{
	struct lttng_ht_iter iter;
	struct ltt_ust_event *uevent;
	struct ltt_ust_context *uctx;
	struct ust_app_event *ua_event;
	struct ust_app_ctx *ua_ctx;

	DBG2("UST app shadow copy of channel %s started", ua_chan->name);

	strncpy(ua_chan->name, uchan->name, sizeof(ua_chan->name));
	ua_chan->name[sizeof(ua_chan->name) - 1] = '\0';

	ua_chan->tracefile_size = uchan->tracefile_size;
	ua_chan->tracefile_count = uchan->tracefile_count;

	/* Copy event attributes since the layout is different. */
	ua_chan->attr.subbuf_size = uchan->attr.subbuf_size;
	ua_chan->attr.num_subbuf = uchan->attr.num_subbuf;
	ua_chan->attr.overwrite = uchan->attr.overwrite;
	ua_chan->attr.switch_timer_interval = uchan->attr.switch_timer_interval;
	ua_chan->attr.read_timer_interval = uchan->attr.read_timer_interval;
	ua_chan->attr.output = uchan->attr.output;
	/*
	 * Note that the attribute channel type is not set since the channel on the
	 * tracing registry side does not have this information.
	 */

	ua_chan->enabled = uchan->enabled;
	ua_chan->tracing_channel_id = uchan->id;

	cds_list_for_each_entry(uctx, &uchan->ctx_list, list) {
		ua_ctx = alloc_ust_app_ctx(&uctx->ctx);
		if (ua_ctx == NULL) {
			continue;
		}
		lttng_ht_node_init_ulong(&ua_ctx->node,
				(unsigned long) ua_ctx->ctx.ctx);
		lttng_ht_add_ulong(ua_chan->ctx, &ua_ctx->node);
		cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list);
	}

	/* Copy all events from ltt ust channel to ust app channel */
	cds_lfht_for_each_entry(uchan->events->ht, &iter.iter, uevent, node.node) {
		ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name,
				uevent->filter, uevent->attr.loglevel, uevent->exclusion);
		if (ua_event == NULL) {
			DBG2("UST event %s not found on shadow copy channel",
					uevent->attr.name);
			ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr);
			if (ua_event == NULL) {
				continue;
			}
			shadow_copy_event(ua_event, uevent);
			add_unique_ust_app_event(ua_chan, ua_event);
		}
	}

	DBG3("UST app shadow copy of channel %s done", ua_chan->name);
}

/*
 * Copy data between a UST app session and a regular LTT session.
 */
static void shadow_copy_session(struct ust_app_session *ua_sess,
		struct ltt_ust_session *usess, struct ust_app *app)
{
	struct lttng_ht_node_str *ua_chan_node;
	struct lttng_ht_iter iter;
	struct ltt_ust_channel *uchan;
	struct ust_app_channel *ua_chan;
	time_t rawtime;
	struct tm *timeinfo;
	char datetime[16];
	int ret;
	char tmp_shm_path[PATH_MAX];

	/* Get date and time for unique app path */
	time(&rawtime);
	timeinfo = localtime(&rawtime);
	strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", timeinfo);

	DBG2("Shadow copy of session handle %d", ua_sess->handle);

	ua_sess->tracing_id = usess->id;
	ua_sess->id = get_next_session_id();
	ua_sess->uid = app->uid;
	ua_sess->gid = app->gid;
	ua_sess->euid = usess->uid;
	ua_sess->egid = usess->gid;
	ua_sess->buffer_type = usess->buffer_type;
	ua_sess->bits_per_long = app->bits_per_long;

	/* There is only one consumer object per session possible. */
	consumer_output_get(usess->consumer);
	ua_sess->consumer = usess->consumer;

	ua_sess->output_traces = usess->output_traces;
	ua_sess->live_timer_interval = usess->live_timer_interval;
	copy_channel_attr_to_ustctl(&ua_sess->metadata_attr,
			&usess->metadata_attr);

	switch (ua_sess->buffer_type) {
	case LTTNG_BUFFER_PER_PID:
		ret = snprintf(ua_sess->path, sizeof(ua_sess->path),
				DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s", app->name, app->pid,
				datetime);
		break;
	case LTTNG_BUFFER_PER_UID:
		ret = snprintf(ua_sess->path, sizeof(ua_sess->path),
				DEFAULT_UST_TRACE_UID_PATH, ua_sess->uid, app->bits_per_long);
		break;
	default:
		assert(0);
		goto error;
	}
	if (ret < 0) {
		PERROR("asprintf UST shadow copy session");
		assert(0);
		goto error;
	}

	strncpy(ua_sess->root_shm_path, usess->root_shm_path,
		sizeof(ua_sess->root_shm_path));
	ua_sess->root_shm_path[sizeof(ua_sess->root_shm_path) - 1] = '\0';
	strncpy(ua_sess->shm_path, usess->shm_path,
		sizeof(ua_sess->shm_path));
	ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0';
	if (ua_sess->shm_path[0]) {
		switch (ua_sess->buffer_type) {
		case LTTNG_BUFFER_PER_PID:
			ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path),
					DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s",
					app->name, app->pid, datetime);
			break;
		case LTTNG_BUFFER_PER_UID:
			ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path),
					DEFAULT_UST_TRACE_UID_PATH,
					app->uid, app->bits_per_long);
			break;
		default:
			assert(0);
			goto error;
		}
		if (ret < 0) {
			PERROR("sprintf UST shadow copy session");
			assert(0);
			goto error;
		}
		strncat(ua_sess->shm_path, tmp_shm_path,
			sizeof(ua_sess->shm_path) - strlen(ua_sess->shm_path) - 1);
		ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0';
	}

	/* Iterate over all channels in global domain. */
	cds_lfht_for_each_entry(usess->domain_global.channels->ht, &iter.iter,
			uchan, node.node) {
		struct lttng_ht_iter uiter;

		lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter);
		ua_chan_node = lttng_ht_iter_get_node_str(&uiter);
		if (ua_chan_node != NULL) {
			/* Session exist. Contiuing. */
			continue;
		}

		DBG2("Channel %s not found on shadow session copy, creating it",
				uchan->name);
		ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr);
		if (ua_chan == NULL) {
			/* malloc failed FIXME: Might want to do handle ENOMEM .. */
			continue;
		}
		shadow_copy_channel(ua_chan, uchan);
		/*
		 * The concept of metadata channel does not exist on the tracing
		 * registry side of the session daemon so this can only be a per CPU
		 * channel and not metadata.
		 */
		ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU;

		lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node);
	}
	return;

error:
	consumer_output_put(ua_sess->consumer);
}

/*
 * Lookup sesison wrapper.
 */
static
void __lookup_session_by_app(struct ltt_ust_session *usess,
			struct ust_app *app, struct lttng_ht_iter *iter)
{
	/* Get right UST app session from app */
	lttng_ht_lookup(app->sessions, &usess->id, iter);
}

/*
 * Return ust app session from the app session hashtable using the UST session
 * id.
 */
static struct ust_app_session *lookup_session_by_app(
		struct ltt_ust_session *usess, struct ust_app *app)
{
	struct lttng_ht_iter iter;
	struct lttng_ht_node_u64 *node;

	__lookup_session_by_app(usess, app, &iter);
	node = lttng_ht_iter_get_node_u64(&iter);
	if (node == NULL) {
		goto error;
	}

	return caa_container_of(node, struct ust_app_session, node);

error:
	return NULL;
}

/*
 * Setup buffer registry per PID for the given session and application. If none
 * is found, a new one is created, added to the global registry and
 * initialized. If regp is valid, it's set with the newly created object.
 *
 * Return 0 on success or else a negative value.
 */
static int setup_buffer_reg_pid(struct ust_app_session *ua_sess,
		struct ust_app *app, struct buffer_reg_pid **regp)
{
	int ret = 0;
	struct buffer_reg_pid *reg_pid;

	assert(ua_sess);
	assert(app);

	rcu_read_lock();

	reg_pid = buffer_reg_pid_find(ua_sess->id);
	if (!reg_pid) {
		/*
		 * This is the create channel path meaning that if there is NO
		 * registry available, we have to create one for this session.
		 */
		ret = buffer_reg_pid_create(ua_sess->id, &reg_pid,
			ua_sess->root_shm_path, ua_sess->shm_path);
		if (ret < 0) {
			goto error;
		}
	} else {
		goto end;
	}

	/* Initialize registry. */
	ret = ust_registry_session_init(&reg_pid->registry->reg.ust, app,
			app->bits_per_long, app->uint8_t_alignment,
			app->uint16_t_alignment, app->uint32_t_alignment,
			app->uint64_t_alignment, app->long_alignment,
			app->byte_order, app->version.major,
			app->version.minor, reg_pid->root_shm_path,
			reg_pid->shm_path,
			ua_sess->euid, ua_sess->egid);
	if (ret < 0) {
		/*
		 * reg_pid->registry->reg.ust is NULL upon error, so we need to
		 * destroy the buffer registry, because it is always expected
		 * that if the buffer registry can be found, its ust registry is
		 * non-NULL.
		 */
		buffer_reg_pid_destroy(reg_pid);
		goto error;
	}

	buffer_reg_pid_add(reg_pid);

	DBG3("UST app buffer registry per PID created successfully");

end:
	if (regp) {
		*regp = reg_pid;
	}
error:
	rcu_read_unlock();
	return ret;
}

/*
 * Setup buffer registry per UID for the given session and application. If none
 * is found, a new one is created, added to the global registry and
 * initialized. If regp is valid, it's set with the newly created object.
 *
 * Return 0 on success or else a negative value.
 */
static int setup_buffer_reg_uid(struct ltt_ust_session *usess,
		struct ust_app_session *ua_sess,
		struct ust_app *app, struct buffer_reg_uid **regp)
{
	int ret = 0;
	struct buffer_reg_uid *reg_uid;

	assert(usess);
	assert(app);

	rcu_read_lock();

	reg_uid = buffer_reg_uid_find(usess->id, app->bits_per_long, app->uid);
	if (!reg_uid) {
		/*
		 * This is the create channel path meaning that if there is NO
		 * registry available, we have to create one for this session.
		 */
		ret = buffer_reg_uid_create(usess->id, app->bits_per_long, app->uid,
				LTTNG_DOMAIN_UST, &reg_uid,
				ua_sess->root_shm_path, ua_sess->shm_path);
		if (ret < 0) {
			goto error;
		}
	} else {
		goto end;
	}

	/* Initialize registry. */
	ret = ust_registry_session_init(&reg_uid->registry->reg.ust, NULL,
			app->bits_per_long, app->uint8_t_alignment,
			app->uint16_t_alignment, app->uint32_t_alignment,
			app->uint64_t_alignment, app->long_alignment,
			app->byte_order, app->version.major,
			app->version.minor, reg_uid->root_shm_path,
			reg_uid->shm_path, usess->uid, usess->gid);
	if (ret < 0) {
		/*
		 * reg_uid->registry->reg.ust is NULL upon error, so we need to
		 * destroy the buffer registry, because it is always expected
		 * that if the buffer registry can be found, its ust registry is
		 * non-NULL.
		 */
		buffer_reg_uid_destroy(reg_uid, NULL);
		goto error;
	}
	/* Add node to teardown list of the session. */
	cds_list_add(&reg_uid->lnode, &usess->buffer_reg_uid_list);

	buffer_reg_uid_add(reg_uid);

	DBG3("UST app buffer registry per UID created successfully");
end:
	if (regp) {
		*regp = reg_uid;
	}
error:
	rcu_read_unlock();
	return ret;
}

/*
 * Create a session on the tracer side for the given app.
 *
 * On success, ua_sess_ptr is populated with the session pointer or else left
 * untouched. If the session was created, is_created is set to 1. On error,
 * it's left untouched. Note that ua_sess_ptr is mandatory but is_created can
 * be NULL.
 *
 * Returns 0 on success or else a negative code which is either -ENOMEM or
 * -ENOTCONN which is the default code if the ustctl_create_session fails.
 */
static int create_ust_app_session(struct ltt_ust_session *usess,
		struct ust_app *app, struct ust_app_session **ua_sess_ptr,
		int *is_created)
{
	int ret, created = 0;
	struct ust_app_session *ua_sess;

	assert(usess);
	assert(app);
	assert(ua_sess_ptr);

	health_code_update();

	ua_sess = lookup_session_by_app(usess, app);
	if (ua_sess == NULL) {
		DBG2("UST app pid: %d session id %" PRIu64 " not found, creating it",
				app->pid, usess->id);
		ua_sess = alloc_ust_app_session(app);
		if (ua_sess == NULL) {
			/* Only malloc can failed so something is really wrong */
			ret = -ENOMEM;
			goto error;
		}
		shadow_copy_session(ua_sess, usess, app);
		created = 1;
	}

	switch (usess->buffer_type) {
	case LTTNG_BUFFER_PER_PID:
		/* Init local registry. */
		ret = setup_buffer_reg_pid(ua_sess, app, NULL);
		if (ret < 0) {
			delete_ust_app_session(-1, ua_sess, app);
			goto error;
		}
		break;
	case LTTNG_BUFFER_PER_UID:
		/* Look for a global registry. If none exists, create one. */
		ret = setup_buffer_reg_uid(usess, ua_sess, app, NULL);
		if (ret < 0) {
			delete_ust_app_session(-1, ua_sess, app);
			goto error;
		}
		break;
	default:
		assert(0);
		ret = -EINVAL;
		goto error;
	}

	health_code_update();

	if (ua_sess->handle == -1) {
		ret = ustctl_create_session(app->sock);
		if (ret < 0) {
			if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) {
				ERR("Creating session for app pid %d with ret %d",
						app->pid, ret);
			} else {
				DBG("UST app creating session failed. Application is dead");
				/*
				 * This is normal behavior, an application can die during the
				 * creation process. Don't report an error so the execution can
				 * continue normally. This will get flagged ENOTCONN and the
				 * caller will handle it.
				 */
				ret = 0;
			}
			delete_ust_app_session(-1, ua_sess, app);
			if (ret != -ENOMEM) {
				/*
				 * Tracer is probably gone or got an internal error so let's
				 * behave like it will soon unregister or not usable.
				 */
				ret = -ENOTCONN;
			}
			goto error;
		}

		ua_sess->handle = ret;

		/* Add ust app session to app's HT */
		lttng_ht_node_init_u64(&ua_sess->node,
				ua_sess->tracing_id);
		lttng_ht_add_unique_u64(app->sessions, &ua_sess->node);

		DBG2("UST app session created successfully with handle %d", ret);
	}

	*ua_sess_ptr = ua_sess;
	if (is_created) {
		*is_created = created;
	}

	/* Everything went well. */
	ret = 0;

error:
	health_code_update();
	return ret;
}

/*
 * Match function for a hash table lookup of ust_app_ctx.
 *
 * It matches an ust app context based on the context type and, in the case
 * of perf counters, their name.
 */
static int ht_match_ust_app_ctx(struct cds_lfht_node *node, const void *_key)
{
	struct ust_app_ctx *ctx;
	const struct lttng_ust_context *key;

	assert(node);
	assert(_key);

	ctx = caa_container_of(node, struct ust_app_ctx, node.node);
	key = _key;

	/* Context type */
	if (ctx->ctx.ctx != key->ctx) {
		goto no_match;
	}

	/* Check the name in the case of perf thread counters. */
	if (key->ctx == LTTNG_UST_CONTEXT_PERF_THREAD_COUNTER) {
		if (strncmp(key->u.perf_counter.name,
			ctx->ctx.u.perf_counter.name,
			sizeof(key->u.perf_counter.name))) {
			goto no_match;
		}
	}

	/* Match. */
	return 1;

no_match:
	return 0;
}

/*
 * Lookup for an ust app context from an lttng_ust_context.
 *
 * Must be called while holding RCU read side lock.
 * Return an ust_app_ctx object or NULL on error.
 */
static
struct ust_app_ctx *find_ust_app_context(struct lttng_ht *ht,
		struct lttng_ust_context *uctx)
{
	struct lttng_ht_iter iter;
	struct lttng_ht_node_ulong *node;
	struct ust_app_ctx *app_ctx = NULL;

	assert(uctx);
	assert(ht);

	/* Lookup using the lttng_ust_context_type and a custom match fct. */
	cds_lfht_lookup(ht->ht, ht->hash_fct((void *) uctx->ctx, lttng_ht_seed),
			ht_match_ust_app_ctx, uctx, &iter.iter);
	node = lttng_ht_iter_get_node_ulong(&iter);
	if (!node) {
		goto end;
	}

	app_ctx = caa_container_of(node, struct ust_app_ctx, node);

end:
	return app_ctx;
}

/*
 * Create a context for the channel on the tracer.
 *
 * Called with UST app session lock held and a RCU read side lock.
 */
static
int create_ust_app_channel_context(struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan, struct lttng_ust_context *uctx,
		struct ust_app *app)
{
	int ret = 0;
	struct ust_app_ctx *ua_ctx;

	DBG2("UST app adding context to channel %s", ua_chan->name);

	ua_ctx = find_ust_app_context(ua_chan->ctx, uctx);
	if (ua_ctx) {
		ret = -EEXIST;
		goto error;
	}

	ua_ctx = alloc_ust_app_ctx(uctx);
	if (ua_ctx == NULL) {
		/* malloc failed */
		ret = -1;
		goto error;
	}

	lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx);
	lttng_ht_add_ulong(ua_chan->ctx, &ua_ctx->node);
	cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list);

	ret = create_ust_channel_context(ua_chan, ua_ctx, app);
	if (ret < 0) {
		goto error;
	}

error:
	return ret;
}

/*
 * Enable on the tracer side a ust app event for the session and channel.
 *
 * Called with UST app session lock held.
 */
static
int enable_ust_app_event(struct ust_app_session *ua_sess,
		struct ust_app_event *ua_event, struct ust_app *app)
{
	int ret;

	ret = enable_ust_event(app, ua_sess, ua_event);
	if (ret < 0) {
		goto error;
	}

	ua_event->enabled = 1;

error:
	return ret;
}

/*
 * Disable on the tracer side a ust app event for the session and channel.
 */
static int disable_ust_app_event(struct ust_app_session *ua_sess,
		struct ust_app_event *ua_event, struct ust_app *app)
{
	int ret;

	ret = disable_ust_event(app, ua_sess, ua_event);
	if (ret < 0) {
		goto error;
	}

	ua_event->enabled = 0;

error:
	return ret;
}

/*
 * Lookup ust app channel for session and disable it on the tracer side.
 */
static
int disable_ust_app_channel(struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan, struct ust_app *app)
{
	int ret;

	ret = disable_ust_channel(app, ua_sess, ua_chan);
	if (ret < 0) {
		goto error;
	}

	ua_chan->enabled = 0;

error:
	return ret;
}

/*
 * Lookup ust app channel for session and enable it on the tracer side. This
 * MUST be called with a RCU read side lock acquired.
 */
static int enable_ust_app_channel(struct ust_app_session *ua_sess,
		struct ltt_ust_channel *uchan, struct ust_app *app)
{
	int ret = 0;
	struct lttng_ht_iter iter;
	struct lttng_ht_node_str *ua_chan_node;
	struct ust_app_channel *ua_chan;

	lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter);
	ua_chan_node = lttng_ht_iter_get_node_str(&iter);
	if (ua_chan_node == NULL) {
		DBG2("Unable to find channel %s in ust session id %" PRIu64,
				uchan->name, ua_sess->tracing_id);
		goto error;
	}

	ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node);

	ret = enable_ust_channel(app, ua_sess, ua_chan);
	if (ret < 0) {
		goto error;
	}

error:
	return ret;
}

/*
 * Ask the consumer to create a channel and get it if successful.
 *
 * Return 0 on success or else a negative value.
 */
static int do_consumer_create_channel(struct ltt_ust_session *usess,
		struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan,
		int bitness, struct ust_registry_session *registry)
{
	int ret;
	unsigned int nb_fd = 0;
	struct consumer_socket *socket;

	assert(usess);
	assert(ua_sess);
	assert(ua_chan);
	assert(registry);

	rcu_read_lock();
	health_code_update();

	/* Get the right consumer socket for the application. */
	socket = consumer_find_socket_by_bitness(bitness, usess->consumer);
	if (!socket) {
		ret = -EINVAL;
		goto error;
	}

	health_code_update();

	/* Need one fd for the channel. */
	ret = lttng_fd_get(LTTNG_FD_APPS, 1);
	if (ret < 0) {
		ERR("Exhausted number of available FD upon create channel");
		goto error;
	}

	/*
	 * Ask consumer to create channel. The consumer will return the number of
	 * stream we have to expect.
	 */
	ret = ust_consumer_ask_channel(ua_sess, ua_chan, usess->consumer, socket,
			registry);
	if (ret < 0) {
		goto error_ask;
	}

	/*
	 * Compute the number of fd needed before receiving them. It must be 2 per
	 * stream (2 being the default value here).
	 */
	nb_fd = DEFAULT_UST_STREAM_FD_NUM * ua_chan->expected_stream_count;

	/* Reserve the amount of file descriptor we need. */
	ret = lttng_fd_get(LTTNG_FD_APPS, nb_fd);
	if (ret < 0) {
		ERR("Exhausted number of available FD upon create channel");
		goto error_fd_get_stream;
	}

	health_code_update();

	/*
	 * Now get the channel from the consumer. This call wil populate the stream
	 * list of that channel and set the ust objects.
	 */
	if (usess->consumer->enabled) {
		ret = ust_consumer_get_channel(socket, ua_chan);
		if (ret < 0) {
			goto error_destroy;
		}
	}

	rcu_read_unlock();
	return 0;

error_destroy:
	lttng_fd_put(LTTNG_FD_APPS, nb_fd);
error_fd_get_stream:
	/*
	 * Initiate a destroy channel on the consumer since we had an error
	 * handling it on our side. The return value is of no importance since we
	 * already have a ret value set by the previous error that we need to
	 * return.
	 */
	(void) ust_consumer_destroy_channel(socket, ua_chan);
error_ask:
	lttng_fd_put(LTTNG_FD_APPS, 1);
error:
	health_code_update();
	rcu_read_unlock();
	return ret;
}

/*
 * Duplicate the ust data object of the ust app stream and save it in the
 * buffer registry stream.
 *
 * Return 0 on success or else a negative value.
 */
static int duplicate_stream_object(struct buffer_reg_stream *reg_stream,
		struct ust_app_stream *stream)
{
	int ret;

	assert(reg_stream);
	assert(stream);

	/* Reserve the amount of file descriptor we need. */
	ret = lttng_fd_get(LTTNG_FD_APPS, 2);
	if (ret < 0) {
		ERR("Exhausted number of available FD upon duplicate stream");
		goto error;
	}

	/* Duplicate object for stream once the original is in the registry. */
	ret = ustctl_duplicate_ust_object_data(&stream->obj,
			reg_stream->obj.ust);
	if (ret < 0) {
		ERR("Duplicate stream obj from %p to %p failed with ret %d",
				reg_stream->obj.ust, stream->obj, ret);
		lttng_fd_put(LTTNG_FD_APPS, 2);
		goto error;
	}
	stream->handle = stream->obj->handle;

error:
	return ret;
}

/*
 * Duplicate the ust data object of the ust app. channel and save it in the
 * buffer registry channel.
 *
 * Return 0 on success or else a negative value.
 */
static int duplicate_channel_object(struct buffer_reg_channel *reg_chan,
		struct ust_app_channel *ua_chan)
{
	int ret;

	assert(reg_chan);
	assert(ua_chan);

	/* Need two fds for the channel. */
	ret = lttng_fd_get(LTTNG_FD_APPS, 1);
	if (ret < 0) {
		ERR("Exhausted number of available FD upon duplicate channel");
		goto error_fd_get;
	}

	/* Duplicate object for stream once the original is in the registry. */
	ret = ustctl_duplicate_ust_object_data(&ua_chan->obj, reg_chan->obj.ust);
	if (ret < 0) {
		ERR("Duplicate channel obj from %p to %p failed with ret: %d",
				reg_chan->obj.ust, ua_chan->obj, ret);
		goto error;
	}
	ua_chan->handle = ua_chan->obj->handle;

	return 0;

error:
	lttng_fd_put(LTTNG_FD_APPS, 1);
error_fd_get:
	return ret;
}

/*
 * For a given channel buffer registry, setup all streams of the given ust
 * application channel.
 *
 * Return 0 on success or else a negative value.
 */
static int setup_buffer_reg_streams(struct buffer_reg_channel *reg_chan,
		struct ust_app_channel *ua_chan)
{
	int ret = 0;
	struct ust_app_stream *stream, *stmp;

	assert(reg_chan);
	assert(ua_chan);

	DBG2("UST app setup buffer registry stream");

	/* Send all streams to application. */
	cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) {
		struct buffer_reg_stream *reg_stream;

		ret = buffer_reg_stream_create(&reg_stream);
		if (ret < 0) {
			goto error;
		}

		/*
		 * Keep original pointer and nullify it in the stream so the delete
		 * stream call does not release the object.
		 */
		reg_stream->obj.ust = stream->obj;
		stream->obj = NULL;
		buffer_reg_stream_add(reg_stream, reg_chan);

		/* We don't need the streams anymore. */
		cds_list_del(&stream->list);
		delete_ust_app_stream(-1, stream);
	}

error:
	return ret;
}

/*
 * Create a buffer registry channel for the given session registry and
 * application channel object. If regp pointer is valid, it's set with the
 * created object. Important, the created object is NOT added to the session
 * registry hash table.
 *
 * Return 0 on success else a negative value.
 */
static int create_buffer_reg_channel(struct buffer_reg_session *reg_sess,
		struct ust_app_channel *ua_chan, struct buffer_reg_channel **regp)
{
	int ret;
	struct buffer_reg_channel *reg_chan = NULL;

	assert(reg_sess);
	assert(ua_chan);

	DBG2("UST app creating buffer registry channel for %s", ua_chan->name);

	/* Create buffer registry channel. */
	ret = buffer_reg_channel_create(ua_chan->tracing_channel_id, &reg_chan);
	if (ret < 0) {
		goto error_create;
	}
	assert(reg_chan);
	reg_chan->consumer_key = ua_chan->key;
	reg_chan->subbuf_size = ua_chan->attr.subbuf_size;
	reg_chan->num_subbuf = ua_chan->attr.num_subbuf;

	/* Create and add a channel registry to session. */
	ret = ust_registry_channel_add(reg_sess->reg.ust,
			ua_chan->tracing_channel_id);
	if (ret < 0) {
		goto error;
	}
	buffer_reg_channel_add(reg_sess, reg_chan);

	if (regp) {
		*regp = reg_chan;
	}

	return 0;

error:
	/* Safe because the registry channel object was not added to any HT. */
	buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST);
error_create:
	return ret;
}

/*
 * Setup buffer registry channel for the given session registry and application
 * channel object. If regp pointer is valid, it's set with the created object.
 *
 * Return 0 on success else a negative value.
 */
static int setup_buffer_reg_channel(struct buffer_reg_session *reg_sess,
		struct ust_app_channel *ua_chan, struct buffer_reg_channel *reg_chan)
{
	int ret;

	assert(reg_sess);
	assert(reg_chan);
	assert(ua_chan);
	assert(ua_chan->obj);

	DBG2("UST app setup buffer registry channel for %s", ua_chan->name);

	/* Setup all streams for the registry. */
	ret = setup_buffer_reg_streams(reg_chan, ua_chan);
	if (ret < 0) {
		goto error;
	}

	reg_chan->obj.ust = ua_chan->obj;
	ua_chan->obj = NULL;

	return 0;

error:
	buffer_reg_channel_remove(reg_sess, reg_chan);
	buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST);
	return ret;
}

/*
 * Send buffer registry channel to the application.
 *
 * Return 0 on success else a negative value.
 */
static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan,
		struct ust_app *app, struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan)
{
	int ret;
	struct buffer_reg_stream *reg_stream;

	assert(reg_chan);
	assert(app);
	assert(ua_sess);
	assert(ua_chan);

	DBG("UST app sending buffer registry channel to ust sock %d", app->sock);

	ret = duplicate_channel_object(reg_chan, ua_chan);
	if (ret < 0) {
		goto error;
	}

	/* Send channel to the application. */
	ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan);
	if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
		ret = -ENOTCONN;	/* Caused by app exiting. */
		goto error;
	} else if (ret < 0) {
		goto error;
	}

	health_code_update();

	/* Send all streams to application. */
	pthread_mutex_lock(&reg_chan->stream_list_lock);
	cds_list_for_each_entry(reg_stream, &reg_chan->streams, lnode) {
		struct ust_app_stream stream;

		ret = duplicate_stream_object(reg_stream, &stream);
		if (ret < 0) {
			goto error_stream_unlock;
		}

		ret = ust_consumer_send_stream_to_ust(app, ua_chan, &stream);
		if (ret < 0) {
			(void) release_ust_app_stream(-1, &stream);
			if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) {
				ret = -ENOTCONN; /* Caused by app exiting. */
				goto error_stream_unlock;
			} else if (ret < 0) {
				goto error_stream_unlock;
			}
			goto error_stream_unlock;
		}

		/*
		 * The return value is not important here. This function will output an
		 * error if needed.
		 */
		(void) release_ust_app_stream(-1, &stream);
	}
	ua_chan->is_sent = 1;

error_stream_unlock:
	pthread_mutex_unlock(&reg_chan->stream_list_lock);
error:
	return ret;
}

/*
 * Create and send to the application the created buffers with per UID buffers.
 *
 * Return 0 on success else a negative value.
 */
static int create_channel_per_uid(struct ust_app *app,
		struct ltt_ust_session *usess, struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan)
{
	int ret;
	struct buffer_reg_uid *reg_uid;
	struct buffer_reg_channel *reg_chan;

	assert(app);
	assert(usess);
	assert(ua_sess);
	assert(ua_chan);

	DBG("UST app creating channel %s with per UID buffers", ua_chan->name);

	reg_uid = buffer_reg_uid_find(usess->id, app->bits_per_long, app->uid);
	/*
	 * The session creation handles the creation of this global registry
	 * object. If none can be find, there is a code flow problem or a
	 * teardown race.
	 */
	assert(reg_uid);

	reg_chan = buffer_reg_channel_find(ua_chan->tracing_channel_id,
			reg_uid);
	if (!reg_chan) {
		/* Create the buffer registry channel object. */
		ret = create_buffer_reg_channel(reg_uid->registry, ua_chan, &reg_chan);
		if (ret < 0) {
			ERR("Error creating the UST channel \"%s\" registry instance",
				ua_chan->name);
			goto error;
		}
		assert(reg_chan);

		/*
		 * Create the buffers on the consumer side. This call populates the
		 * ust app channel object with all streams and data object.
		 */
		ret = do_consumer_create_channel(usess, ua_sess, ua_chan,
				app->bits_per_long, reg_uid->registry->reg.ust);
		if (ret < 0) {
			ERR("Error creating UST channel \"%s\" on the consumer daemon",
				ua_chan->name);

			/*
			 * Let's remove the previously created buffer registry channel so
			 * it's not visible anymore in the session registry.
			 */
			ust_registry_channel_del_free(reg_uid->registry->reg.ust,
					ua_chan->tracing_channel_id);
			buffer_reg_channel_remove(reg_uid->registry, reg_chan);
			buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST);
			goto error;
		}

		/*
		 * Setup the streams and add it to the session registry.
		 */
		ret = setup_buffer_reg_channel(reg_uid->registry, ua_chan, reg_chan);
		if (ret < 0) {
			ERR("Error setting up UST channel \"%s\"",
				ua_chan->name);
			goto error;
		}

	}

	/* Send buffers to the application. */
	ret = send_channel_uid_to_ust(reg_chan, app, ua_sess, ua_chan);
	if (ret < 0) {
		if (ret != -ENOTCONN) {
			ERR("Error sending channel to application");
		}
		goto error;
	}

error:
	return ret;
}

/*
 * Create and send to the application the created buffers with per PID buffers.
 *
 * Return 0 on success else a negative value.
 */
static int create_channel_per_pid(struct ust_app *app,
		struct ltt_ust_session *usess, struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan)
{
	int ret;
	struct ust_registry_session *registry;

	assert(app);
	assert(usess);
	assert(ua_sess);
	assert(ua_chan);

	DBG("UST app creating channel %s with per PID buffers", ua_chan->name);

	rcu_read_lock();

	registry = get_session_registry(ua_sess);
	assert(registry);

	/* Create and add a new channel registry to session. */
	ret = ust_registry_channel_add(registry, ua_chan->key);
	if (ret < 0) {
		ERR("Error creating the UST channel \"%s\" registry instance",
			ua_chan->name);
		goto error;
	}

	/* Create and get channel on the consumer side. */
	ret = do_consumer_create_channel(usess, ua_sess, ua_chan,
			app->bits_per_long, registry);
	if (ret < 0) {
		ERR("Error creating UST channel \"%s\" on the consumer daemon",
			ua_chan->name);
		goto error;
	}

	ret = send_channel_pid_to_ust(app, ua_sess, ua_chan);
	if (ret < 0) {
		if (ret != -ENOTCONN) {
			ERR("Error sending channel to application");
		}
		goto error;
	}

error:
	rcu_read_unlock();
	return ret;
}

/*
 * From an already allocated ust app channel, create the channel buffers if
 * need and send it to the application. This MUST be called with a RCU read
 * side lock acquired.
 *
 * Return 0 on success or else a negative value. Returns -ENOTCONN if
 * the application exited concurrently.
 */
static int do_create_channel(struct ust_app *app,
		struct ltt_ust_session *usess, struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan)
{
	int ret;

	assert(app);
	assert(usess);
	assert(ua_sess);
	assert(ua_chan);

	/* Handle buffer type before sending the channel to the application. */
	switch (usess->buffer_type) {
	case LTTNG_BUFFER_PER_UID:
	{
		ret = create_channel_per_uid(app, usess, ua_sess, ua_chan);
		if (ret < 0) {
			goto error;
		}
		break;
	}
	case LTTNG_BUFFER_PER_PID:
	{
		ret = create_channel_per_pid(app, usess, ua_sess, ua_chan);
		if (ret < 0) {
			goto error;
		}
		break;
	}
	default:
		assert(0);
		ret = -EINVAL;
		goto error;
	}

	/* Initialize ust objd object using the received handle and add it. */
	lttng_ht_node_init_ulong(&ua_chan->ust_objd_node, ua_chan->handle);
	lttng_ht_add_unique_ulong(app->ust_objd, &ua_chan->ust_objd_node);

	/* If channel is not enabled, disable it on the tracer */
	if (!ua_chan->enabled) {
		ret = disable_ust_channel(app, ua_sess, ua_chan);
		if (ret < 0) {
			goto error;
		}
	}

error:
	return ret;
}

/*
 * Create UST app channel and create it on the tracer. Set ua_chanp of the
 * newly created channel if not NULL.
 *
 * Called with UST app session lock and RCU read-side lock held.
 *
 * Return 0 on success or else a negative value. Returns -ENOTCONN if
 * the application exited concurrently.
 */
static int create_ust_app_channel(struct ust_app_session *ua_sess,
		struct ltt_ust_channel *uchan, struct ust_app *app,
		enum lttng_ust_chan_type type, struct ltt_ust_session *usess,
		struct ust_app_channel **ua_chanp)
{
	int ret = 0;
	struct lttng_ht_iter iter;
	struct lttng_ht_node_str *ua_chan_node;
	struct ust_app_channel *ua_chan;

	/* Lookup channel in the ust app session */
	lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter);
	ua_chan_node = lttng_ht_iter_get_node_str(&iter);
	if (ua_chan_node != NULL) {
		ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node);
		goto end;
	}

	ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr);
	if (ua_chan == NULL) {
		/* Only malloc can fail here */
		ret = -ENOMEM;
		goto error_alloc;
	}
	shadow_copy_channel(ua_chan, uchan);

	/* Set channel type. */
	ua_chan->attr.type = type;

	ret = do_create_channel(app, usess, ua_sess, ua_chan);
	if (ret < 0) {
		goto error;
	}

	DBG2("UST app create channel %s for PID %d completed", ua_chan->name,
			app->pid);

	/* Only add the channel if successful on the tracer side. */
	lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node);

end:
	if (ua_chanp) {
		*ua_chanp = ua_chan;
	}

	/* Everything went well. */
	return 0;

error:
	delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app);
error_alloc:
	return ret;
}

/*
 * Create UST app event and create it on the tracer side.
 *
 * Called with ust app session mutex held.
 */
static
int create_ust_app_event(struct ust_app_session *ua_sess,
		struct ust_app_channel *ua_chan, struct ltt_ust_event *uevent,
		struct ust_app *app)
{
	int ret = 0;
	struct ust_app_event *ua_event;

	/* Get event node */
	ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name,
			uevent->filter, uevent->attr.loglevel, uevent->exclusion);
	if (ua_event != NULL) {
		ret = -EEXIST;
		goto end;
	}

	/* Does not exist so create one */
	ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr);
	if (ua_event == NULL) {
		/* Only malloc can failed so something is really wrong */
		ret = -ENOMEM;
		goto end;
	}
	shadow_copy_event(ua_event, uevent);

	/* Create it on the tracer side */
	ret = create_ust_event(app, ua_sess, ua_chan, ua_event);
	if (ret < 0) {
		/* Not found previously means that it does not exist on the tracer */
		assert(ret != -LTTNG_UST_ERR_EXIST);
		goto error;
	}

	add_unique_ust_app_event(ua_chan, ua_event);

	DBG2("UST app create event %s for PID %d completed", ua_event->name,
			app->pid);

end:
	return ret;

error:
	/* Valid. Calling here is already in a read side lock */
	delete_ust_app_event(-1, ua_event);
	return ret;
}

/*
 * Create UST metadata and open it on the tracer side.
 *
 * Called with UST app session lock held and RCU read side lock.
 */
static int create_ust_app_metadata(struct ust_app_session *ua_sess,
		struct ust_app *app, struct consumer_output *consumer)
{
	int ret = 0;
	struct ust_app_channel *metadata;
	struct consumer_socket *socket;
	struct ust_registry_session *registry;

	assert(ua_sess);
	assert(app);
	assert(consumer);

	registry = get_session_registry(ua_sess);
	assert(registry);

	pthread_mutex_lock(&registry->lock);

	/* Metadata already exists for this registry or it was closed previously */
	if (registry->metadata_key || registry->metadata_closed) {
		ret = 0;
		goto error;
	}

	/* Allocate UST metadata */
	metadata = alloc_ust_app_channel(DEFAULT_METADATA_NAME, ua_sess, NULL);
	if (!metadata) {
		/* malloc() failed */
		ret = -ENOMEM;
		goto error;
	}

	memcpy(&metadata->attr, &ua_sess->metadata_attr, sizeof(metadata->attr));

	/* Need one fd for the channel. */
	ret = lttng_fd_get(LTTNG_FD_APPS, 1);
	if (ret < 0) {
		ERR("Exhausted number of available FD upon create metadata");
		goto error;
	}

	/* Get the right consumer socket for the application. */
	socket = consumer_find_socket_by_bitness(app->bits_per_long, consumer);
	if (!socket) {
		ret = -EINVAL;
		goto error_consumer;
	}

	/*
	 * Keep metadata key so we can identify it on the consumer side. Assign it
	 * to the registry *before* we ask the consumer so we avoid the race of the
	 * consumer requesting the metadata and the ask_channel call on our side
	 * did not returned yet.
	 */
	registry->metadata_key = metadata->key;

	/*
	 * Ask the metadata channel creation to the consumer. The metadata object
	 * will be created by the consumer and kept their. However, the stream is
	 * never added or monitored until we do a first push metadata to the
	 * consumer.
	 */
	ret = ust_consumer_ask_channel(ua_sess, metadata, consumer, socket,
			registry);
	if (ret < 0) {
		/* Nullify the metadata key so we don't try to close it later on. */
		registry->metadata_key = 0;
		goto error_consumer;
	}

	/*
	 * The setup command will make the metadata stream be sent to the relayd,
	 * if applicable, and the thread managing the metadatas. This is important
	 * because after this point, if an error occurs, the only way the stream
	 * can be deleted is to be monitored in the consumer.
	 */
	ret = consumer_setup_metadata(socket, metadata->key);
	if (ret < 0) {
		/* Nullify the metadata key so we don't try to close it later on. */
		registry->metadata_key = 0;
		goto error_consumer;
	}

	DBG2("UST metadata with key %" PRIu64 " created for app pid %d",
			metadata->key, app->pid);

error_consumer:
	lttng_fd_put(LTTNG_FD_APPS, 1);
	delete_ust_app_channel(-1, metadata, app);
error:
	pthread_mutex_unlock(&registry->lock);
	return ret;
}

/*
 * Return ust app pointer or NULL if not found. RCU read side lock MUST be
 * acquired before calling this function.
 */
struct ust_app *ust_app_find_by_pid(pid_t pid)
{
	struct ust_app *app = NULL;
	struct lttng_ht_node_ulong *node;
	struct lttng_ht_iter iter;

	lttng_ht_lookup(ust_app_ht, (void *)((unsigned long) pid), &iter);
	node = lttng_ht_iter_get_node_ulong(&iter);
	if (node == NULL) {
		DBG2("UST app no found with pid %d", pid);
		goto error;
	}

	DBG2("Found UST app by pid %d", pid);

	app = caa_container_of(node, struct ust_app, pid_n);

error:
	return app;
}

/*
 * Allocate and init an UST app object using the registration information and
 * the command socket. This is called when the command socket connects to the
 * session daemon.
 *
 * The object is returned on success or else NULL.
 */
struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock)
{
	struct ust_app *lta = NULL;

	assert(msg);
	assert(sock >= 0);

	DBG3("UST app creating application for socket %d", sock);

	if ((msg->bits_per_long == 64 &&
				(uatomic_read(&ust_consumerd64_fd) == -EINVAL))
			|| (msg->bits_per_long == 32 &&
				(uatomic_read(&ust_consumerd32_fd) == -EINVAL))) {
		ERR("Registration failed: application \"%s\" (pid: %d) has "
				"%d-bit long, but no consumerd for this size is available.\n",
				msg->name, msg->pid, msg->bits_per_long);
		goto error;
	}

	lta = zmalloc(sizeof(struct ust_app));
	if (lta == NULL) {
		PERROR("malloc");
		goto error;
	}

	lta->ppid = msg->ppid;
	lta->uid = msg->uid;
	lta->gid = msg->gid;

	lta->bits_per_long = msg->bits_per_long;
	lta->uint8_t_alignment = msg->uint8_t_alignment;
	lta->uint16_t_alignment = msg->uint16_t_alignment;
	lta->uint32_t_alignment = msg->uint32_t_alignment;
	lta->uint64_t_alignment = msg->uint64_t_alignment;
	lta->long_alignment = msg->long_alignment;
	lta->byte_order = msg->byte_order;

	lta->v_major = msg->major;
	lta->v_minor = msg->minor;
	lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
	lta->ust_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
	lta->notify_sock = -1;

	/* Copy name and make sure it's NULL terminated. */
	strncpy(lta->name, msg->name, sizeof(lta->name));
	lta->name[UST_APP_PROCNAME_LEN] = '\0';

	/*
	 * Before this can be called, when receiving the registration information,
	 * the application compatibility is checked. So, at this point, the
	 * application can work with this session daemon.
	 */
	lta->compatible = 1;

	lta->pid = msg->pid;
	lttng_ht_node_init_ulong(&lta->pid_n, (unsigned long) lta->pid);
	lta->sock = sock;
	lttng_ht_node_init_ulong(&lta->sock_n, (unsigned long) lta->sock);

	CDS_INIT_LIST_HEAD(&lta->teardown_head);

error:
	return lta;
}

/*
 * For a given application object, add it to every hash table.
 */
void ust_app_add(struct ust_app *app)
{
	assert(app);
	assert(app->notify_sock >= 0);

	rcu_read_lock();

	/*
	 * On a re-registration, we want to kick out the previous registration of
	 * that pid
	 */
	lttng_ht_add_replace_ulong(ust_app_ht, &app->pid_n);

	/*
	 * The socket _should_ be unique until _we_ call close. So, a add_unique
	 * for the ust_app_ht_by_sock is used which asserts fail if the entry was
	 * already in the table.
	 */
	lttng_ht_add_unique_ulong(ust_app_ht_by_sock, &app->sock_n);

	/* Add application to the notify socket hash table. */
	lttng_ht_node_init_ulong(&app->notify_sock_n, app->notify_sock);
	lttng_ht_add_unique_ulong(ust_app_ht_by_notify_sock, &app->notify_sock_n);

	DBG("App registered with pid:%d ppid:%d uid:%d gid:%d sock:%d name:%s "
			"notify_sock:%d (version %d.%d)", app->pid, app->ppid, app->uid,
			app->gid, app->sock, app->name, app->notify_sock, app->v_major,
			app->v_minor);

	rcu_read_unlock();
}

/*
 * Set the application version into the object.
 *
 * Return 0 on success else a negative value either an errno code or a
 * LTTng-UST error code.
 */
int ust_app_version(struct ust_app *app)
{
	int ret;

	assert(app);

	ret = ustctl_tracer_version(app->sock, &app->version);
	if (ret < 0) {
		if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) {
			ERR("UST app %d version failed with ret %d", app->sock, ret);
		} else {
			DBG3("UST app %d version failed. Application is dead", app->sock);
		}
	}

	return ret;
}

/*
 * Unregister app by removing it from the global traceable app list and freeing
 * the data struct.
 *
 * The socket is already closed at this point so no close to sock.
 */
void ust_app_unregister(int sock)
{
	struct ust_app *lta;
	struct lttng_ht_node_ulong *node;
	struct lttng_ht_iter ust_app_sock_iter;
	struct lttng_ht_iter iter;
	struct ust_app_session *ua_sess;
	int ret;

	rcu_read_lock();

	/* Get the node reference for a call_rcu */
	lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &ust_app_sock_iter);
	node = lttng_ht_iter_get_node_ulong(&ust_app_sock_iter);
	assert(node);

	lta = caa_container_of(node, struct ust_app, sock_n);
	DBG("PID %d unregistering with sock %d", lta->pid, sock);

	/*
	 * For per-PID buffers, perform "push metadata" and flush all
	 * application streams before removing app from hash tables,
	 * ensuring proper behavior of data_pending check.
	 * Remove sessions so they are not visible during deletion.
	 */
	cds_lfht_for_each_entry(lta->sessions->ht, &iter.iter, ua_sess,
			node.node) {
		struct ust_registry_session *registry;