X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fsession.c;h=6c10aaeba9259286cdac4a43db72619d8a4e8f51;hp=e57463d3e58777eb78612c78c4c77da4296d73dc;hb=fb9a95c4d6242bd8336b638c90a7d8f846125659;hpb=9c6518bcd2b538c09de5e1d1df303b97ec203cc6

diff --git a/src/bin/lttng-sessiond/session.c b/src/bin/lttng-sessiond/session.c
index e57463d3e..6c10aaeba 100644
--- a/src/bin/lttng-sessiond/session.c
+++ b/src/bin/lttng-sessiond/session.c
@@ -25,12 +25,20 @@
 #include <urcu.h>
 #include <dirent.h>
 #include <sys/types.h>
+#include <pthread.h>
 
 #include <common/common.h>
+#include <common/utils.h>
+#include <common/trace-chunk.h>
 #include <common/sessiond-comm/sessiond-comm.h>
+#include <lttng/location-internal.h>
+#include "lttng-sessiond.h"
+#include "kernel.h"
 
 #include "session.h"
 #include "utils.h"
+#include "trace-ust.h"
+#include "timer.h"
 
 /*
  * NOTES:
@@ -49,6 +57,7 @@
 static struct ltt_session_list ltt_session_list = {
 	.head = CDS_LIST_HEAD_INIT(ltt_session_list.head),
 	.lock = PTHREAD_MUTEX_INITIALIZER,
+	.removal_cond = PTHREAD_COND_INITIALIZER,
 	.next_uuid = 0,
 };
 
@@ -58,6 +67,13 @@ static const char *forbidden_name_chars = "/";
 /* Global hash table to keep the sessions, indexed by id. */
 static struct lttng_ht *ltt_sessions_ht_by_id = NULL;
 
+struct consumer_create_chunk_transaction {
+	struct consumer_socket *socket;
+	struct lttng_trace_chunk *new_chunk;
+	struct lttng_trace_chunk *previous_chunk;
+	bool new_chunk_created;
+};
+
 /*
  * Validate the session name for forbidden characters.
  *
@@ -125,6 +141,19 @@ struct ltt_session_list *session_get_list(void)
 	return &ltt_session_list;
 }
 
+/*
+ * Returns once the session list is empty.
+ */
+void session_list_wait_empty(void)
+{
+	pthread_mutex_lock(&ltt_session_list.lock);
+	while (!cds_list_empty(&ltt_session_list.head)) {
+		pthread_cond_wait(&ltt_session_list.removal_cond,
+				&ltt_session_list.lock);
+	}
+	pthread_mutex_unlock(&ltt_session_list.lock);
+}
+
 /*
  * Acquire session list lock
  */
@@ -133,6 +162,14 @@ void session_lock_list(void)
 	pthread_mutex_lock(&ltt_session_list.lock);
 }
 
+/*
+ * Try to acquire session list lock
+ */
+int session_trylock_list(void)
+{
+	return pthread_mutex_trylock(&ltt_session_list.lock);
+}
+
 /*
  * Release session list lock
  */
@@ -141,6 +178,116 @@ void session_unlock_list(void)
 	pthread_mutex_unlock(&ltt_session_list.lock);
 }
 
+/*
+ * Get the session's consumer destination type.
+ *
+ * The caller must hold the session lock.
+ */
+enum consumer_dst_type session_get_consumer_destination_type(
+		const struct ltt_session *session)
+{
+	/*
+	 * The output information is duplicated in both of those session types.
+	 * Hence, it doesn't matter from which it is retrieved. However, it is
+	 * possible for only one of them to be set.
+	 */
+	return session->kernel_session ?
+			session->kernel_session->consumer->type :
+			session->ust_session->consumer->type;
+}
+
+/*
+ * Get the session's consumer network hostname.
+ * The caller must ensure that the destination is of type "net".
+ *
+ * The caller must hold the session lock.
+ */
+const char *session_get_net_consumer_hostname(const struct ltt_session *session)
+{
+	const char *hostname = NULL;
+	const struct consumer_output *output;
+
+	output = session->kernel_session ?
+			session->kernel_session->consumer :
+			session->ust_session->consumer;
+
+	/*
+	 * hostname is assumed to be the same for both control and data
+	 * connections.
+	 */
+	switch (output->dst.net.control.dtype) {
+	case LTTNG_DST_IPV4:
+		hostname = output->dst.net.control.dst.ipv4;
+		break;
+	case LTTNG_DST_IPV6:
+		hostname = output->dst.net.control.dst.ipv6;
+		break;
+	default:
+		abort();
+	}
+	return hostname;
+}
+
+/*
+ * Get the session's consumer network control and data ports.
+ * The caller must ensure that the destination is of type "net".
+ *
+ * The caller must hold the session lock.
+ */
+void session_get_net_consumer_ports(const struct ltt_session *session,
+		uint16_t *control_port, uint16_t *data_port)
+{
+	const struct consumer_output *output;
+
+	output = session->kernel_session ?
+			session->kernel_session->consumer :
+			session->ust_session->consumer;
+	*control_port = output->dst.net.control.port;
+	*data_port = output->dst.net.data.port;
+}
+
+/*
+ * Get the location of the latest trace archive produced by a rotation.
+ *
+ * The caller must hold the session lock.
+ */
+struct lttng_trace_archive_location *session_get_trace_archive_location(
+		struct ltt_session *session)
+{
+	struct lttng_trace_archive_location *location = NULL;
+
+	if (session->rotation_state != LTTNG_ROTATION_STATE_COMPLETED) {
+		goto end;
+	}
+
+	switch (session_get_consumer_destination_type(session)) {
+	case CONSUMER_DST_LOCAL:
+		location = lttng_trace_archive_location_local_create(
+				session->rotation_chunk.current_rotate_path);
+		break;
+	case CONSUMER_DST_NET:
+	{
+		const char *hostname;
+		uint16_t control_port, data_port;
+
+		hostname = session_get_net_consumer_hostname(session);
+		session_get_net_consumer_ports(session,
+				&control_port,
+				&data_port);
+		location = lttng_trace_archive_location_relay_create(
+				hostname,
+				LTTNG_TRACE_ARCHIVE_LOCATION_RELAY_PROTOCOL_TYPE_TCP,
+				control_port, data_port,
+				session->rotation_chunk.current_rotate_path);
+		break;
+	}
+	default:
+		abort();
+	}
+end:
+	return location;
+}
+
 /*
  * Allocate the ltt_sessions_ht_by_id HT.
  *
@@ -166,7 +313,7 @@ end:
  *
  * The session list lock must be held.
  */
-void ltt_sessions_ht_destroy(void)
+static void ltt_sessions_ht_destroy(void)
 {
 	if (!ltt_sessions_ht_by_id) {
 		return;
@@ -262,29 +409,454 @@ void session_unlock(struct ltt_session *session)
 	pthread_mutex_unlock(&session->lock);
 }
 
+static
+int _session_set_trace_chunk_no_lock_check(struct ltt_session *session,
+		struct lttng_trace_chunk *new_trace_chunk)
+{
+	int ret;
+	unsigned int i, refs_to_acquire = 0, refs_acquired = 0, refs_to_release = 0;
+	unsigned int consumer_count = 0;
+	/*
+	 * The maximum amount of consumers to reach is 3
+	 * (32/64 userspace + kernel).
+	 */
+	struct consumer_create_chunk_transaction transactions[3] = {};
+	struct cds_lfht_iter iter;
+	struct consumer_socket *socket;
+	bool close_error_occured = false;
+
+	if (new_trace_chunk) {
+		uint64_t chunk_id;
+		enum lttng_trace_chunk_status chunk_status =
+				lttng_trace_chunk_get_id(new_trace_chunk,
+					&chunk_id);
+
+		assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
+		LTTNG_OPTIONAL_SET(&session->last_trace_chunk_id, chunk_id)
+	}
+
+	if (new_trace_chunk) {
+		refs_to_acquire = 1;
+		refs_to_acquire += !!session->ust_session;
+		refs_to_acquire += !!session->kernel_session;
+	}
+
+	/*
+	 * Build a list of consumers to reach to announce the new trace chunk.
+	 *
+	 * Rolling back the annoucement in case of an error is important since
+	 * not doing so would result in a leak; the chunk will not be
+	 * "reclaimed" by the consumer(s) since they have no concept of the
+	 * lifetime of a session.
+	 */
+	if (session->ust_session) {
+		cds_lfht_for_each_entry(
+				session->ust_session->consumer->socks->ht,
+				&iter, socket, node.node) {
+			transactions[consumer_count].socket = socket;
+			transactions[consumer_count].new_chunk = new_trace_chunk;
+			transactions[consumer_count].previous_chunk =
+					session->current_trace_chunk;
+			consumer_count++;
+			assert(consumer_count <= 3);
+		}
+	}
+	if (session->kernel_session) {
+		cds_lfht_for_each_entry(
+				session->kernel_session->consumer->socks->ht,
+				&iter, socket, node.node) {
+			transactions[consumer_count].socket = socket;
+			transactions[consumer_count].new_chunk = new_trace_chunk;
+			transactions[consumer_count].previous_chunk =
+					session->current_trace_chunk;
+			consumer_count++;
+			assert(consumer_count <= 3);
+		}
+	}
+	for (refs_acquired = 0; refs_acquired < refs_to_acquire; refs_acquired++) {
+		if (new_trace_chunk && !lttng_trace_chunk_get(new_trace_chunk)) {
+			ERR("Failed to acquire reference to new current trace chunk of session \"%s\"",
+					session->name);
+			goto error;
+		}
+	}
+
+	/*
+	 * Close the previous chunk on remote peers (consumers and relayd).
+	 */
+	for (i = 0; i < consumer_count; i++) {
+		if (!transactions[i].previous_chunk) {
+			continue;
+		}
+		pthread_mutex_lock(transactions[i].socket->lock);
+		ret = consumer_close_trace_chunk(transactions[i].socket,
+				session->consumer->net_seq_index,
+				session->id,
+				transactions[i].previous_chunk);
+		pthread_mutex_unlock(transactions[i].socket->lock);
+		if (ret) {
+			ERR("Failed to close trace chunk on consumer");
+			close_error_occured = true;
+		}
+	}
+
+	if (close_error_occured) {
+		/*
+		 * Skip the creation of the new trace chunk and report the
+		 * error.
+		 */
+		goto error;
+	}
+
+	/* Create the new chunk on remote peers (consumers and relayd) */
+	if (new_trace_chunk) {
+		for (i = 0; i < consumer_count; i++) {
+			pthread_mutex_lock(transactions[i].socket->lock);
+			ret = consumer_create_trace_chunk(transactions[i].socket,
+					session->consumer->net_seq_index,
+					session->id,
+					transactions[i].new_chunk);
+			pthread_mutex_unlock(transactions[i].socket->lock);
+			if (ret) {
+				ERR("Failed to create trace chunk on consumer");
+				goto error;
+			}
+			/* This will have to be rolled-back on error. */
+			transactions[i].new_chunk_created = true;
+		}
+	}
+
+	lttng_trace_chunk_put(session->current_trace_chunk);
+	session->current_trace_chunk = NULL;
+	if (session->ust_session) {
+		lttng_trace_chunk_put(
+				session->ust_session->current_trace_chunk);
+		session->ust_session->current_trace_chunk = NULL;
+	}
+	if (session->kernel_session) {
+		lttng_trace_chunk_put(
+				session->kernel_session->current_trace_chunk);
+		session->kernel_session->current_trace_chunk = NULL;
+	}
+
+	/*
+	 * Update local current trace chunk state last, only if all remote
+	 * annoucements succeeded.
+	 */
+	session->current_trace_chunk = new_trace_chunk;
+	if (session->ust_session) {
+		session->ust_session->current_trace_chunk = new_trace_chunk;
+	}
+	if (session->kernel_session) {
+		session->kernel_session->current_trace_chunk =
+				new_trace_chunk;
+	}
+
+	return 0;
+error:
+	/*
+	 * Release references taken in the case where all references could not
+	 * be acquired.
+	 */
+	refs_to_release = refs_to_acquire - refs_acquired;
+	for (i = 0; i < refs_to_release; i++) {
+		lttng_trace_chunk_put(new_trace_chunk);
+	}
+
+	/*
+	 * Close the newly-created chunk from remote peers (consumers and
+	 * relayd).
+	 */
+	DBG("Rolling back the creation of the new trace chunk on consumers");
+	for (i = 0; i < consumer_count; i++) {
+		if (!transactions[i].new_chunk_created) {
+			continue;
+		}
+
+		pthread_mutex_lock(transactions[i].socket->lock);
+		ret = consumer_close_trace_chunk(transactions[i].socket,
+				session->consumer->net_seq_index,
+				session->id,
+				transactions[i].new_chunk);
+		pthread_mutex_unlock(transactions[i].socket->lock);
+		if (ret) {
+			ERR("Failed to close trace chunk on consumer");
+			close_error_occured = true;
+		}
+	}
+
+	return -1;
+}
+
+static
+bool output_supports_chunks(const struct ltt_session *session)
+{
+	if (session->consumer->type == CONSUMER_DST_LOCAL) {
+		return true;
+	} else {
+		struct consumer_output *output;
+
+		if (session->ust_session) {
+			output = session->ust_session->consumer;
+		} else if (session->kernel_session) {
+			output = session->kernel_session->consumer;
+		} else {
+			abort();
+		}
+
+		if (output->relay_major_version > 2) {
+			return true;
+		} else if (output->relay_major_version == 2 &&
+				output->relay_minor_version >= 11) {
+			return true;
+		}
+	}
+	return false;
+}
+
+enum lttng_error_code session_switch_trace_chunk(struct ltt_session *session,
+		const char *session_base_path_override,
+		const char *chunk_name_override)
+{
+	int ret;
+	enum lttng_error_code ret_code = LTTNG_OK;
+	struct lttng_trace_chunk *trace_chunk = NULL;
+	enum lttng_trace_chunk_status chunk_status;
+	const time_t timestamp_begin = time(NULL);
+	const bool is_local_trace =
+			session->consumer->type == CONSUMER_DST_LOCAL;
+	const char *base_path = session_base_path_override ? :
+			session_get_base_path(session);
+	struct lttng_directory_handle session_output_directory;
+	const struct lttng_credentials session_credentials = {
+		.uid = session->uid,
+		.gid = session->gid,
+	};
+	uint64_t next_chunk_id;
+
+	if (timestamp_begin == (time_t) -1) {
+		PERROR("Failed to sample time while changing session \"%s\" trace chunk",
+				session->name);
+		ret_code = LTTNG_ERR_FATAL;
+		goto error;
+	}
+	session->current_chunk_start_ts = timestamp_begin;
+
+	if (!output_supports_chunks(session)) {
+		goto end;
+	}
+	next_chunk_id = session->last_trace_chunk_id.is_set ?
+			session->last_trace_chunk_id.value + 1 : 0;
+
+	trace_chunk = lttng_trace_chunk_create(next_chunk_id, timestamp_begin);
+	if (!trace_chunk) {
+		ret_code = LTTNG_ERR_FATAL;
+		goto error;
+	}
+
+	if (chunk_name_override) {
+		chunk_status = lttng_trace_chunk_override_name(trace_chunk,
+				chunk_name_override);
+		switch (chunk_status) {
+		case LTTNG_TRACE_CHUNK_STATUS_OK:
+			break;
+		case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
+			ret_code = LTTNG_ERR_INVALID;
+			goto error;
+		default:
+			ret_code = LTTNG_ERR_NOMEM;
+			goto error;
+		}
+	}
+
+	if (!is_local_trace) {
+		/*
+		 * No need to set crendentials and output directory
+		 * for remote trace chunks.
+		 */
+		goto publish;
+	}
+
+	chunk_status = lttng_trace_chunk_set_credentials(trace_chunk,
+			&session_credentials);
+	if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+		ret_code = LTTNG_ERR_FATAL;
+		goto error;
+	}
+
+	if (!session->current_trace_chunk) {
+		DBG("Creating base output directory of session \"%s\" at %s",
+				session->name, base_path);
+	}
+	ret = utils_mkdir_recursive(base_path, S_IRWXU | S_IRWXG,
+			session->uid, session->gid);
+	if (ret) {
+		ret = LTTNG_ERR_FATAL;
+		goto error;
+	}
+	ret = lttng_directory_handle_init(&session_output_directory,
+			base_path);
+	if (ret) {
+		ret = LTTNG_ERR_FATAL;
+		goto error;
+	}
+	chunk_status = lttng_trace_chunk_set_as_owner(trace_chunk,
+			&session_output_directory);
+	lttng_directory_handle_fini(&session_output_directory);
+	if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+		ret = LTTNG_ERR_CREATE_DIR_FAIL;
+		goto error;
+	}
+publish:
+	ret = session_set_trace_chunk(session, trace_chunk);
+	if (ret) {
+		ret_code = LTTNG_ERR_FATAL;
+		goto error;
+	}
+error:
+	lttng_trace_chunk_put(trace_chunk);
+end:
+	return ret_code;
+}
+
+/*
+ * Set a session's current trace chunk.
+ *
+ * Must be called with the session lock held.
+ */
+int session_set_trace_chunk(struct ltt_session *session,
+		struct lttng_trace_chunk *new_trace_chunk)
+{
+	ASSERT_LOCKED(session->lock);
+	return _session_set_trace_chunk_no_lock_check(session, new_trace_chunk);
+}
+
+static
+void session_release(struct urcu_ref *ref)
+{
+	int ret;
+	struct ltt_ust_session *usess;
+	struct ltt_kernel_session *ksess;
+	struct ltt_session *session = container_of(ref, typeof(*session), ref);
+
+	usess = session->ust_session;
+	ksess = session->kernel_session;
+	(void) _session_set_trace_chunk_no_lock_check(session, NULL);
+
+	/* Clean kernel session teardown */
+	kernel_destroy_session(ksess);
+	session->kernel_session = NULL;
+
+	/* UST session teardown */
+	if (usess) {
+		/* Close any relayd session */
+		consumer_output_send_destroy_relayd(usess->consumer);
+
+		/* Destroy every UST application related to this session. */
+		ret = ust_app_destroy_trace_all(usess);
+		if (ret) {
+			ERR("Error in ust_app_destroy_trace_all");
+		}
+
+		/* Clean up the rest. */
+		trace_ust_destroy_session(usess);
+		session->ust_session = NULL;
+	}
+
+	/*
+	 * Must notify the kernel thread here to update it's poll set in order to
+	 * remove the channel(s)' fd just destroyed.
+	 */
+	ret = notify_thread_pipe(kernel_poll_pipe[1]);
+	if (ret < 0) {
+		PERROR("write kernel poll pipe");
+	}
+
+	DBG("Destroying session %s (id %" PRIu64 ")", session->name, session->id);
+
+	consumer_output_put(session->consumer);
+	snapshot_destroy(&session->snapshot);
+
+	pthread_mutex_destroy(&session->lock);
+
+	if (session->published) {
+		ASSERT_LOCKED(ltt_session_list.lock);
+		del_session_list(session);
+		del_session_ht(session);
+		pthread_cond_broadcast(&ltt_session_list.removal_cond);
+	}
+	free(session);
+}
+
+/*
+ * Acquire a reference to a session.
+ * This function may fail (return false); its return value must be checked.
+ */
+bool session_get(struct ltt_session *session)
+{
+	return urcu_ref_get_unless_zero(&session->ref);
+}
+
+/*
+ * Release a reference to a session.
+ */
+void session_put(struct ltt_session *session)
+{
+	if (!session) {
+		return;
+	}
+	/*
+	 * The session list lock must be held as any session_put()
+	 * may cause the removal of the session from the session_list.
+	 */
+	ASSERT_LOCKED(ltt_session_list.lock);
+	assert(session->ref.refcount);
+	urcu_ref_put(&session->ref, session_release);
+}
+
+/*
+ * Destroy a session.
+ *
+ * This method does not immediately release/free the session as other
+ * components may still hold a reference to the session. However,
+ * the session should no longer be presented to the user.
+ *
+ * Releases the session list's reference to the session
+ * and marks it as destroyed. Iterations on the session list should be
+ * mindful of the "destroyed" flag.
+ */
+void session_destroy(struct ltt_session *session)
+{
+	assert(!session->destroyed);
+	session->destroyed = true;
+	session_put(session);
+}
+
 /*
  * Return a ltt_session structure ptr that matches name. If no session found,
  * NULL is returned. This must be called with the session list lock held using
  * session_lock_list and session_unlock_list.
+ * A reference to the session is implicitly acquired by this function.
  */
 struct ltt_session *session_find_by_name(const char *name)
 {
 	struct ltt_session *iter;
 
 	assert(name);
+	ASSERT_LOCKED(ltt_session_list.lock);
 
 	DBG2("Trying to find session by name %s", name);
 
 	cds_list_for_each_entry(iter, &ltt_session_list.head, list) {
-		if (strncmp(iter->name, name, NAME_MAX) == 0) {
+		if (!strncmp(iter->name, name, NAME_MAX) &&
+				!iter->destroyed) {
 			goto found;
 		}
 	}
 
-	iter = NULL;
-
+	return NULL;
 found:
-	return iter;
+	return session_get(iter) ? iter : NULL;
 }
 
 /*
@@ -298,107 +870,166 @@ struct ltt_session *session_find_by_id(uint64_t id)
 	struct lttng_ht_iter iter;
 	struct ltt_session *ls;
 
+	ASSERT_LOCKED(ltt_session_list.lock);
+
+	if (!ltt_sessions_ht_by_id) {
+		goto end;
+	}
+
 	lttng_ht_lookup(ltt_sessions_ht_by_id, &id, &iter);
 	node = lttng_ht_iter_get_node_u64(&iter);
 	if (node == NULL) {
-		goto error;
+		goto end;
 	}
 	ls = caa_container_of(node, struct ltt_session, node);
 
 	DBG3("Session %" PRIu64 " found by id.", id);
-	return ls;
+	return session_get(ls) ? ls : NULL;
 
-error:
+end:
 	DBG3("Session %" PRIu64 " NOT found by id", id);
 	return NULL;
 }
 
 /*
- * Delete session from the session list and free the memory.
- *
- * Return -1 if no session is found.  On success, return 1;
- * Should *NOT* be called with RCU read-side lock held.
+ * Create a new session and add it to the session list.
+ * Session list lock must be held by the caller.
  */
-int session_destroy(struct ltt_session *session)
-{
-	/* Safety check */
-	assert(session);
-
-	DBG("Destroying session %s", session->name);
-	del_session_list(session);
-	pthread_mutex_destroy(&session->lock);
-	del_session_ht(session);
-
-	consumer_output_put(session->consumer);
-	snapshot_destroy(&session->snapshot);
-	free(session);
-
-	return LTTNG_OK;
-}
-
-/*
- * Create a brand new session and add it to the session list.
- */
-int session_create(char *name, uid_t uid, gid_t gid)
+enum lttng_error_code session_create(const char *name, uid_t uid, gid_t gid,
+		struct ltt_session **out_session)
 {
 	int ret;
-	struct ltt_session *new_session;
+	enum lttng_error_code ret_code;
+	struct ltt_session *new_session = NULL;
+
+	ASSERT_LOCKED(ltt_session_list.lock);
+	if (name) {
+		struct ltt_session *clashing_session;
 
-	/* Allocate session data structure */
+		clashing_session = session_find_by_name(name);
+		if (clashing_session) {
+			session_put(clashing_session);
+			ret_code = LTTNG_ERR_EXIST_SESS;
+			goto error;
+		}
+	}
 	new_session = zmalloc(sizeof(struct ltt_session));
-	if (new_session == NULL) {
-		PERROR("zmalloc");
-		ret = LTTNG_ERR_FATAL;
-		goto error_malloc;
+	if (!new_session) {
+		PERROR("Failed to allocate an ltt_session structure");
+		ret_code = LTTNG_ERR_NOMEM;
+		goto error;
 	}
 
-	/* Define session name */
-	if (name != NULL) {
-		if (snprintf(new_session->name, NAME_MAX, "%s", name) < 0) {
-			ret = LTTNG_ERR_FATAL;
-			goto error_asprintf;
-		}
-	} else {
-		ERR("No session name given");
-		ret = LTTNG_ERR_FATAL;
+	urcu_ref_init(&new_session->ref);
+	pthread_mutex_init(&new_session->lock, NULL);
+
+	new_session->creation_time = time(NULL);
+	if (new_session->creation_time == (time_t) -1) {
+		PERROR("Failed to sample session creation time");
+		ret_code = LTTNG_ERR_SESSION_FAIL;
 		goto error;
 	}
 
-	ret = validate_name(name);
-	if (ret < 0) {
-		ret = LTTNG_ERR_SESSION_INVALID_CHAR;
+	/* Create default consumer output. */
+	new_session->consumer = consumer_create_output(CONSUMER_DST_LOCAL);
+	if (new_session->consumer == NULL) {
+		ret_code = LTTNG_ERR_NOMEM;
 		goto error;
 	}
 
+	if (name) {
+		ret = lttng_strncpy(new_session->name, name, sizeof(new_session->name));
+		if (ret) {
+			ret_code = LTTNG_ERR_SESSION_INVALID_CHAR;
+			goto error;
+		}
+		ret = validate_name(name);
+		if (ret < 0) {
+			ret_code = LTTNG_ERR_SESSION_INVALID_CHAR;
+			goto error;
+		}
+	} else {
+		int i = 0;
+		bool found_name = false;
+		char datetime[16];
+		struct tm *timeinfo;
+
+		timeinfo = localtime(&new_session->creation_time);
+		if (!timeinfo) {
+			ret_code = LTTNG_ERR_SESSION_FAIL;
+			goto error;
+		}
+		strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", timeinfo);
+		for (i = 0; i < INT_MAX; i++) {
+			struct ltt_session *clashing_session;
+
+			if (i == 0) {
+				ret = snprintf(new_session->name,
+						sizeof(new_session->name),
+						"%s-%s",
+						DEFAULT_SESSION_NAME,
+						datetime);
+			} else {
+				ret = snprintf(new_session->name,
+						sizeof(new_session->name),
+						"%s%d-%s",
+						DEFAULT_SESSION_NAME, i,
+						datetime);
+			}
+			if (ret == -1 || ret >= sizeof(new_session->name)) {
+				/*
+				 * Null-terminate in case the name is used
+				 * in logging statements.
+				 */
+				new_session->name[sizeof(new_session->name) - 1] = '\0';
+				ret_code = LTTNG_ERR_SESSION_FAIL;
+				goto error;
+			}
+
+			clashing_session =
+					session_find_by_name(new_session->name);
+			session_put(clashing_session);
+			if (!clashing_session) {
+				found_name = true;
+				break;
+			}
+		}
+		if (found_name) {
+			DBG("Generated session name \"%s\"", new_session->name);
+			new_session->has_auto_generated_name = true;
+		} else {
+			ERR("Failed to auto-generate a session name");
+			ret_code = LTTNG_ERR_SESSION_FAIL;
+			goto error;
+		}
+	}
+
 	ret = gethostname(new_session->hostname, sizeof(new_session->hostname));
 	if (ret < 0) {
 		if (errno == ENAMETOOLONG) {
 			new_session->hostname[sizeof(new_session->hostname) - 1] = '\0';
+			ERR("Hostname exceeds the maximal permitted length and has been truncated to %s",
+					new_session->hostname);
 		} else {
-			ret = LTTNG_ERR_FATAL;
+			ret_code = LTTNG_ERR_SESSION_FAIL;
 			goto error;
 		}
 	}
 
-	/* Init kernel session */
-	new_session->kernel_session = NULL;
-	new_session->ust_session = NULL;
-
-	/* Init lock */
-	pthread_mutex_init(&new_session->lock, NULL);
-
 	new_session->uid = uid;
 	new_session->gid = gid;
 
 	ret = snapshot_init(&new_session->snapshot);
 	if (ret < 0) {
-		ret = LTTNG_ERR_NOMEM;
+		ret_code = LTTNG_ERR_NOMEM;
 		goto error;
 	}
 
-	/* Add new session to the session list */
-	session_lock_list();
+	new_session->rotation_state = LTTNG_ROTATION_STATE_NO_ROTATION;
+
+	/* Add new session to the session list. */
 	new_session->id = add_session_list(new_session);
+
 	/*
 	 * Add the new session to the ltt_sessions_ht_by_id.
 	 * No ownership is taken by the hash table; it is merely
@@ -406,23 +1037,26 @@ int session_create(char *name, uid_t uid, gid_t gid)
 	 * by session id.
 	 */
 	add_session_ht(new_session);
-	session_unlock_list();
+	new_session->published = true;
 
 	/*
-	 * Consumer is let to NULL since the create_session_uri command will set it
-	 * up and, if valid, assign it to the session.
+	 * Consumer is left to NULL since the create_session_uri command will
+	 * set it up and, if valid, assign it to the session.
 	 */
-	DBG("Tracing session %s created with ID %" PRIu64 " by UID %d GID %d",
-			name, new_session->id, new_session->uid, new_session->gid);
-
-	return LTTNG_OK;
-
+	DBG("Tracing session %s created with ID %" PRIu64 " by uid = %d, gid = %d",
+			new_session->name, new_session->id, new_session->uid,
+			new_session->gid);
+	ret_code = LTTNG_OK;
+end:
+	if (new_session) {
+		(void) session_get(new_session);
+		*out_session = new_session;
+	}
+	return ret_code;
 error:
-error_asprintf:
-	free(new_session);
-
-error_malloc:
-	return ret;
+	session_put(new_session);
+	new_session = NULL;
+	goto end;
 }
 
 /*
@@ -439,3 +1073,36 @@ int session_access_ok(struct ltt_session *session, uid_t uid, gid_t gid)
 		return 1;
 	}
 }
+
+/*
+ * Set a session's rotation state and reset all associated state.
+ *
+ * This function resets the rotation state (check timers, pending
+ * flags, etc.) and sets the result of the last rotation. The result
+ * can be queries by a liblttng-ctl client.
+ *
+ * Be careful of the result passed to this function. For instance,
+ * on failure to launch a rotation, a client will expect the rotation
+ * state to be set to "NO_ROTATION". If an error occured while the
+ * rotation was "ONGOING", result should be set to "ERROR", which will
+ * allow a client to report it.
+ *
+ * Must be called with the session and session_list locks held.
+ */
+int session_reset_rotation_state(struct ltt_session *session,
+		enum lttng_rotation_state result)
+{
+	int ret = 0;
+
+	ASSERT_LOCKED(ltt_session_list.lock);
+	ASSERT_LOCKED(session->lock);
+
+	session->rotation_pending_local = false;
+	session->rotation_pending_relay = false;
+	session->rotated_after_last_stop = false;
+	session->rotation_state = result;
+	if (session->rotation_pending_check_timer_enabled) {
+		ret = timer_session_rotation_pending_check_stop(session);
+	}
+	return ret;
+}