+/*
+ * relay_send_version: send relayd version number
+ */
+static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct lttcomm_relayd_version reply, msg;
+ bool compatible = true;
+
+ conn->version_check_done = true;
+
+ /* Get version from the other side. */
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end;
+ }
+
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.major = be32toh(msg.major);
+ msg.minor = be32toh(msg.minor);
+
+ memset(&reply, 0, sizeof(reply));
+ reply.major = RELAYD_VERSION_COMM_MAJOR;
+ reply.minor = RELAYD_VERSION_COMM_MINOR;
+
+ /* Major versions must be the same */
+ if (reply.major != msg.major) {
+ DBG("Incompatible major versions (%u vs %u), deleting session",
+ reply.major, msg.major);
+ compatible = false;
+ }
+
+ conn->major = reply.major;
+ /* We adapt to the lowest compatible version */
+ if (reply.minor <= msg.minor) {
+ conn->minor = reply.minor;
+ } else {
+ conn->minor = msg.minor;
+ }
+
+ reply.major = htobe32(reply.major);
+ reply.minor = htobe32(reply.minor);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"send version\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ goto end;
+ } else {
+ ret = 0;
+ }
+
+ if (!compatible) {
+ ret = -1;
+ goto end;
+ }
+
+ DBG("Version check done using protocol %u.%u", conn->major,
+ conn->minor);
+
+end:
+ return ret;
+}
+
+/*
+ * Check for data pending for a given stream id from the session daemon.
+ */
+static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_data_pending msg;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+ ssize_t send_ret;
+ int ret;
+ uint64_t stream_seq;
+
+ DBG("Data pending command received");
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end_no_session;
+ }
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.stream_id = be64toh(msg.stream_id);
+ msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
+
+ stream = stream_get_by_id(msg.stream_id);
+ if (stream == NULL) {
+ ret = -1;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+
+ if (session_streams_have_index(session)) {
+ /*
+ * Ensure that both the index and stream data have been
+ * flushed up to the requested point.
+ */
+ stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
+ } else {
+ stream_seq = stream->prev_data_seq;
+ }
+ DBG("Data pending for stream id %" PRIu64 ": prev_data_seq %" PRIu64
+ ", prev_index_seq %" PRIu64
+ ", and last_seq %" PRIu64, msg.stream_id,
+ stream->prev_data_seq, stream->prev_index_seq,
+ msg.last_net_seq_num);
+
+ /* Avoid wrapping issue */
+ if (((int64_t) (stream_seq - msg.last_net_seq_num)) >= 0) {
+ /* Data has in fact been written and is NOT pending */
+ ret = 0;
+ } else {
+ /* Data still being streamed thus pending */
+ ret = 1;
+ }
+
+ stream->data_pending_check_done = true;
+ pthread_mutex_unlock(&stream->lock);
+
+ stream_put(stream);
+end:
+
+ memset(&reply, 0, sizeof(reply));
+ reply.ret_code = htobe32(ret);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * Wait for the control socket to reach a quiescent state.
+ *
+ * Note that for now, when receiving this command from the session
+ * daemon, this means that every subsequent commands or data received on
+ * the control socket has been handled. So, this is why we simply return
+ * OK here.
+ */
+static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct relay_stream *stream;
+ struct lttcomm_relayd_quiescent_control msg;
+ struct lttcomm_relayd_generic_reply reply;
+
+ DBG("Checking quiescent state on control socket");
+
+ if (!conn->session || !conn->version_check_done) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end_no_session;
+ }
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.stream_id = be64toh(msg.stream_id);
+
+ stream = stream_get_by_id(msg.stream_id);
+ if (!stream) {
+ goto reply;
+ }
+ pthread_mutex_lock(&stream->lock);
+ stream->data_pending_check_done = true;
+ pthread_mutex_unlock(&stream->lock);
+
+ DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
+ stream_put(stream);
+reply:
+ memset(&reply, 0, sizeof(reply));
+ reply.ret_code = htobe32(LTTNG_OK);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * Initialize a data pending command. This means that a consumer is about
+ * to ask for data pending for each stream it holds. Simply iterate over
+ * all streams of a session and set the data_pending_check_done flag.
+ *
+ * This command returns to the client a LTTNG_OK code.
+ */
+static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct lttng_ht_iter iter;
+ struct lttcomm_relayd_begin_data_pending msg;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+
+ assert(recv_hdr);
+ assert(conn);
+
+ DBG("Init streams for data pending");
+
+ if (!conn->session || !conn->version_check_done) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end_no_session;
+ }
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.session_id = be64toh(msg.session_id);
+
+ /*
+ * Iterate over all streams to set the begin data pending flag.
+ * For now, the streams are indexed by stream handle so we have
+ * to iterate over all streams to find the one associated with
+ * the right session_id.
+ */
+ rcu_read_lock();
+ cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
+ node.node) {
+ if (!stream_get(stream)) {
+ continue;
+ }
+ if (stream->trace->session->id == msg.session_id) {
+ pthread_mutex_lock(&stream->lock);
+ stream->data_pending_check_done = false;
+ pthread_mutex_unlock(&stream->lock);
+ DBG("Set begin data pending flag to stream %" PRIu64,
+ stream->stream_handle);
+ }
+ stream_put(stream);
+ }
+ rcu_read_unlock();
+
+ memset(&reply, 0, sizeof(reply));
+ /* All good, send back reply. */
+ reply.ret_code = htobe32(LTTNG_OK);
+
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * End data pending command. This will check, for a given session id, if
+ * each stream associated with it has its data_pending_check_done flag
+ * set. If not, this means that the client lost track of the stream but
+ * the data is still being streamed on our side. In this case, we inform
+ * the client that data is in flight.
+ *
+ * Return to the client if there is data in flight or not with a ret_code.
+ */
+static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct lttng_ht_iter iter;
+ struct lttcomm_relayd_end_data_pending msg;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+ uint32_t is_data_inflight = 0;
+
+ DBG("End data pending command");
+
+ if (!conn->session || !conn->version_check_done) {
+ ERR("Trying to check for data before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ if (payload->size < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end_no_session;
+ }
+ memcpy(&msg, payload->data, sizeof(msg));
+ msg.session_id = be64toh(msg.session_id);
+
+ /*
+ * Iterate over all streams to see if the begin data pending
+ * flag is set.
+ */
+ rcu_read_lock();
+ cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
+ node.node) {
+ if (!stream_get(stream)) {
+ continue;
+ }
+ if (stream->trace->session->id != msg.session_id) {
+ stream_put(stream);
+ continue;
+ }
+ pthread_mutex_lock(&stream->lock);
+ if (!stream->data_pending_check_done) {
+ uint64_t stream_seq;
+
+ if (session_streams_have_index(conn->session)) {
+ /*
+ * Ensure that both the index and stream data have been
+ * flushed up to the requested point.
+ */
+ stream_seq = min(stream->prev_data_seq, stream->prev_index_seq);
+ } else {
+ stream_seq = stream->prev_data_seq;
+ }
+ if (!stream->closed || !(((int64_t) (stream_seq - stream->last_net_seq_num)) >= 0)) {
+ is_data_inflight = 1;
+ DBG("Data is still in flight for stream %" PRIu64,
+ stream->stream_handle);
+ pthread_mutex_unlock(&stream->lock);
+ stream_put(stream);
+ break;
+ }
+ }
+ pthread_mutex_unlock(&stream->lock);
+ stream_put(stream);
+ }
+ rcu_read_unlock();
+
+ memset(&reply, 0, sizeof(reply));
+ /* All good, send back reply. */
+ reply.ret_code = htobe32(is_data_inflight);
+
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ ret = 0;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * Receive an index for a specific stream.
+ *
+ * Return 0 on success else a negative value.
+ */
+static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_index index_info;
+ struct lttcomm_relayd_generic_reply reply;
+ struct relay_stream *stream;
+ size_t msg_len;
+
+ assert(conn);
+
+ DBG("Relay receiving index");
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to close a stream before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ msg_len = lttcomm_relayd_index_len(
+ lttng_to_index_major(conn->major, conn->minor),
+ lttng_to_index_minor(conn->major, conn->minor));
+ if (payload->size < msg_len) {
+ ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
+ msg_len, payload->size);
+ ret = -1;
+ goto end_no_session;
+ }
+ memcpy(&index_info, payload->data, msg_len);
+ index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
+ index_info.net_seq_num = be64toh(index_info.net_seq_num);
+ index_info.packet_size = be64toh(index_info.packet_size);
+ index_info.content_size = be64toh(index_info.content_size);
+ index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
+ index_info.timestamp_end = be64toh(index_info.timestamp_end);
+ index_info.events_discarded = be64toh(index_info.events_discarded);
+ index_info.stream_id = be64toh(index_info.stream_id);
+
+ if (conn->minor >= 8) {
+ index_info.stream_instance_id =
+ be64toh(index_info.stream_instance_id);
+ index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
+ }
+
+ stream = stream_get_by_id(index_info.relay_stream_id);
+ if (!stream) {
+ ERR("stream_get_by_id not found");
+ ret = -1;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+ ret = stream_add_index(stream, &index_info);
+ pthread_mutex_unlock(&stream->lock);
+ if (ret) {
+ goto end_stream_put;
+ }
+
+end_stream_put:
+ stream_put(stream);
+end:
+ memset(&reply, 0, sizeof(reply));
+ if (ret < 0) {
+ reply.ret_code = htobe32(LTTNG_ERR_UNK);
+ } else {
+ reply.ret_code = htobe32(LTTNG_OK);
+ }
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
+ ret = -1;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * Receive the streams_sent message.
+ *
+ * Return 0 on success else a negative value.
+ */
+static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret;
+ ssize_t send_ret;
+ struct lttcomm_relayd_generic_reply reply;
+
+ assert(conn);
+
+ DBG("Relay receiving streams_sent");
+
+ if (!conn->session || !conn->version_check_done) {
+ ERR("Trying to close a stream before version check");
+ ret = -1;
+ goto end_no_session;
+ }
+
+ /*
+ * Publish every pending stream in the connection recv list which are
+ * now ready to be used by the viewer.
+ */
+ publish_connection_local_streams(conn);
+
+ memset(&reply, 0, sizeof(reply));
+ reply.ret_code = htobe32(LTTNG_OK);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ } else {
+ /* Success. */
+ ret = 0;
+ }
+
+end_no_session:
+ return ret;
+}
+
+/*
+ * relay_rotate_session_stream: rotate a stream to a new tracefile for the
+ * session rotation feature (not the tracefile rotation feature).
+ */
+static int relay_rotate_session_streams(
+ const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ uint32_t i;
+ ssize_t send_ret;
+ enum lttng_error_code reply_code = LTTNG_ERR_UNK;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_rotate_streams rotate_streams;
+ struct lttcomm_relayd_generic_reply reply = {};
+ struct relay_stream *stream = NULL;
+ const size_t header_len = sizeof(struct lttcomm_relayd_rotate_streams);
+ struct lttng_trace_chunk *next_trace_chunk = NULL;
+ struct lttng_buffer_view stream_positions;
+ char chunk_id_buf[MAX_INT_DEC_LEN(uint64_t)];
+ const char *chunk_id_str = "none";
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to rotate a stream before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Unsupported feature before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (payload->size < header_len) {
+ ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
+ header_len, payload->size);
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ memcpy(&rotate_streams, payload->data, header_len);
+
+ /* Convert header to host endianness. */
+ rotate_streams = (typeof(rotate_streams)) {
+ .stream_count = be32toh(rotate_streams.stream_count),
+ .new_chunk_id = (typeof(rotate_streams.new_chunk_id)) {
+ .is_set = !!rotate_streams.new_chunk_id.is_set,
+ .value = be64toh(rotate_streams.new_chunk_id.value),
+ }
+ };
+
+ if (rotate_streams.new_chunk_id.is_set) {
+ /*
+ * Retrieve the trace chunk the stream must transition to. As
+ * per the protocol, this chunk should have been created
+ * before this command is received.
+ */
+ next_trace_chunk = sessiond_trace_chunk_registry_get_chunk(
+ sessiond_trace_chunk_registry,
+ session->sessiond_uuid, session->id,
+ rotate_streams.new_chunk_id.value);
+ if (!next_trace_chunk) {
+ char uuid_str[UUID_STR_LEN];
+
+ lttng_uuid_to_str(session->sessiond_uuid, uuid_str);
+ ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
+ ", trace_chunk_id = %" PRIu64,
+ uuid_str, session->id,
+ rotate_streams.new_chunk_id.value);
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end;
+ }
+
+ ret = snprintf(chunk_id_buf, sizeof(chunk_id_buf), "%" PRIu64,
+ rotate_streams.new_chunk_id.value);
+ if (ret < 0 || ret >= sizeof(chunk_id_buf)) {
+ chunk_id_str = "formatting error";
+ } else {
+ chunk_id_str = chunk_id_buf;
+ }
+ session->has_rotated = true;
+ }
+
+ DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
+ rotate_streams.stream_count, session->session_name,
+ chunk_id_str);
+
+ stream_positions = lttng_buffer_view_from_view(payload,
+ sizeof(rotate_streams), -1);
+ if (!stream_positions.data ||
+ stream_positions.size <
+ (rotate_streams.stream_count *
+ sizeof(struct lttcomm_relayd_stream_rotation_position))) {
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end;
+ }
+
+ for (i = 0; i < rotate_streams.stream_count; i++) {
+ struct lttcomm_relayd_stream_rotation_position *position_comm =
+ &((typeof(position_comm)) stream_positions.data)[i];
+ const struct lttcomm_relayd_stream_rotation_position pos = {
+ .stream_id = be64toh(position_comm->stream_id),
+ .rotate_at_seq_num = be64toh(
+ position_comm->rotate_at_seq_num),
+ };
+
+ stream = stream_get_by_id(pos.stream_id);
+ if (!stream) {
+ reply_code = LTTNG_ERR_INVALID;
+ ret = -1;
+ goto end;
+ }
+
+ pthread_mutex_lock(&stream->lock);
+ ret = stream_set_pending_rotation(stream, next_trace_chunk,
+ pos.rotate_at_seq_num);
+ pthread_mutex_unlock(&stream->lock);
+ if (ret) {
+ reply_code = LTTNG_ERR_FILE_CREATION_ERROR;
+ goto end;
+ }
+
+ stream_put(stream);
+ stream = NULL;
+ }
+
+ reply_code = LTTNG_OK;
+end:
+ if (stream) {
+ stream_put(stream);
+ }
+
+ reply.ret_code = htobe32((uint32_t) reply_code);
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(struct lttcomm_relayd_generic_reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+
+ ret = 0;
+end_no_reply:
+ lttng_trace_chunk_put(next_trace_chunk);
+ return ret;
+}
+
+static int init_session_output_directory_handle(struct relay_session *session,
+ struct lttng_directory_handle *handle)
+{
+ int ret;
+ /*
+ * relayd_output_path/session_directory
+ * e.g. /home/user/lttng-traces/hostname/session_name
+ */
+ char *full_session_path = NULL;
+
+ pthread_mutex_lock(&session->lock);
+ full_session_path = create_output_path(session->output_path);
+ if (!full_session_path) {
+ ret = -1;
+ goto end;
+ }
+
+ ret = utils_mkdir_recursive(
+ full_session_path, S_IRWXU | S_IRWXG, -1, -1);
+ if (ret) {
+ ERR("Failed to create session output path \"%s\"",
+ full_session_path);
+ goto end;
+ }
+
+ ret = lttng_directory_handle_init(handle, full_session_path);
+ if (ret) {
+ goto end;
+ }
+end:
+ pthread_mutex_unlock(&session->lock);
+ free(full_session_path);
+ return ret;
+}
+
+/*
+ * relay_create_trace_chunk: create a new trace chunk
+ */
+static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_create_trace_chunk *msg;
+ struct lttcomm_relayd_generic_reply reply = {};
+ struct lttng_buffer_view header_view;
+ struct lttng_buffer_view chunk_name_view;
+ struct lttng_trace_chunk *chunk = NULL, *published_chunk = NULL;
+ enum lttng_error_code reply_code = LTTNG_OK;
+ enum lttng_trace_chunk_status chunk_status;
+ struct lttng_directory_handle session_output;
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to create a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk creation command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk creation command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ msg->chunk_id = be64toh(msg->chunk_id);
+ msg->creation_timestamp = be64toh(msg->creation_timestamp);
+ msg->override_name_length = be32toh(msg->override_name_length);
+
+ chunk = lttng_trace_chunk_create(
+ msg->chunk_id, msg->creation_timestamp);
+ if (!chunk) {
+ ERR("Failed to create trace chunk in trace chunk creation command");
+ ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ if (msg->override_name_length) {
+ const char *name;
+
+ chunk_name_view = lttng_buffer_view_from_view(payload,
+ sizeof(*msg),
+ msg->override_name_length);
+ name = chunk_name_view.data;
+ if (!name || name[msg->override_name_length - 1]) {
+ ERR("Failed to receive payload of chunk creation command");
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID;
+ goto end;
+ }
+
+ chunk_status = lttng_trace_chunk_override_name(
+ chunk, chunk_name_view.data);
+ switch (chunk_status) {
+ case LTTNG_TRACE_CHUNK_STATUS_OK:
+ break;
+ case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT:
+ ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
+ reply_code = LTTNG_ERR_INVALID;
+ ret = -1;
+ goto end;
+ default:
+ ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
+ reply_code = LTTNG_ERR_UNK;
+ ret = -1;
+ goto end;
+ }
+ }
+
+ ret = init_session_output_directory_handle(
+ conn->session, &session_output);
+ if (ret) {
+ reply_code = LTTNG_ERR_CREATE_DIR_FAIL;
+ goto end;
+ }
+
+ chunk_status = lttng_trace_chunk_set_credentials_current_user(chunk);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ reply_code = LTTNG_ERR_UNK;
+ ret = -1;
+ goto end;
+ }
+
+ chunk_status = lttng_trace_chunk_set_as_owner(chunk, &session_output);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ reply_code = LTTNG_ERR_UNK;
+ ret = -1;
+ goto end;
+ }
+
+ published_chunk = sessiond_trace_chunk_registry_publish_chunk(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk);
+ if (!published_chunk) {
+ char uuid_str[UUID_STR_LEN];
+
+ lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
+ ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
+ uuid_str,
+ conn->session->id,
+ msg->chunk_id);
+ ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ pthread_mutex_lock(&conn->session->lock);
+ if (conn->session->pending_closure_trace_chunk) {
+ /*
+ * Invalid; this means a second create_trace_chunk command was
+ * received before a close_trace_chunk.
+ */
+ ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end_unlock_session;
+ }
+ conn->session->pending_closure_trace_chunk =
+ conn->session->current_trace_chunk;
+ conn->session->current_trace_chunk = published_chunk;
+ published_chunk = NULL;
+end_unlock_session:
+ pthread_mutex_unlock(&conn->session->lock);
+end:
+ reply.ret_code = htobe32((uint32_t) reply_code);
+ send_ret = conn->sock->ops->sendmsg(conn->sock,
+ &reply,
+ sizeof(struct lttcomm_relayd_generic_reply),
+ 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ lttng_trace_chunk_put(chunk);
+ lttng_trace_chunk_put(published_chunk);
+ lttng_directory_handle_fini(&session_output);
+ return ret;
+}
+
+/*
+ * relay_close_trace_chunk: close a trace chunk
+ */
+static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_close_trace_chunk *msg;
+ struct lttcomm_relayd_close_trace_chunk_reply reply = {};
+ struct lttng_buffer_view header_view;
+ struct lttng_trace_chunk *chunk = NULL;
+ enum lttng_error_code reply_code = LTTNG_OK;
+ enum lttng_trace_chunk_status chunk_status;
+ uint64_t chunk_id;
+ LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type) close_command = {};
+ time_t close_timestamp;
+ char closed_trace_chunk_path[LTTNG_PATH_MAX];
+ size_t path_length = 0;
+ const char *chunk_name = NULL;
+ struct lttng_dynamic_buffer reply_payload;
+
+ lttng_dynamic_buffer_init(&reply_payload);
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to close a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk close command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk close command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ chunk_id = be64toh(msg->chunk_id);
+ close_timestamp = (time_t) be64toh(msg->close_timestamp);
+ close_command = (typeof(close_command)){
+ .value = be32toh(msg->close_command.value),
+ .is_set = msg->close_command.is_set,
+ };
+
+ chunk = sessiond_trace_chunk_registry_get_chunk(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk_id);
+ if (!chunk) {
+ char uuid_str[UUID_STR_LEN];
+
+ lttng_uuid_to_str(conn->session->sessiond_uuid, uuid_str);
+ ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64 ", chunk_id = %" PRIu64,
+ uuid_str,
+ conn->session->id,
+ msg->chunk_id);
+ ret = -1;
+ reply_code = LTTNG_ERR_NOMEM;
+ goto end;
+ }
+
+ pthread_mutex_lock(&session->lock);
+ if (session->pending_closure_trace_chunk &&
+ session->pending_closure_trace_chunk != chunk) {
+ ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
+ session->session_name);
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ ret = -1;
+ goto end_unlock_session;
+ }
+
+ chunk_status = lttng_trace_chunk_set_close_timestamp(
+ chunk, close_timestamp);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ERR("Failed to set trace chunk close timestamp");
+ ret = -1;
+ reply_code = LTTNG_ERR_UNK;
+ goto end_unlock_session;
+ }
+
+ if (close_command.is_set) {
+ chunk_status = lttng_trace_chunk_set_close_command(
+ chunk, close_command.value);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID;
+ goto end_unlock_session;
+ }
+ }
+ chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name, NULL);
+ if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
+ ERR("Failed to get chunk name");
+ ret = -1;
+ reply_code = LTTNG_ERR_UNK;
+ goto end_unlock_session;
+ }
+ if (!session->has_rotated && !session->snapshot) {
+ ret = lttng_strncpy(closed_trace_chunk_path,
+ session->output_path,
+ sizeof(closed_trace_chunk_path));
+ if (ret) {
+ ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
+ strlen(session->output_path),
+ sizeof(closed_trace_chunk_path));
+ reply_code = LTTNG_ERR_NOMEM;
+ ret = -1;
+ goto end_unlock_session;
+ }
+ } else {
+ if (session->snapshot) {
+ ret = snprintf(closed_trace_chunk_path,
+ sizeof(closed_trace_chunk_path),
+ "%s/%s", session->output_path,
+ chunk_name);
+ } else {
+ ret = snprintf(closed_trace_chunk_path,
+ sizeof(closed_trace_chunk_path),
+ "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
+ "/%s",
+ session->output_path, chunk_name);
+ }
+ if (ret < 0 || ret == sizeof(closed_trace_chunk_path)) {
+ ERR("Failed to format closed trace chunk resulting path");
+ reply_code = ret < 0 ? LTTNG_ERR_UNK : LTTNG_ERR_NOMEM;
+ ret = -1;
+ goto end_unlock_session;
+ }
+ }
+ DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
+ path_length = strlen(closed_trace_chunk_path) + 1;
+ if (path_length > UINT32_MAX) {
+ ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
+ ret = -1;
+ reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+ goto end_unlock_session;
+ }
+
+ if (session->current_trace_chunk == chunk) {
+ /*
+ * After a trace chunk close command, no new streams
+ * referencing the chunk may be created. Hence, on the
+ * event that no new trace chunk have been created for
+ * the session, the reference to the current trace chunk
+ * is released in order to allow it to be reclaimed when
+ * the last stream releases its reference to it.
+ */
+ lttng_trace_chunk_put(session->current_trace_chunk);
+ session->current_trace_chunk = NULL;
+ }
+ lttng_trace_chunk_put(session->pending_closure_trace_chunk);
+ session->pending_closure_trace_chunk = NULL;
+end_unlock_session:
+ pthread_mutex_unlock(&session->lock);
+
+end:
+ reply.generic.ret_code = htobe32((uint32_t) reply_code);
+ reply.path_length = htobe32((uint32_t) path_length);
+ ret = lttng_dynamic_buffer_append(
+ &reply_payload, &reply, sizeof(reply));
+ if (ret) {
+ ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
+ goto end_no_reply;
+ }
+
+ if (reply_code == LTTNG_OK) {
+ ret = lttng_dynamic_buffer_append(&reply_payload,
+ closed_trace_chunk_path, path_length);
+ if (ret) {
+ ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
+ goto end_no_reply;
+ }
+ }
+
+ send_ret = conn->sock->ops->sendmsg(conn->sock,
+ reply_payload.data,
+ reply_payload.size,
+ 0);
+ if (send_ret < reply_payload.size) {
+ ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
+ reply_payload.size, send_ret);
+ ret = -1;
+ goto end_no_reply;
+ }
+end_no_reply:
+ lttng_trace_chunk_put(chunk);
+ lttng_dynamic_buffer_reset(&reply_payload);
+ return ret;
+}
+
+/*
+ * relay_trace_chunk_exists: check if a trace chunk exists
+ */
+static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr *recv_hdr,
+ struct relay_connection *conn,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+ ssize_t send_ret;
+ struct relay_session *session = conn->session;
+ struct lttcomm_relayd_trace_chunk_exists *msg;
+ struct lttcomm_relayd_trace_chunk_exists_reply reply = {};
+ struct lttng_buffer_view header_view;
+ struct lttng_trace_chunk *chunk = NULL;
+ uint64_t chunk_id;
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to close a trace chunk before version check");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ if (session->major == 2 && session->minor < 11) {
+ ERR("Chunk close command is unsupported before 2.11");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ header_view = lttng_buffer_view_from_view(payload, 0, sizeof(*msg));
+ if (!header_view.data) {
+ ERR("Failed to receive payload of chunk close command");
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ /* Convert to host endianness. */
+ msg = (typeof(msg)) header_view.data;
+ chunk_id = be64toh(msg->chunk_id);
+
+ chunk = sessiond_trace_chunk_registry_get_chunk(
+ sessiond_trace_chunk_registry,
+ conn->session->sessiond_uuid,
+ conn->session->id,
+ chunk_id);
+
+ reply = (typeof(reply)) {
+ .generic.ret_code = htobe32((uint32_t) LTTNG_OK),
+ .trace_chunk_exists = !!chunk,
+ };
+ send_ret = conn->sock->ops->sendmsg(conn->sock,
+ &reply, sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+end_no_reply:
+ lttng_trace_chunk_put(chunk);
+ return ret;
+}
+
+#define DBG_CMD(cmd_name, conn) \
+ DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
+
+static int relay_process_control_command(struct relay_connection *conn,
+ const struct lttcomm_relayd_hdr *header,
+ const struct lttng_buffer_view *payload)
+{
+ int ret = 0;
+
+ switch (header->cmd) {
+ case RELAYD_CREATE_SESSION:
+ DBG_CMD("RELAYD_CREATE_SESSION", conn);
+ ret = relay_create_session(header, conn, payload);
+ break;
+ case RELAYD_ADD_STREAM:
+ DBG_CMD("RELAYD_ADD_STREAM", conn);
+ ret = relay_add_stream(header, conn, payload);
+ break;
+ case RELAYD_START_DATA:
+ DBG_CMD("RELAYD_START_DATA", conn);
+ ret = relay_start(header, conn, payload);
+ break;
+ case RELAYD_SEND_METADATA:
+ DBG_CMD("RELAYD_SEND_METADATA", conn);
+ ret = relay_recv_metadata(header, conn, payload);
+ break;
+ case RELAYD_VERSION:
+ DBG_CMD("RELAYD_VERSION", conn);
+ ret = relay_send_version(header, conn, payload);
+ break;
+ case RELAYD_CLOSE_STREAM:
+ DBG_CMD("RELAYD_CLOSE_STREAM", conn);
+ ret = relay_close_stream(header, conn, payload);
+ break;
+ case RELAYD_DATA_PENDING:
+ DBG_CMD("RELAYD_DATA_PENDING", conn);
+ ret = relay_data_pending(header, conn, payload);
+ break;
+ case RELAYD_QUIESCENT_CONTROL:
+ DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
+ ret = relay_quiescent_control(header, conn, payload);
+ break;
+ case RELAYD_BEGIN_DATA_PENDING:
+ DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
+ ret = relay_begin_data_pending(header, conn, payload);
+ break;
+ case RELAYD_END_DATA_PENDING:
+ DBG_CMD("RELAYD_END_DATA_PENDING", conn);
+ ret = relay_end_data_pending(header, conn, payload);
+ break;
+ case RELAYD_SEND_INDEX:
+ DBG_CMD("RELAYD_SEND_INDEX", conn);
+ ret = relay_recv_index(header, conn, payload);
+ break;
+ case RELAYD_STREAMS_SENT:
+ DBG_CMD("RELAYD_STREAMS_SENT", conn);
+ ret = relay_streams_sent(header, conn, payload);
+ break;
+ case RELAYD_RESET_METADATA:
+ DBG_CMD("RELAYD_RESET_METADATA", conn);
+ ret = relay_reset_metadata(header, conn, payload);
+ break;
+ case RELAYD_ROTATE_STREAMS:
+ DBG_CMD("RELAYD_ROTATE_STREAMS", conn);
+ ret = relay_rotate_session_streams(header, conn, payload);
+ break;
+ case RELAYD_CREATE_TRACE_CHUNK:
+ DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn);
+ ret = relay_create_trace_chunk(header, conn, payload);
+ break;
+ case RELAYD_CLOSE_TRACE_CHUNK:
+ DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn);
+ ret = relay_close_trace_chunk(header, conn, payload);
+ break;
+ case RELAYD_TRACE_CHUNK_EXISTS:
+ DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn);
+ ret = relay_trace_chunk_exists(header, conn, payload);
+ break;
+ case RELAYD_UPDATE_SYNC_INFO:
+ default:
+ ERR("Received unknown command (%u)", header->cmd);
+ relay_unknown_command(conn);
+ ret = -1;
+ goto end;
+ }
+
+end:
+ return ret;
+}
+
+static enum relay_connection_status relay_process_control_receive_payload(
+ struct relay_connection *conn)
+{
+ int ret = 0;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct lttng_dynamic_buffer *reception_buffer =
+ &conn->protocol.ctrl.reception_buffer;
+ struct ctrl_connection_state_receive_payload *state =
+ &conn->protocol.ctrl.state.receive_payload;
+ struct lttng_buffer_view payload_view;
+
+ if (state->left_to_receive == 0) {
+ /* Short-circuit for payload-less commands. */
+ goto reception_complete;
+ }
+
+ ret = conn->sock->ops->recvmsg(conn->sock,
+ reception_buffer->data + state->received,
+ state->left_to_receive, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Unable to receive command payload on sock %d",
+ conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+ goto end;
+ } else if (ret == 0) {
+ DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ goto end;
+ }
+
+ assert(ret > 0);
+ assert(ret <= state->left_to_receive);
+
+ state->left_to_receive -= ret;
+ state->received += ret;
+
+ if (state->left_to_receive > 0) {
+ /*
+ * Can't transition to the protocol's next state, wait to
+ * receive the rest of the header.
+ */
+ DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
+ state->received, state->left_to_receive,
+ conn->sock->fd);
+ goto end;
+ }
+
+reception_complete:
+ DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
+ conn->sock->fd, state->received);
+ /*
+ * The payload required to process the command has been received.
+ * A view to the reception buffer is forwarded to the various
+ * commands and the state of the control is reset on success.
+ *
+ * Commands are responsible for sending their reply to the peer.
+ */
+ payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
+ 0, -1);
+ ret = relay_process_control_command(conn,
+ &state->header, &payload_view);
+ if (ret < 0) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ ret = connection_reset_protocol_state(conn);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+end:
+ return status;
+}
+
+static enum relay_connection_status relay_process_control_receive_header(
+ struct relay_connection *conn)
+{
+ int ret = 0;
+ enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
+ struct lttcomm_relayd_hdr header;
+ struct lttng_dynamic_buffer *reception_buffer =
+ &conn->protocol.ctrl.reception_buffer;
+ struct ctrl_connection_state_receive_header *state =
+ &conn->protocol.ctrl.state.receive_header;
+
+ assert(state->left_to_receive != 0);
+
+ ret = conn->sock->ops->recvmsg(conn->sock,
+ reception_buffer->data + state->received,
+ state->left_to_receive, MSG_DONTWAIT);
+ if (ret < 0) {
+ if (errno != EAGAIN && errno != EWOULDBLOCK) {
+ PERROR("Unable to receive control command header on sock %d",
+ conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ }
+ goto end;
+ } else if (ret == 0) {
+ DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
+ status = RELAY_CONNECTION_STATUS_CLOSED;
+ goto end;
+ }
+
+ assert(ret > 0);
+ assert(ret <= state->left_to_receive);
+
+ state->left_to_receive -= ret;
+ state->received += ret;
+
+ if (state->left_to_receive > 0) {
+ /*
+ * Can't transition to the protocol's next state, wait to
+ * receive the rest of the header.
+ */
+ DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
+ state->received, state->left_to_receive,
+ conn->sock->fd);
+ goto end;
+ }