+int relay_rotate_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_rotate_pending msg;
+ struct lttcomm_relayd_rotate_pending_reply reply;
+ struct lttng_ht_iter iter;
+ struct relay_stream *stream;
+ int ret = 0;
+ ssize_t send_ret;
+ uint64_t chunk_id;
+ bool rotate_pending = false;
+
+ DBG("Rotate pending command received");
+
+ if (!session || !conn->version_check_done) {
+ ERR("Trying to check for data 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 < sizeof(msg)) {
+ ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
+ sizeof(msg), payload->size);
+ ret = -1;
+ goto end_no_reply;
+ }
+
+ memcpy(&msg, payload->data, sizeof(msg));
+
+ chunk_id = be64toh(msg.chunk_id);
+
+ DBG("Evaluating rotate pending for session \"%s\" and chunk id %" PRIu64,
+ session->session_name, chunk_id);
+
+ /*
+ * Iterate over all the streams in the session and check if they are
+ * still waiting for data to perform their rotation.
+ */
+ 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 != session) {
+ stream_put(stream);
+ continue;
+ }
+ pthread_mutex_lock(&stream->lock);
+ if (stream->rotate_at_seq_num != -1ULL) {
+ /* We have not yet performed the rotation. */
+ rotate_pending = true;
+ DBG("Stream %" PRIu64 " is still rotating",
+ stream->stream_handle);
+ } else if (stream->current_chunk_id.value < chunk_id) {
+ /*
+ * Stream closed on the consumer but still active on the
+ * relay.
+ */
+ rotate_pending = true;
+ DBG("Stream %" PRIu64 " did not exist on the consumer "
+ "when the last rotation started, but is"
+ "still waiting for data before getting"
+ "closed",
+ stream->stream_handle);
+ }
+ pthread_mutex_unlock(&stream->lock);
+ stream_put(stream);
+ if (rotate_pending) {
+ goto send_reply;
+ }
+ }
+
+send_reply:
+ rcu_read_unlock();
+ memset(&reply, 0, sizeof(reply));
+ reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK);
+ reply.is_pending = (uint8_t) !!rotate_pending;
+ send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
+ sizeof(reply), 0);
+ if (send_ret < (ssize_t) sizeof(reply)) {
+ ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
+ send_ret);
+ ret = -1;
+ }
+
+end_no_reply:
+ 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_STREAM:
+ DBG_CMD("RELAYD_ROTATE_STREAM", conn);
+ ret = relay_rotate_session_stream(header, conn, payload);
+ break;
+ case RELAYD_ROTATE_RENAME:
+ DBG_CMD("RELAYD_ROTATE_RENAME", conn);
+ ret = relay_rotate_rename(header, conn, payload);
+ break;
+ case RELAYD_ROTATE_PENDING:
+ DBG_CMD("RELAYD_ROTATE_PENDING", conn);
+ ret = relay_rotate_pending(header, conn, payload);
+ break;
+ case RELAYD_MKDIR:
+ DBG_CMD("RELAYD_MKDIR", conn);
+ ret = relay_mkdir(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;
+ }
+
+ /* Transition to next state: receiving the command's payload. */
+ conn->protocol.ctrl.state_id =
+ CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
+ memcpy(&header, reception_buffer->data, sizeof(header));
+ header.circuit_id = be64toh(header.circuit_id);
+ header.data_size = be64toh(header.data_size);
+ header.cmd = be32toh(header.cmd);
+ header.cmd_version = be32toh(header.cmd_version);
+ memcpy(&conn->protocol.ctrl.state.receive_payload.header,
+ &header, sizeof(header));
+
+ DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
+ conn->sock->fd, header.cmd, header.cmd_version,
+ header.data_size);
+
+ if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
+ ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
+ header.data_size);
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ conn->protocol.ctrl.state.receive_payload.left_to_receive =
+ header.data_size;
+ conn->protocol.ctrl.state.receive_payload.received = 0;
+ ret = lttng_dynamic_buffer_set_size(reception_buffer,
+ header.data_size);
+ if (ret) {
+ status = RELAY_CONNECTION_STATUS_ERROR;
+ goto end;
+ }
+
+ if (header.data_size == 0) {
+ /*
+ * Manually invoke the next state as the poll loop
+ * will not wake-up to allow us to proceed further.
+ */
+ status = relay_process_control_receive_payload(conn);
+ }
+end:
+ return status;
+}
+
+/*
+ * Process the commands received on the control socket
+ */
+static enum relay_connection_status relay_process_control(
+ struct relay_connection *conn)
+{
+ enum relay_connection_status status;
+
+ switch (conn->protocol.ctrl.state_id) {
+ case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
+ status = relay_process_control_receive_header(conn);
+ break;
+ case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
+ status = relay_process_control_receive_payload(conn);
+ break;
+ default:
+ ERR("Unknown control connection protocol state encountered.");
+ abort();
+ }
+
+ return status;
+}
+
+/*
+ * Handle index for a data stream.
+ *
+ * Called with the stream lock held.
+ *
+ * Return 0 on success else a negative value.
+ */
+static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
+ bool rotate_index, bool *flushed, uint64_t total_size)
+{
+ int ret = 0;
+ uint64_t data_offset;
+ struct relay_index *index;
+
+ /* Get data offset because we are about to update the index. */
+ data_offset = htobe64(stream->tracefile_size_current);
+
+ DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
+ stream->stream_handle, net_seq_num, stream->tracefile_size_current);
+
+ /*
+ * Lookup for an existing index for that stream id/sequence
+ * number. If it exists, the control thread has already received the
+ * data for it, thus we need to write it to disk.
+ */
+ index = relay_index_get_by_id_or_create(stream, net_seq_num);
+ if (!index) {
+ ret = -1;
+ goto end;
+ }
+
+ if (rotate_index || !stream->index_file) {
+ ret = create_rotate_index_file(stream);
+ if (ret < 0) {
+ ERR("Failed to rotate index");
+ /* Put self-ref for this index due to error. */
+ relay_index_put(index);
+ index = NULL;
+ goto end;
+ }
+ }
+
+ if (relay_index_set_file(index, stream->index_file, data_offset)) {
+ ret = -1;
+ /* Put self-ref for this index due to error. */
+ relay_index_put(index);
+ index = NULL;
+ goto end;
+ }
+
+ ret = relay_index_try_flush(index);
+ if (ret == 0) {
+ tracefile_array_commit_seq(stream->tfa);
+ stream->index_received_seqcount++;
+ *flushed = true;
+ } else if (ret > 0) {
+ index->total_size = total_size;
+ /* No flush. */
+ ret = 0;
+ } else {
+ /*
+ * ret < 0
+ *
+ * relay_index_try_flush is responsible for the self-reference
+ * put of the index object on error.
+ */
+ ERR("relay_index_try_flush error %d", ret);
+ ret = -1;
+ }
+end:
+ return ret;
+}
+
+static enum relay_connection_status relay_process_data_receive_header(
+ struct relay_connection *conn)