+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
+ if (ret < 0) {
+ PERROR("send lost packets");
+ goto error_fatal;
+ }
+
+ break;
+ }
+ case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
+ {
+ int channel_monitor_pipe;
+
+ ret_code = LTTCOMM_CONSUMERD_SUCCESS;
+ /* Successfully received the command's type. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ goto error_fatal;
+ }
+
+ ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
+ 1);
+ if (ret != sizeof(channel_monitor_pipe)) {
+ ERR("Failed to receive channel monitor pipe");
+ goto error_fatal;
+ }
+
+ DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
+ ret = consumer_timer_thread_set_channel_monitor_pipe(
+ channel_monitor_pipe);
+ if (!ret) {
+ int flags;
+
+ ret_code = LTTCOMM_CONSUMERD_SUCCESS;
+ /* Set the pipe as non-blocking. */
+ ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
+ if (ret == -1) {
+ PERROR("fcntl get flags of the channel monitoring pipe");
+ goto error_fatal;
+ }
+ flags = ret;
+
+ ret = fcntl(channel_monitor_pipe, F_SETFL,
+ flags | O_NONBLOCK);
+ if (ret == -1) {
+ PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
+ goto error_fatal;
+ }
+ DBG("Channel monitor pipe set as non-blocking");
+ } else {
+ ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
+ }
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ goto error_fatal;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_ROTATE_CHANNEL:
+ {
+ struct lttng_consumer_channel *channel;
+ uint64_t key = msg.u.rotate_channel.key;
+
+ DBG("Consumer rotate channel %" PRIu64, key);
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("Channel %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ } else {
+ /*
+ * Sample the rotate position of all the streams in this channel.
+ */
+ ret = lttng_consumer_rotate_channel(channel, key,
+ msg.u.rotate_channel.relayd_id,
+ msg.u.rotate_channel.metadata,
+ ctx);
+ if (ret < 0) {
+ ERR("Rotate channel failed");
+ ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
+ }
+
+ health_code_update();
+ }
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_rotate_channel;
+ }
+ if (channel) {
+ /* Rotate the streams that are ready right now. */
+ ret = lttng_consumer_rotate_ready_streams(
+ channel, key, ctx);
+ if (ret < 0) {
+ ERR("Rotate ready streams failed");
+ }
+ }
+ break;
+error_rotate_channel:
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_CLEAR_CHANNEL:
+ {
+ struct lttng_consumer_channel *channel;
+ uint64_t key = msg.u.clear_channel.key;
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ DBG("Channel %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ } else {
+ ret = lttng_consumer_clear_channel(channel);
+ if (ret) {
+ ERR("Clear channel failed");
+ ret_code = ret;
+ }
+
+ health_code_update();
+ }
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
+
+ break;
+ }
+ case LTTNG_CONSUMER_INIT:
+ {
+ ret_code = lttng_consumer_init_command(ctx,
+ msg.u.init.sessiond_uuid);
+ health_code_update();
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
+ break;
+ }
+ case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
+ {
+ const struct lttng_credentials credentials = {
+ .uid = msg.u.create_trace_chunk.credentials.value.uid,
+ .gid = msg.u.create_trace_chunk.credentials.value.gid,
+ };
+ const bool is_local_trace =
+ !msg.u.create_trace_chunk.relayd_id.is_set;
+ const uint64_t relayd_id =
+ msg.u.create_trace_chunk.relayd_id.value;
+ const char *chunk_override_name =
+ *msg.u.create_trace_chunk.override_name ?
+ msg.u.create_trace_chunk.override_name :
+ NULL;
+ struct lttng_directory_handle *chunk_directory_handle = NULL;
+
+ /*
+ * The session daemon will only provide a chunk directory file
+ * descriptor for local traces.
+ */
+ if (is_local_trace) {
+ int chunk_dirfd;
+
+ /* Acnowledge the reception of the command. */
+ ret = consumer_send_status_msg(sock,
+ LTTCOMM_CONSUMERD_SUCCESS);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_nosignal;
+ }
+
+ ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
+ if (ret != sizeof(chunk_dirfd)) {
+ ERR("Failed to receive trace chunk directory file descriptor");
+ goto error_fatal;
+ }
+
+ DBG("Received trace chunk directory fd (%d)",
+ chunk_dirfd);
+ chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
+ chunk_dirfd);
+ if (!chunk_directory_handle) {
+ ERR("Failed to initialize chunk directory handle from directory file descriptor");
+ if (close(chunk_dirfd)) {
+ PERROR("Failed to close chunk directory file descriptor");
+ }
+ goto error_fatal;
+ }
+ }
+
+ ret_code = lttng_consumer_create_trace_chunk(
+ !is_local_trace ? &relayd_id : NULL,
+ msg.u.create_trace_chunk.session_id,
+ msg.u.create_trace_chunk.chunk_id,
+ (time_t) msg.u.create_trace_chunk
+ .creation_timestamp,
+ chunk_override_name,
+ msg.u.create_trace_chunk.credentials.is_set ?
+ &credentials :
+ NULL,
+ chunk_directory_handle);
+ lttng_directory_handle_put(chunk_directory_handle);
+ goto end_msg_sessiond;
+ }
+ case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
+ {
+ enum lttng_trace_chunk_command_type close_command =
+ msg.u.close_trace_chunk.close_command.value;
+ const uint64_t relayd_id =
+ msg.u.close_trace_chunk.relayd_id.value;
+ struct lttcomm_consumer_close_trace_chunk_reply reply;
+ char path[LTTNG_PATH_MAX];
+
+ ret_code = lttng_consumer_close_trace_chunk(
+ msg.u.close_trace_chunk.relayd_id.is_set ?
+ &relayd_id :
+ NULL,
+ msg.u.close_trace_chunk.session_id,
+ msg.u.close_trace_chunk.chunk_id,
+ (time_t) msg.u.close_trace_chunk.close_timestamp,
+ msg.u.close_trace_chunk.close_command.is_set ?
+ &close_command :
+ NULL, path);
+ reply.ret_code = ret_code;
+ reply.path_length = strlen(path) + 1;
+ ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
+ if (ret != sizeof(reply)) {
+ goto error_fatal;
+ }
+ ret = lttcomm_send_unix_sock(sock, path, reply.path_length);
+ if (ret != reply.path_length) {
+ goto error_fatal;
+ }
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
+ {
+ const uint64_t relayd_id =
+ msg.u.trace_chunk_exists.relayd_id.value;
+
+ ret_code = lttng_consumer_trace_chunk_exists(
+ msg.u.trace_chunk_exists.relayd_id.is_set ?
+ &relayd_id : NULL,
+ msg.u.trace_chunk_exists.session_id,
+ msg.u.trace_chunk_exists.chunk_id);
+ goto end_msg_sessiond;
+ }
+ default:
+ goto end_nosignal;
+ }
+
+end_nosignal:
+ /*
+ * Return 1 to indicate success since the 0 value can be a socket
+ * shutdown during the recv() or send() call.
+ */
+ ret = 1;
+ goto end;
+error_fatal:
+ /* This will issue a consumer stop. */
+ ret = -1;
+ goto end;
+end_msg_sessiond:
+ /*
+ * The returned value here is not useful since either way we'll return 1 to
+ * the caller because the session daemon socket management is done
+ * elsewhere. Returning a negative code or 0 will shutdown the consumer.
+ */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ goto error_fatal;
+ }
+ ret = 1;
+end:
+ health_code_update();
+ rcu_read_unlock();
+ return ret;
+}
+
+/*
+ * Populate index values of a kernel stream. Values are set in big endian order.
+ *
+ * Return 0 on success or else a negative value.
+ */
+static int get_index_values(struct ctf_packet_index *index, int infd)
+{
+ int ret;
+ uint64_t packet_size, content_size, timestamp_begin, timestamp_end,
+ events_discarded, stream_id, stream_instance_id,
+ packet_seq_num;
+
+ ret = kernctl_get_timestamp_begin(infd, ×tamp_begin);
+ if (ret < 0) {
+ PERROR("kernctl_get_timestamp_begin");
+ goto error;
+ }
+
+ ret = kernctl_get_timestamp_end(infd, ×tamp_end);
+ if (ret < 0) {
+ PERROR("kernctl_get_timestamp_end");
+ goto error;
+ }
+
+ ret = kernctl_get_events_discarded(infd, &events_discarded);
+ if (ret < 0) {
+ PERROR("kernctl_get_events_discarded");
+ goto error;
+ }
+
+ ret = kernctl_get_content_size(infd, &content_size);
+ if (ret < 0) {
+ PERROR("kernctl_get_content_size");
+ goto error;
+ }
+
+ ret = kernctl_get_packet_size(infd, &packet_size);
+ if (ret < 0) {
+ PERROR("kernctl_get_packet_size");
+ goto error;
+ }
+
+ ret = kernctl_get_stream_id(infd, &stream_id);
+ if (ret < 0) {
+ PERROR("kernctl_get_stream_id");
+ goto error;
+ }
+
+ ret = kernctl_get_instance_id(infd, &stream_instance_id);
+ if (ret < 0) {
+ if (ret == -ENOTTY) {
+ /* Command not implemented by lttng-modules. */
+ stream_instance_id = -1ULL;
+ } else {
+ PERROR("kernctl_get_instance_id");
+ goto error;
+ }
+ }
+
+ ret = kernctl_get_sequence_number(infd, &packet_seq_num);
+ if (ret < 0) {
+ if (ret == -ENOTTY) {
+ /* Command not implemented by lttng-modules. */
+ packet_seq_num = -1ULL;
+ ret = 0;
+ } else {
+ PERROR("kernctl_get_sequence_number");
+ goto error;
+ }
+ }
+ index->packet_seq_num = htobe64(index->packet_seq_num);
+
+ *index = (typeof(*index)) {
+ .offset = index->offset,
+ .packet_size = htobe64(packet_size),
+ .content_size = htobe64(content_size),
+ .timestamp_begin = htobe64(timestamp_begin),
+ .timestamp_end = htobe64(timestamp_end),
+ .events_discarded = htobe64(events_discarded),
+ .stream_id = htobe64(stream_id),
+ .stream_instance_id = htobe64(stream_instance_id),
+ .packet_seq_num = htobe64(packet_seq_num),
+ };
+
+error:
+ return ret;
+}
+/*
+ * Sync metadata meaning request them to the session daemon and snapshot to the
+ * metadata thread can consumer them.
+ *
+ * Metadata stream lock MUST be acquired.
+ *
+ * Return 0 if new metadatda is available, EAGAIN if the metadata stream
+ * is empty or a negative value on error.
+ */
+int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
+{
+ int ret;
+
+ assert(metadata);
+
+ ret = kernctl_buffer_flush(metadata->wait_fd);
+ if (ret < 0) {
+ ERR("Failed to flush kernel stream");
+ goto end;
+ }
+
+ ret = kernctl_snapshot(metadata->wait_fd);
+ if (ret < 0) {
+ if (ret != -EAGAIN) {
+ ERR("Sync metadata, taking kernel snapshot failed.");
+ goto end;
+ }
+ DBG("Sync metadata, no new kernel metadata");
+ /* No new metadata, exit. */
+ ret = ENODATA;
+ goto end;
+ }
+
+end:
+ return ret;
+}
+
+static
+int update_stream_stats(struct lttng_consumer_stream *stream)
+{
+ int ret;
+ uint64_t seq, discarded;
+
+ ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
+ if (ret < 0) {
+ if (ret == -ENOTTY) {
+ /* Command not implemented by lttng-modules. */
+ seq = -1ULL;
+ stream->sequence_number_unavailable = true;
+ } else {
+ PERROR("kernctl_get_sequence_number");
+ goto end;
+ }
+ }
+
+ /*
+ * Start the sequence when we extract the first packet in case we don't
+ * start at 0 (for example if a consumer is not connected to the
+ * session immediately after the beginning).
+ */
+ if (stream->last_sequence_number == -1ULL) {
+ stream->last_sequence_number = seq;
+ } else if (seq > stream->last_sequence_number) {
+ stream->chan->lost_packets += seq -
+ stream->last_sequence_number - 1;
+ } else {
+ /* seq <= last_sequence_number */
+ ERR("Sequence number inconsistent : prev = %" PRIu64
+ ", current = %" PRIu64,
+ stream->last_sequence_number, seq);
+ ret = -1;
+ goto end;
+ }
+ stream->last_sequence_number = seq;
+
+ ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
+ if (ret < 0) {
+ PERROR("kernctl_get_events_discarded");
+ goto end;
+ }
+ if (discarded < stream->last_discarded_events) {
+ /*
+ * Overflow has occurred. We assume only one wrap-around
+ * has occurred.
+ */
+ stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
+ stream->last_discarded_events + discarded;
+ } else {
+ stream->chan->discarded_events += discarded -
+ stream->last_discarded_events;
+ }
+ stream->last_discarded_events = discarded;
+ ret = 0;
+
+end:
+ return ret;
+}
+
+/*
+ * Check if the local version of the metadata stream matches with the version
+ * of the metadata stream in the kernel. If it was updated, set the reset flag
+ * on the stream.
+ */
+static
+int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
+{
+ int ret;
+ uint64_t cur_version;
+
+ ret = kernctl_get_metadata_version(infd, &cur_version);
+ if (ret < 0) {
+ if (ret == -ENOTTY) {
+ /*
+ * LTTng-modules does not implement this
+ * command.
+ */
+ ret = 0;
+ goto end;
+ }
+ ERR("Failed to get the metadata version");
+ goto end;
+ }
+
+ if (stream->metadata_version == cur_version) {
+ ret = 0;
+ goto end;
+ }
+
+ DBG("New metadata version detected");
+ stream->metadata_version = cur_version;
+ stream->reset_metadata_flag = 1;
+ ret = 0;
+
+end:
+ return ret;
+}
+
+/*
+ * Consume data on a file descriptor and write it on a trace file.
+ * The stream and channel locks must be held by the caller.
+ */
+ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
+ struct lttng_consumer_local_data *ctx)
+{
+ unsigned long len, subbuf_size, padding;
+ int err, write_index = 1, rotation_ret;
+ ssize_t ret = 0;
+ int infd = stream->wait_fd;
+ struct ctf_packet_index index = {};
+
+ DBG("In read_subbuffer (infd : %d)", infd);
+
+ /*
+ * If the stream was flagged to be ready for rotation before we extract the
+ * next packet, rotate it now.
+ */
+ if (stream->rotate_ready) {
+ DBG("Rotate stream before extracting data");
+ rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
+ if (rotation_ret < 0) {
+ ERR("Stream rotation error");
+ ret = -1;
+ goto error;
+ }
+ }
+
+ /* Get the next subbuffer */
+ err = kernctl_get_next_subbuf(infd);
+ if (err != 0) {
+ /*
+ * This is a debug message even for single-threaded consumer,
+ * because poll() have more relaxed criterions than get subbuf,
+ * so get_subbuf may fail for short race windows where poll()
+ * would issue wakeups.
+ */
+ DBG("Reserving sub buffer failed (everything is normal, "
+ "it is due to concurrency)");
+ ret = err;
+ goto error;
+ }
+
+ /* Get the full subbuffer size including padding */
+ err = kernctl_get_padded_subbuf_size(infd, &len);
+ if (err != 0) {
+ PERROR("Getting sub-buffer len failed.");
+ err = kernctl_put_subbuf(infd);
+ if (err != 0) {
+ if (err == -EFAULT) {
+ PERROR("Error in unreserving sub buffer\n");
+ } else if (err == -EIO) {
+ /* Should never happen with newer LTTng versions */
+ PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
+ }
+ ret = err;
+ goto error;
+ }
+ ret = err;
+ goto error;
+ }
+
+ if (!stream->metadata_flag) {
+ ret = get_index_values(&index, infd);
+ if (ret < 0) {
+ err = kernctl_put_subbuf(infd);
+ if (err != 0) {
+ if (err == -EFAULT) {
+ PERROR("Error in unreserving sub buffer\n");
+ } else if (err == -EIO) {
+ /* Should never happen with newer LTTng versions */
+ PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
+ }
+ ret = err;
+ goto error;
+ }
+ goto error;
+ }
+ ret = update_stream_stats(stream);
+ if (ret < 0) {
+ err = kernctl_put_subbuf(infd);
+ if (err != 0) {
+ if (err == -EFAULT) {
+ PERROR("Error in unreserving sub buffer\n");
+ } else if (err == -EIO) {
+ /* Should never happen with newer LTTng versions */
+ PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
+ }
+ ret = err;
+ goto error;
+ }
+ goto error;
+ }
+ } else {
+ write_index = 0;
+ ret = metadata_stream_check_version(infd, stream);
+ if (ret < 0) {
+ err = kernctl_put_subbuf(infd);
+ if (err != 0) {
+ if (err == -EFAULT) {
+ PERROR("Error in unreserving sub buffer\n");
+ } else if (err == -EIO) {
+ /* Should never happen with newer LTTng versions */
+ PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
+ }
+ ret = err;
+ goto error;
+ }
+ goto error;
+ }
+ }
+
+ switch (stream->chan->output) {
+ case CONSUMER_CHANNEL_SPLICE:
+ /*
+ * XXX: The lttng-modules splice "actor" does not handle copying
+ * partial pages hence only using the subbuffer size without the
+ * padding makes the splice fail.
+ */
+ subbuf_size = len;
+ padding = 0;
+
+ /* splice the subbuffer to the tracefile */
+ ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
+ padding, &index);
+ /*
+ * XXX: Splice does not support network streaming so the return value
+ * is simply checked against subbuf_size and not like the mmap() op.
+ */
+ if (ret != subbuf_size) {
+ /*
+ * display the error but continue processing to try
+ * to release the subbuffer
+ */
+ ERR("Error splicing to tracefile (ret: %zd != len: %lu)",