+ /* Session daemon status message are handled in the following call. */
+ consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
+ msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
+ &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
+ msg.u.relayd_sock.relayd_session_id);
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_ADD_CHANNEL:
+ {
+ struct lttng_consumer_channel *new_channel;
+ int ret_recv;
+ const uint64_t chunk_id = msg.u.channel.chunk_id.value;
+
+ health_code_update();
+
+ /* First send a status message before receiving the fds. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_fatal;
+ }
+
+ health_code_update();
+
+ DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
+ new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
+ msg.u.channel.session_id,
+ msg.u.channel.chunk_id.is_set ?
+ &chunk_id : NULL,
+ msg.u.channel.pathname,
+ msg.u.channel.name,
+ msg.u.channel.relayd_id, msg.u.channel.output,
+ msg.u.channel.tracefile_size,
+ msg.u.channel.tracefile_count, 0,
+ msg.u.channel.monitor,
+ msg.u.channel.live_timer_interval,
+ NULL, NULL);
+ if (new_channel == NULL) {
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
+ goto end_nosignal;
+ }
+ new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
+ switch (msg.u.channel.output) {
+ case LTTNG_EVENT_SPLICE:
+ new_channel->output = CONSUMER_CHANNEL_SPLICE;
+ break;
+ case LTTNG_EVENT_MMAP:
+ new_channel->output = CONSUMER_CHANNEL_MMAP;
+ break;
+ default:
+ ERR("Channel output unknown %d", msg.u.channel.output);
+ goto end_nosignal;
+ }
+
+ /* Translate and save channel type. */
+ switch (msg.u.channel.type) {
+ case CONSUMER_CHANNEL_TYPE_DATA:
+ case CONSUMER_CHANNEL_TYPE_METADATA:
+ new_channel->type = msg.u.channel.type;
+ break;
+ default:
+ assert(0);
+ goto end_nosignal;
+ };
+
+ health_code_update();
+
+ if (ctx->on_recv_channel != NULL) {
+ ret_recv = ctx->on_recv_channel(new_channel);
+ if (ret_recv == 0) {
+ ret = consumer_add_channel(new_channel, ctx);
+ } else if (ret_recv < 0) {
+ goto end_nosignal;
+ }
+ } else {
+ ret = consumer_add_channel(new_channel, ctx);
+ }
+ if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
+ int monitor_start_ret;
+
+ DBG("Consumer starting monitor timer");
+ consumer_timer_live_start(new_channel,
+ msg.u.channel.live_timer_interval);
+ monitor_start_ret = consumer_timer_monitor_start(
+ new_channel,
+ msg.u.channel.monitor_timer_interval);
+ if (monitor_start_ret < 0) {
+ ERR("Starting channel monitoring timer failed");
+ goto end_nosignal;
+ }
+
+ }
+
+ health_code_update();
+
+ /* If we received an error in add_channel, we need to report it. */
+ if (ret < 0) {
+ ret = consumer_send_status_msg(sock, ret);
+ if (ret < 0) {
+ goto error_fatal;
+ }
+ goto end_nosignal;
+ }
+
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_ADD_STREAM:
+ {
+ int fd;
+ struct lttng_pipe *stream_pipe;
+ struct lttng_consumer_stream *new_stream;
+ struct lttng_consumer_channel *channel;
+ int alloc_ret = 0;
+
+ /*
+ * Get stream's channel reference. Needed when adding the stream to the
+ * global hash table.
+ */
+ channel = consumer_find_channel(msg.u.stream.channel_key);
+ if (!channel) {
+ /*
+ * We could not find the channel. Can happen if cpu hotplug
+ * happens while tearing down.
+ */
+ ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ }
+
+ health_code_update();
+
+ /* First send a status message before receiving the fds. */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_add_stream_fatal;
+ }
+
+ health_code_update();
+
+ if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
+ /* Channel was not found. */
+ goto error_add_stream_nosignal;
+ }
+
+ /* Blocking call */
+ health_poll_entry();
+ ret = lttng_consumer_poll_socket(consumer_sockpoll);
+ health_poll_exit();
+ if (ret) {
+ goto error_add_stream_fatal;
+ }
+
+ health_code_update();
+
+ /* Get stream file descriptor from socket */
+ ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
+ if (ret != sizeof(fd)) {
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
+ goto end;
+ }
+
+ health_code_update();
+
+ /*
+ * Send status code to session daemon only if the recv works. If the
+ * above recv() failed, the session daemon is notified through the
+ * error socket and the teardown is eventually done.
+ */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_add_stream_nosignal;
+ }
+
+ health_code_update();
+
+ pthread_mutex_lock(&channel->lock);
+ new_stream = consumer_allocate_stream(
+ channel,
+ channel->key,
+ fd,
+ channel->name,
+ channel->relayd_id,
+ channel->session_id,
+ channel->trace_chunk,
+ msg.u.stream.cpu,
+ &alloc_ret,
+ channel->type,
+ channel->monitor);
+ if (new_stream == NULL) {
+ switch (alloc_ret) {
+ case -ENOMEM:
+ case -EINVAL:
+ default:
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
+ break;
+ }
+ pthread_mutex_unlock(&channel->lock);
+ goto error_add_stream_nosignal;
+ }
+
+ new_stream->wait_fd = fd;
+ ret = kernctl_get_max_subbuf_size(new_stream->wait_fd,
+ &new_stream->max_sb_size);
+ if (ret < 0) {
+ pthread_mutex_unlock(&channel->lock);
+ ERR("Failed to get kernel maximal subbuffer size");
+ goto error_add_stream_nosignal;
+ }
+
+ consumer_stream_update_channel_attributes(new_stream,
+ channel);
+ switch (channel->output) {
+ case CONSUMER_CHANNEL_SPLICE:
+ new_stream->output = LTTNG_EVENT_SPLICE;
+ ret = utils_create_pipe(new_stream->splice_pipe);
+ if (ret < 0) {
+ pthread_mutex_unlock(&channel->lock);
+ goto error_add_stream_nosignal;
+ }
+ break;
+ case CONSUMER_CHANNEL_MMAP:
+ new_stream->output = LTTNG_EVENT_MMAP;
+ break;
+ default:
+ ERR("Stream output unknown %d", channel->output);
+ pthread_mutex_unlock(&channel->lock);
+ goto error_add_stream_nosignal;
+ }
+
+ /*
+ * We've just assigned the channel to the stream so increment the
+ * refcount right now. We don't need to increment the refcount for
+ * streams in no monitor because we handle manually the cleanup of
+ * those. It is very important to make sure there is NO prior
+ * consumer_del_stream() calls or else the refcount will be unbalanced.
+ */
+ if (channel->monitor) {
+ uatomic_inc(&new_stream->chan->refcount);
+ }
+
+ /*
+ * The buffer flush is done on the session daemon side for the kernel
+ * so no need for the stream "hangup_flush_done" variable to be
+ * tracked. This is important for a kernel stream since we don't rely
+ * on the flush state of the stream to read data. It's not the case for
+ * user space tracing.
+ */
+ new_stream->hangup_flush_done = 0;
+
+ health_code_update();
+
+ pthread_mutex_lock(&new_stream->lock);
+ if (ctx->on_recv_stream) {
+ ret = ctx->on_recv_stream(new_stream);
+ if (ret < 0) {
+ pthread_mutex_unlock(&new_stream->lock);
+ pthread_mutex_unlock(&channel->lock);
+ consumer_stream_free(new_stream);
+ goto error_add_stream_nosignal;
+ }
+ }
+ health_code_update();
+
+ if (new_stream->metadata_flag) {
+ channel->metadata_stream = new_stream;
+ }
+
+ /* Do not monitor this stream. */
+ if (!channel->monitor) {
+ DBG("Kernel consumer add stream %s in no monitor mode with "
+ "relayd id %" PRIu64, new_stream->name,
+ new_stream->net_seq_idx);
+ cds_list_add(&new_stream->send_node, &channel->streams.head);
+ pthread_mutex_unlock(&new_stream->lock);
+ pthread_mutex_unlock(&channel->lock);
+ goto end_add_stream;
+ }
+
+ /* Send stream to relayd if the stream has an ID. */
+ if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
+ ret = consumer_send_relayd_stream(new_stream,
+ new_stream->chan->pathname);
+ if (ret < 0) {
+ pthread_mutex_unlock(&new_stream->lock);
+ pthread_mutex_unlock(&channel->lock);
+ consumer_stream_free(new_stream);
+ goto error_add_stream_nosignal;
+ }
+
+ /*
+ * If adding an extra stream to an already
+ * existing channel (e.g. cpu hotplug), we need
+ * to send the "streams_sent" command to relayd.
+ */
+ if (channel->streams_sent_to_relayd) {
+ ret = consumer_send_relayd_streams_sent(
+ new_stream->net_seq_idx);
+ if (ret < 0) {
+ pthread_mutex_unlock(&new_stream->lock);
+ pthread_mutex_unlock(&channel->lock);
+ goto error_add_stream_nosignal;
+ }
+ }
+ }
+ pthread_mutex_unlock(&new_stream->lock);
+ pthread_mutex_unlock(&channel->lock);
+
+ /* Get the right pipe where the stream will be sent. */
+ if (new_stream->metadata_flag) {
+ consumer_add_metadata_stream(new_stream);
+ stream_pipe = ctx->consumer_metadata_pipe;
+ } else {
+ consumer_add_data_stream(new_stream);
+ stream_pipe = ctx->consumer_data_pipe;
+ }
+
+ /* Visible to other threads */
+ new_stream->globally_visible = 1;
+
+ health_code_update();
+
+ ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
+ if (ret < 0) {
+ ERR("Consumer write %s stream to pipe %d",
+ new_stream->metadata_flag ? "metadata" : "data",
+ lttng_pipe_get_writefd(stream_pipe));
+ if (new_stream->metadata_flag) {
+ consumer_del_stream_for_metadata(new_stream);
+ } else {
+ consumer_del_stream_for_data(new_stream);
+ }
+ goto error_add_stream_nosignal;
+ }
+
+ DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
+ new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
+end_add_stream:
+ break;
+error_add_stream_nosignal:
+ goto end_nosignal;
+error_add_stream_fatal:
+ goto error_fatal;
+ }
+ case LTTNG_CONSUMER_STREAMS_SENT:
+ {
+ struct lttng_consumer_channel *channel;
+
+ /*
+ * Get stream's channel reference. Needed when adding the stream to the
+ * global hash table.
+ */
+ channel = consumer_find_channel(msg.u.sent_streams.channel_key);
+ if (!channel) {
+ /*
+ * We could not find the channel. Can happen if cpu hotplug
+ * happens while tearing down.
+ */
+ ERR("Unable to find channel key %" PRIu64,
+ msg.u.sent_streams.channel_key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ }
+
+ health_code_update();
+
+ /*
+ * Send status code to session daemon.
+ */
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_streams_sent_nosignal;
+ }
+
+ health_code_update();
+
+ /*
+ * We should not send this message if we don't monitor the
+ * streams in this channel.
+ */
+ if (!channel->monitor) {
+ goto end_error_streams_sent;
+ }
+
+ health_code_update();
+ /* Send stream to relayd if the stream has an ID. */
+ if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
+ ret = consumer_send_relayd_streams_sent(
+ msg.u.sent_streams.net_seq_idx);
+ if (ret < 0) {
+ goto error_streams_sent_nosignal;
+ }
+ channel->streams_sent_to_relayd = true;
+ }
+end_error_streams_sent:
+ break;
+error_streams_sent_nosignal:
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_UPDATE_STREAM:
+ {
+ rcu_read_unlock();
+ return -ENOSYS;
+ }
+ case LTTNG_CONSUMER_DESTROY_RELAYD:
+ {
+ uint64_t index = msg.u.destroy_relayd.net_seq_idx;
+ struct consumer_relayd_sock_pair *relayd;
+
+ DBG("Kernel consumer destroying relayd %" PRIu64, index);
+
+ /* Get relayd reference if exists. */
+ relayd = consumer_find_relayd(index);
+ if (relayd == NULL) {
+ DBG("Unable to find relayd %" PRIu64, index);
+ ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
+ }
+
+ /*
+ * Each relayd socket pair has a refcount of stream attached to it
+ * which tells if the relayd is still active or not depending on the
+ * refcount value.
+ *
+ * This will set the destroy flag of the relayd object and destroy it
+ * if the refcount reaches zero when called.
+ *
+ * The destroy can happen either here or when a stream fd hangs up.
+ */
+ if (relayd) {
+ consumer_flag_relayd_for_destroy(relayd);
+ }
+
+ health_code_update();
+
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto error_fatal;
+ }
+
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_DATA_PENDING:
+ {
+ int32_t ret;
+ uint64_t id = msg.u.data_pending.session_id;
+
+ DBG("Kernel consumer data pending command for id %" PRIu64, id);
+
+ ret = consumer_data_pending(id);
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
+ if (ret < 0) {
+ PERROR("send data pending ret code");
+ goto error_fatal;
+ }
+
+ /*
+ * No need to send back a status message since the data pending
+ * returned value is the response.
+ */
+ break;
+ }
+ case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
+ {
+ struct lttng_consumer_channel *channel;
+ uint64_t key = msg.u.snapshot_channel.key;
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("Channel %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ } else {
+ pthread_mutex_lock(&channel->lock);
+ if (msg.u.snapshot_channel.metadata == 1) {
+ ret = lttng_kconsumer_snapshot_metadata(channel, key,
+ msg.u.snapshot_channel.pathname,
+ msg.u.snapshot_channel.relayd_id, ctx);
+ if (ret < 0) {
+ ERR("Snapshot metadata failed");
+ ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
+ }
+ } else {
+ ret = lttng_kconsumer_snapshot_channel(channel, key,
+ msg.u.snapshot_channel.pathname,
+ msg.u.snapshot_channel.relayd_id,
+ msg.u.snapshot_channel.nb_packets_per_stream,
+ ctx);
+ if (ret < 0) {
+ ERR("Snapshot channel failed");
+ ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
+ }
+ }
+ pthread_mutex_unlock(&channel->lock);
+ }
+ 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_DESTROY_CHANNEL:
+ {
+ uint64_t key = msg.u.destroy_channel.key;
+ struct lttng_consumer_channel *channel;
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
+ ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
+ }
+
+ health_code_update();
+
+ ret = consumer_send_status_msg(sock, ret_code);
+ if (ret < 0) {
+ /* Somehow, the session daemon is not responding anymore. */
+ goto end_destroy_channel;
+ }
+
+ health_code_update();
+
+ /* Stop right now if no channel was found. */
+ if (!channel) {
+ goto end_destroy_channel;
+ }
+
+ /*
+ * This command should ONLY be issued for channel with streams set in
+ * no monitor mode.
+ */
+ assert(!channel->monitor);
+
+ /*
+ * The refcount should ALWAYS be 0 in the case of a channel in no
+ * monitor mode.
+ */
+ assert(!uatomic_sub_return(&channel->refcount, 1));
+
+ consumer_del_channel(channel);
+end_destroy_channel:
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_DISCARDED_EVENTS:
+ {
+ ssize_t ret;
+ uint64_t count;
+ struct lttng_consumer_channel *channel;
+ uint64_t id = msg.u.discarded_events.session_id;
+ uint64_t key = msg.u.discarded_events.channel_key;
+
+ DBG("Kernel consumer discarded events command for session id %"
+ PRIu64 ", channel key %" PRIu64, id, key);
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("Kernel consumer discarded events channel %"
+ PRIu64 " not found", key);
+ count = 0;
+ } else {
+ count = channel->discarded_events;
+ }
+
+ health_code_update();
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
+ if (ret < 0) {
+ PERROR("send discarded events");
+ goto error_fatal;
+ }
+
+ break;
+ }
+ case LTTNG_CONSUMER_LOST_PACKETS:
+ {
+ ssize_t ret;
+ uint64_t count;
+ struct lttng_consumer_channel *channel;
+ uint64_t id = msg.u.lost_packets.session_id;
+ uint64_t key = msg.u.lost_packets.channel_key;
+
+ DBG("Kernel consumer lost packets command for session id %"
+ PRIu64 ", channel key %" PRIu64, id, key);
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("Kernel consumer lost packets channel %"
+ PRIu64 " not found", key);
+ count = 0;
+ } else {
+ count = channel->lost_packets;
+ }
+
+ 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)",
+ ret, subbuf_size);
+ write_index = 0;
+ }
+ break;
+ case CONSUMER_CHANNEL_MMAP:
+ {
+ const char *subbuf_addr;
+ struct lttng_buffer_view subbuf_view;
+
+ /* Get subbuffer size without padding */
+ err = kernctl_get_subbuf_size(infd, &subbuf_size);
+ 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;
+ }
+
+ ret = get_current_subbuf_addr(stream, &subbuf_addr);
+ if (ret) {
+ goto error_put_subbuf;
+ }
+
+ /* Make sure the tracer is not gone mad on us! */
+ assert(len >= subbuf_size);
+
+ padding = len - subbuf_size;
+
+ subbuf_view = lttng_buffer_view_init(subbuf_addr, 0, len);
+
+ /* write the subbuffer to the tracefile */
+ ret = lttng_consumer_on_read_subbuffer_mmap(
+ ctx, stream, &subbuf_view, padding, &index);
+ /*
+ * The mmap operation should write subbuf_size amount of data
+ * when network streaming or the full padding (len) size when we
+ * are _not_ streaming.
+ */
+ if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
+ (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
+ /*
+ * Display the error but continue processing to try to release the
+ * subbuffer. This is a DBG statement since this is possible to
+ * happen without being a critical error.
+ */
+ DBG("Error writing to tracefile "
+ "(ret: %zd != len: %lu != subbuf_size: %lu)",
+ ret, len, subbuf_size);
+ write_index = 0;
+ }
+ break;
+ }
+ default:
+ ERR("Unknown output method");
+ ret = -EPERM;
+ }
+error_put_subbuf:
+ err = kernctl_put_next_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;
+ }
+
+ /* Write index if needed. */
+ if (!write_index) {
+ goto rotate;
+ }
+
+ if (stream->chan->live_timer_interval && !stream->metadata_flag) {
+ /*
+ * In live, block until all the metadata is sent.
+ */
+ pthread_mutex_lock(&stream->metadata_timer_lock);
+ assert(!stream->missed_metadata_flush);
+ stream->waiting_on_metadata = true;
+ pthread_mutex_unlock(&stream->metadata_timer_lock);
+
+ err = consumer_stream_sync_metadata(ctx, stream->session_id);
+
+ pthread_mutex_lock(&stream->metadata_timer_lock);
+ stream->waiting_on_metadata = false;
+ if (stream->missed_metadata_flush) {
+ stream->missed_metadata_flush = false;
+ pthread_mutex_unlock(&stream->metadata_timer_lock);
+ (void) consumer_flush_kernel_index(stream);
+ } else {
+ pthread_mutex_unlock(&stream->metadata_timer_lock);
+ }
+ if (err < 0) {
+ goto error;
+ }
+ }
+
+ err = consumer_stream_write_index(stream, &index);
+ if (err < 0) {
+ goto error;
+ }
+
+rotate:
+ /*
+ * After extracting the packet, we check if the stream is now ready to be
+ * rotated and perform the action immediately.
+ */
+ rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
+ if (rotation_ret == 1) {
+ rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
+ if (rotation_ret < 0) {
+ ERR("Stream rotation error");
+ ret = -1;
+ goto error;
+ }
+ } else if (rotation_ret < 0) {
+ ERR("Checking if stream is ready to rotate");
+ ret = -1;
+ goto error;
+ }
+
+error:
+ return ret;
+}
+
+int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
+{
+ int ret;
+
+ assert(stream);
+
+ /*
+ * Don't create anything if this is set for streaming or if there is
+ * no current trace chunk on the parent channel.
+ */
+ if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
+ stream->chan->trace_chunk) {
+ ret = consumer_stream_create_output_files(stream, true);
+ if (ret) {
+ goto error;
+ }
+ }
+
+ if (stream->output == LTTNG_EVENT_MMAP) {
+ /* get the len of the mmap region */
+ unsigned long mmap_len;
+
+ ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
+ if (ret != 0) {
+ PERROR("kernctl_get_mmap_len");
+ goto error_close_fd;
+ }
+ stream->mmap_len = (size_t) mmap_len;
+
+ stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
+ MAP_PRIVATE, stream->wait_fd, 0);
+ if (stream->mmap_base == MAP_FAILED) {
+ PERROR("Error mmaping");
+ ret = -1;
+ goto error_close_fd;
+ }
+ }
+
+ /* we return 0 to let the library handle the FD internally */
+ return 0;
+
+error_close_fd:
+ if (stream->out_fd >= 0) {
+ int err;
+
+ err = close(stream->out_fd);
+ assert(!err);
+ stream->out_fd = -1;