+ ret = send_streams_to_thread(metadata, ctx);
+ if (ret < 0) {
+ /*
+ * If we are unable to send the stream to the thread, there is
+ * a big problem so just stop everything.
+ */
+ ret = LTTCOMM_CONSUMERD_FATAL;
+ goto error;
+ }
+ /* List MUST be empty after or else it could be reused. */
+ assert(cds_list_empty(&metadata->streams.head));
+
+ ret = 0;
+ goto end;
+
+error:
+ /*
+ * Delete metadata channel on error. At this point, the metadata stream can
+ * NOT be monitored by the metadata thread thus having the guarantee that
+ * the stream is still in the local stream list of the channel. This call
+ * will make sure to clean that list.
+ */
+ cds_list_del(&metadata->metadata_stream->send_node);
+ consumer_stream_destroy(metadata->metadata_stream, NULL);
+error_no_stream:
+end:
+ return ret;
+}
+
+/*
+ * Snapshot the whole metadata.
+ *
+ * Returns 0 on success, < 0 on error
+ */
+static int snapshot_metadata(uint64_t key, char *path, uint64_t relayd_id,
+ struct lttng_consumer_local_data *ctx)
+{
+ int ret = 0;
+ ssize_t write_len;
+ uint64_t total_len = 0;
+ struct lttng_consumer_channel *metadata_channel;
+ struct lttng_consumer_stream *metadata_stream;
+
+ assert(path);
+ assert(ctx);
+
+ DBG("UST consumer snapshot metadata with key %" PRIu64 " at path %s",
+ key, path);
+
+ rcu_read_lock();
+
+ metadata_channel = consumer_find_channel(key);
+ if (!metadata_channel) {
+ ERR("UST snapshot metadata channel not found for key %lu", key);
+ ret = -1;
+ goto error;
+ }
+ assert(!metadata_channel->monitor);
+
+ /*
+ * Ask the sessiond if we have new metadata waiting and update the
+ * consumer metadata cache.
+ */
+ ret = lttng_ustconsumer_request_metadata(ctx, metadata_channel);
+ if (ret < 0) {
+ goto error;
+ }
+
+ /*
+ * The metadata stream is NOT created in no monitor mode when the channel
+ * is created on a sessiond ask channel command.
+ */
+ ret = create_ust_streams(metadata_channel, ctx);
+ if (ret < 0) {
+ goto error;
+ }
+
+ metadata_stream = metadata_channel->metadata_stream;
+ assert(metadata_stream);
+
+ if (relayd_id != (uint64_t) -1ULL) {
+ metadata_stream->net_seq_idx = relayd_id;
+ ret = consumer_send_relayd_stream(metadata_stream, path);
+ if (ret < 0) {
+ goto error_stream;
+ }
+ } else {
+ ret = utils_create_stream_file(path, metadata_stream->name,
+ metadata_stream->chan->tracefile_size,
+ metadata_stream->tracefile_count_current,
+ metadata_stream->uid, metadata_stream->gid);
+ if (ret < 0) {
+ goto error_stream;
+ }
+ metadata_stream->out_fd = ret;
+ metadata_stream->tracefile_size_current = 0;
+ }
+
+ pthread_mutex_lock(&metadata_channel->metadata_cache->lock);
+ while (total_len < metadata_channel->metadata_cache->total_bytes_written) {
+ /*
+ * Write at most one packet of metadata into the channel
+ * to avoid blocking here.
+ */
+ write_len = ustctl_write_one_packet_to_channel(metadata_channel->uchan,
+ metadata_channel->metadata_cache->data,
+ metadata_channel->metadata_cache->total_bytes_written);
+ if (write_len < 0) {
+ ERR("UST consumer snapshot writing metadata packet");
+ ret = -1;
+ goto error_unlock;
+ }
+ total_len += write_len;
+
+ DBG("Written %" PRIu64 " bytes to metadata (left: %" PRIu64 ")",
+ write_len,
+ metadata_channel->metadata_cache->total_bytes_written - write_len);
+ ustctl_flush_buffer(metadata_stream->ustream, 1);
+ ret = lttng_consumer_read_subbuffer(metadata_stream, ctx);
+ if (ret < 0) {
+ goto error_unlock;
+ }
+ }
+
+error_unlock:
+ pthread_mutex_unlock(&metadata_channel->metadata_cache->lock);
+
+error_stream:
+ /*
+ * Clean up the stream completly because the next snapshot will use a new
+ * metadata stream.
+ */
+ cds_list_del(&metadata_stream->send_node);
+ consumer_stream_destroy(metadata_stream, NULL);
+ metadata_channel->metadata_stream = NULL;
+
+error:
+ rcu_read_unlock();
+ return ret;
+}
+
+/*
+ * Take a snapshot of all the stream of a channel.
+ *
+ * Returns 0 on success, < 0 on error
+ */
+static int snapshot_channel(uint64_t key, char *path, uint64_t relayd_id,
+ uint64_t max_stream_size, struct lttng_consumer_local_data *ctx)
+{
+ int ret;
+ unsigned use_relayd = 0;
+ unsigned long consumed_pos, produced_pos;
+ struct lttng_consumer_channel *channel;
+ struct lttng_consumer_stream *stream;
+
+ assert(path);
+ assert(ctx);
+
+ rcu_read_lock();
+
+ if (relayd_id != (uint64_t) -1ULL) {
+ use_relayd = 1;
+ }
+
+ channel = consumer_find_channel(key);
+ if (!channel) {
+ ERR("UST snapshot channel not found for key %lu", key);
+ ret = -1;
+ goto error;
+ }
+ assert(!channel->monitor);
+ DBG("UST consumer snapshot channel %lu", key);
+
+ cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
+ /* Lock stream because we are about to change its state. */
+ pthread_mutex_lock(&stream->lock);
+ stream->net_seq_idx = relayd_id;
+
+ if (use_relayd) {
+ ret = consumer_send_relayd_stream(stream, path);
+ if (ret < 0) {
+ goto error_unlock;
+ }
+ } else {
+ ret = utils_create_stream_file(path, stream->name,
+ stream->chan->tracefile_size,
+ stream->tracefile_count_current,
+ stream->uid, stream->gid);
+ if (ret < 0) {
+ goto error_unlock;
+ }
+ stream->out_fd = ret;
+ stream->tracefile_size_current = 0;
+
+ DBG("UST consumer snapshot stream %s/%s (%" PRIu64 ")", path,
+ stream->name, stream->key);
+ }
+
+ ustctl_flush_buffer(stream->ustream, 1);
+
+ ret = lttng_ustconsumer_take_snapshot(stream);
+ if (ret < 0) {
+ ERR("Taking UST snapshot");
+ goto error_unlock;
+ }
+
+ ret = lttng_ustconsumer_get_produced_snapshot(stream, &produced_pos);
+ if (ret < 0) {
+ ERR("Produced UST snapshot position");
+ goto error_unlock;
+ }
+
+ ret = lttng_ustconsumer_get_consumed_snapshot(stream, &consumed_pos);
+ if (ret < 0) {
+ ERR("Consumerd UST snapshot position");
+ goto error_unlock;
+ }
+
+ /*
+ * The original value is sent back if max stream size is larger than
+ * the possible size of the snapshot. Also, we asume that the session
+ * daemon should never send a maximum stream size that is lower than
+ * subbuffer size.
+ */
+ consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
+ produced_pos, max_stream_size);
+
+ while (consumed_pos < produced_pos) {
+ ssize_t read_len;
+ unsigned long len, padded_len;
+
+ DBG("UST consumer taking snapshot at pos %lu", consumed_pos);
+
+ ret = ustctl_get_subbuf(stream->ustream, &consumed_pos);
+ if (ret < 0) {
+ if (ret != -EAGAIN) {
+ PERROR("ustctl_get_subbuf snapshot");
+ goto error_close_stream;
+ }
+ DBG("UST consumer get subbuf failed. Skipping it.");
+ consumed_pos += stream->max_sb_size;
+ continue;
+ }
+
+ ret = ustctl_get_subbuf_size(stream->ustream, &len);
+ if (ret < 0) {
+ ERR("Snapshot ustctl_get_subbuf_size");
+ goto error_put_subbuf;
+ }
+
+ ret = ustctl_get_padded_subbuf_size(stream->ustream, &padded_len);
+ if (ret < 0) {
+ ERR("Snapshot ustctl_get_padded_subbuf_size");
+ goto error_put_subbuf;
+ }
+
+ read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
+ padded_len - len);
+ if (use_relayd) {
+ if (read_len != len) {
+ ret = -1;
+ goto error_put_subbuf;
+ }
+ } else {
+ if (read_len != padded_len) {
+ ret = -1;
+ goto error_put_subbuf;
+ }
+ }
+
+ ret = ustctl_put_subbuf(stream->ustream);
+ if (ret < 0) {
+ ERR("Snapshot ustctl_put_subbuf");
+ goto error_close_stream;
+ }
+ consumed_pos += stream->max_sb_size;
+ }
+
+ /* Simply close the stream so we can use it on the next snapshot. */
+ consumer_stream_close(stream);
+ pthread_mutex_unlock(&stream->lock);
+ }
+
+ rcu_read_unlock();
+ return 0;
+
+error_put_subbuf:
+ if (ustctl_put_subbuf(stream->ustream) < 0) {
+ ERR("Snapshot ustctl_put_subbuf");
+ }
+error_close_stream:
+ consumer_stream_close(stream);
+error_unlock:
+ pthread_mutex_unlock(&stream->lock);
+error:
+ rcu_read_unlock();
+ return ret;
+}
+
+/*
+ * Receive the metadata updates from the sessiond.
+ */
+int lttng_ustconsumer_recv_metadata(int sock, uint64_t key, uint64_t offset,
+ uint64_t len, struct lttng_consumer_channel *channel)
+{
+ int ret, ret_code = LTTNG_OK;
+ char *metadata_str;
+
+ DBG("UST consumer push metadata key %" PRIu64 " of len %" PRIu64, key, len);
+
+ metadata_str = zmalloc(len * sizeof(char));
+ if (!metadata_str) {
+ PERROR("zmalloc metadata string");
+ ret_code = LTTCOMM_CONSUMERD_ENOMEM;
+ goto end;
+ }
+
+ /* Receive metadata string. */
+ ret = lttcomm_recv_unix_sock(sock, metadata_str, len);
+ if (ret < 0) {
+ /* Session daemon is dead so return gracefully. */
+ ret_code = ret;
+ goto end_free;
+ }
+
+ /*
+ * XXX: The consumer data lock is acquired before calling metadata cache
+ * write which calls push metadata that MUST be protected by the consumer
+ * lock in order to be able to check the validity of the metadata stream of
+ * the channel.
+ *
+ * Note that this will be subject to change to better fine grained locking
+ * and ultimately try to get rid of this global consumer data lock.
+ */
+ pthread_mutex_lock(&consumer_data.lock);
+
+ pthread_mutex_lock(&channel->metadata_cache->lock);
+ ret = consumer_metadata_cache_write(channel, offset, len, metadata_str);
+ if (ret < 0) {
+ /* Unable to handle metadata. Notify session daemon. */
+ ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
+ /*
+ * Skip metadata flush on write error since the offset and len might
+ * not have been updated which could create an infinite loop below when
+ * waiting for the metadata cache to be flushed.
+ */
+ pthread_mutex_unlock(&channel->metadata_cache->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
+ goto end_free;
+ }
+ pthread_mutex_unlock(&channel->metadata_cache->lock);
+ pthread_mutex_unlock(&consumer_data.lock);
+
+ while (consumer_metadata_cache_flushed(channel, offset + len)) {
+ DBG("Waiting for metadata to be flushed");
+ usleep(DEFAULT_METADATA_AVAILABILITY_WAIT_TIME);
+ }
+
+end_free:
+ free(metadata_str);
+end:
+ return ret_code;
+}
+
+/*
+ * Receive command from session daemon and process it.
+ *
+ * Return 1 on success else a negative value or 0.
+ */
+int lttng_ustconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
+ int sock, struct pollfd *consumer_sockpoll)
+{
+ ssize_t ret;
+ enum lttng_error_code ret_code = LTTNG_OK;
+ struct lttcomm_consumer_msg msg;
+ struct lttng_consumer_channel *channel = NULL;
+
+ ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
+ if (ret != sizeof(msg)) {
+ DBG("Consumer received unexpected message size %zd (expects %zu)",
+ ret, sizeof(msg));
+ /*
+ * The ret value might 0 meaning an orderly shutdown but this is ok
+ * since the caller handles this.
+ */
+ if (ret > 0) {
+ lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
+ ret = -1;
+ }
+ return ret;
+ }
+ if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
+ /*
+ * Notify the session daemon that the command is completed.
+ *
+ * On transport layer error, the function call will print an error
+ * message so handling the returned code is a bit useless since we
+ * return an error code anyway.
+ */
+ (void) consumer_send_status_msg(sock, ret_code);
+ return -ENOENT;
+ }
+
+ /* relayd needs RCU read-side lock */
+ rcu_read_lock();
+
+ switch (msg.cmd_type) {
+ case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
+ {
+ /* Session daemon status message are handled in the following call. */
+ ret = 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);
+ goto end_nosignal;
+ }
+ case LTTNG_CONSUMER_DESTROY_RELAYD:
+ {
+ uint64_t index = msg.u.destroy_relayd.net_seq_idx;
+ struct consumer_relayd_sock_pair *relayd;
+
+ DBG("UST 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 = LTTNG_ERR_NO_CONSUMER;
+ }
+
+ /*
+ * 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);
+ }
+
+ goto end_msg_sessiond;
+ }
+ case LTTNG_CONSUMER_UPDATE_STREAM:
+ {
+ rcu_read_unlock();
+ return -ENOSYS;
+ }
+ case LTTNG_CONSUMER_DATA_PENDING:
+ {
+ int ret, is_data_pending;
+ uint64_t id = msg.u.data_pending.session_id;
+
+ DBG("UST consumer data pending command for id %" PRIu64, id);
+
+ is_data_pending = consumer_data_pending(id);
+
+ /* Send back returned value to session daemon */
+ ret = lttcomm_send_unix_sock(sock, &is_data_pending,
+ sizeof(is_data_pending));
+ if (ret < 0) {
+ DBG("Error when sending the data pending ret code: %d", ret);
+ goto error_fatal;
+ }
+
+ /*
+ * No need to send back a status message since the data pending
+ * returned value is the response.
+ */
+ break;
+ }
+ case LTTNG_CONSUMER_ASK_CHANNEL_CREATION:
+ {
+ int ret;
+ struct ustctl_consumer_channel_attr attr;
+
+ /* Create a plain object and reserve a channel key. */
+ channel = allocate_channel(msg.u.ask_channel.session_id,
+ msg.u.ask_channel.pathname, msg.u.ask_channel.name,
+ msg.u.ask_channel.uid, msg.u.ask_channel.gid,
+ msg.u.ask_channel.relayd_id, msg.u.ask_channel.key,
+ (enum lttng_event_output) msg.u.ask_channel.output,
+ msg.u.ask_channel.tracefile_size,
+ msg.u.ask_channel.tracefile_count,
+ msg.u.ask_channel.session_id_per_pid,
+ msg.u.ask_channel.monitor);
+ if (!channel) {
+ goto end_channel_error;
+ }
+
+ /* Build channel attributes from received message. */
+ attr.subbuf_size = msg.u.ask_channel.subbuf_size;
+ attr.num_subbuf = msg.u.ask_channel.num_subbuf;
+ attr.overwrite = msg.u.ask_channel.overwrite;
+ attr.switch_timer_interval = msg.u.ask_channel.switch_timer_interval;
+ attr.read_timer_interval = msg.u.ask_channel.read_timer_interval;
+ attr.chan_id = msg.u.ask_channel.chan_id;
+ attr.output = msg.u.ask_channel.output;
+ memcpy(attr.uuid, msg.u.ask_channel.uuid, sizeof(attr.uuid));
+
+ /* Translate and save channel type. */
+ switch (msg.u.ask_channel.type) {
+ case LTTNG_UST_CHAN_PER_CPU:
+ channel->type = CONSUMER_CHANNEL_TYPE_DATA;
+ attr.type = LTTNG_UST_CHAN_PER_CPU;
+ /*
+ * Set refcount to 1 for owner. Below, we will
+ * pass ownership to the
+ * consumer_thread_channel_poll() thread.
+ */
+ channel->refcount = 1;
+ break;
+ case LTTNG_UST_CHAN_METADATA:
+ channel->type = CONSUMER_CHANNEL_TYPE_METADATA;
+ attr.type = LTTNG_UST_CHAN_METADATA;
+ break;
+ default:
+ assert(0);
+ goto error_fatal;
+ };