+ */
+
+#define _LGPL_SOURCE
+#include <assert.h>
+#include <inttypes.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#include <common/common.h>
+#include <common/index/index.h>
+#include <common/kernel-consumer/kernel-consumer.h>
+#include <common/relayd/relayd.h>
+#include <common/ust-consumer/ust-consumer.h>
+#include <common/utils.h>
+#include <common/consumer/consumer.h>
+#include <common/consumer/consumer-timer.h>
+#include <common/consumer/metadata-bucket.h>
+#include <common/kernel-ctl/kernel-ctl.h>
+
+#include "consumer-stream.h"
+
+/*
+ * RCU call to free stream. MUST only be used with call_rcu().
+ */
+static void free_stream_rcu(struct rcu_head *head)
+{
+ struct lttng_ht_node_u64 *node =
+ caa_container_of(head, struct lttng_ht_node_u64, head);
+ struct lttng_consumer_stream *stream =
+ caa_container_of(node, struct lttng_consumer_stream, node);
+
+ pthread_mutex_destroy(&stream->lock);
+ free(stream);
+}
+
+static void consumer_stream_data_lock_all(struct lttng_consumer_stream *stream)
+{
+ pthread_mutex_lock(&stream->chan->lock);
+ pthread_mutex_lock(&stream->lock);
+}
+
+static void consumer_stream_data_unlock_all(struct lttng_consumer_stream *stream)
+{
+ pthread_mutex_unlock(&stream->lock);
+ pthread_mutex_unlock(&stream->chan->lock);
+}
+
+static void consumer_stream_metadata_lock_all(struct lttng_consumer_stream *stream)
+{
+ consumer_stream_data_lock_all(stream);
+ pthread_mutex_lock(&stream->metadata_rdv_lock);
+}
+
+static void consumer_stream_metadata_unlock_all(struct lttng_consumer_stream *stream)
+{
+ pthread_mutex_unlock(&stream->metadata_rdv_lock);
+ consumer_stream_data_unlock_all(stream);
+}
+
+/* Only used for data streams. */
+static int consumer_stream_update_stats(struct lttng_consumer_stream *stream,
+ const struct stream_subbuffer *subbuf)
+{
+ int ret = 0;
+ uint64_t sequence_number;
+ const uint64_t discarded_events = subbuf->info.data.events_discarded;
+
+ if (!subbuf->info.data.sequence_number.is_set) {
+ /* Command not supported by the tracer. */
+ sequence_number = -1ULL;
+ stream->sequence_number_unavailable = true;
+ } else {
+ sequence_number = subbuf->info.data.sequence_number.value;
+ }
+
+ /*
+ * 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 = sequence_number;
+ } else if (sequence_number > stream->last_sequence_number) {
+ stream->chan->lost_packets += sequence_number -
+ stream->last_sequence_number - 1;
+ } else {
+ /* seq <= last_sequence_number */
+ ERR("Sequence number inconsistent : prev = %" PRIu64
+ ", current = %" PRIu64,
+ stream->last_sequence_number, sequence_number);
+ ret = -1;
+ goto end;
+ }
+ stream->last_sequence_number = sequence_number;
+
+ if (discarded_events < 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_events;
+ } else {
+ stream->chan->discarded_events += discarded_events -
+ stream->last_discarded_events;
+ }
+ stream->last_discarded_events = discarded_events;
+ ret = 0;
+
+end:
+ return ret;
+}
+
+static
+void ctf_packet_index_populate(struct ctf_packet_index *index,
+ off_t offset, const struct stream_subbuffer *subbuffer)
+{
+ *index = (typeof(*index)){
+ .offset = htobe64(offset),
+ .packet_size = htobe64(subbuffer->info.data.packet_size),
+ .content_size = htobe64(subbuffer->info.data.content_size),
+ .timestamp_begin = htobe64(
+ subbuffer->info.data.timestamp_begin),
+ .timestamp_end = htobe64(
+ subbuffer->info.data.timestamp_end),
+ .events_discarded = htobe64(
+ subbuffer->info.data.events_discarded),
+ .stream_id = htobe64(subbuffer->info.data.stream_id),
+ .stream_instance_id = htobe64(
+ subbuffer->info.data.stream_instance_id.is_set ?
+ subbuffer->info.data.stream_instance_id.value : -1ULL),
+ .packet_seq_num = htobe64(
+ subbuffer->info.data.sequence_number.is_set ?
+ subbuffer->info.data.sequence_number.value : -1ULL),
+ };
+}
+
+static ssize_t consumer_stream_consume_mmap(
+ struct lttng_consumer_local_data *ctx,
+ struct lttng_consumer_stream *stream,
+ const struct stream_subbuffer *subbuffer)
+{
+ const unsigned long padding_size =
+ subbuffer->info.data.padded_subbuf_size -
+ subbuffer->info.data.subbuf_size;
+ const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_mmap(
+ stream, &subbuffer->buffer.buffer, padding_size);
+
+ if (stream->net_seq_idx == -1ULL) {
+ /*
+ * When writing on disk, check that only the subbuffer (no
+ * padding) was written to disk.
+ */
+ if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
+ DBG("Failed to write the entire padded subbuffer on disk (written_bytes: %zd, padded subbuffer size %lu)",
+ written_bytes,
+ subbuffer->info.data.padded_subbuf_size);
+ }
+ } else {
+ /*
+ * When streaming over the network, check that the entire
+ * subbuffer including padding was successfully written.
+ */
+ if (written_bytes != subbuffer->info.data.subbuf_size) {
+ DBG("Failed to write only the subbuffer over the network (written_bytes: %zd, subbuffer size %lu)",
+ written_bytes,
+ subbuffer->info.data.subbuf_size);
+ }
+ }
+
+ /*
+ * If `lttng_consumer_on_read_subbuffer_mmap()` returned an error, pass
+ * it along to the caller, else return zero.
+ */
+ if (written_bytes < 0) {
+ ERR("Error reading mmap subbuffer: %zd", written_bytes);
+ }
+
+ return written_bytes;
+}
+
+static ssize_t consumer_stream_consume_splice(
+ struct lttng_consumer_local_data *ctx,
+ struct lttng_consumer_stream *stream,
+ const struct stream_subbuffer *subbuffer)
+{
+ const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_splice(
+ ctx, stream, subbuffer->info.data.padded_subbuf_size, 0);
+
+ if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
+ DBG("Failed to write the entire padded subbuffer (written_bytes: %zd, padded subbuffer size %lu)",
+ written_bytes,
+ subbuffer->info.data.padded_subbuf_size);
+ }
+
+ /*
+ * If `lttng_consumer_on_read_subbuffer_splice()` returned an error,
+ * pass it along to the caller, else return zero.
+ */
+ if (written_bytes < 0) {
+ ERR("Error reading splice subbuffer: %zd", written_bytes);
+ }
+
+ return written_bytes;
+}
+
+static int consumer_stream_send_index(
+ struct lttng_consumer_stream *stream,
+ const struct stream_subbuffer *subbuffer,
+ struct lttng_consumer_local_data *ctx)
+{
+ off_t packet_offset = 0;
+ struct ctf_packet_index index = {};
+
+ /*
+ * This is called after consuming the sub-buffer; substract the
+ * effect this sub-buffer from the offset.
+ */
+ if (stream->net_seq_idx == (uint64_t) -1ULL) {
+ packet_offset = stream->out_fd_offset -
+ subbuffer->info.data.padded_subbuf_size;
+ }
+
+ ctf_packet_index_populate(&index, packet_offset, subbuffer);
+ return consumer_stream_write_index(stream, &index);
+}
+
+/*
+ * Actually do the metadata sync using the given metadata stream.
+ *
+ * Return 0 on success else a negative value. ENODATA can be returned also
+ * indicating that there is no metadata available for that stream.
+ */
+static int do_sync_metadata(struct lttng_consumer_stream *metadata,
+ struct lttng_consumer_local_data *ctx)
+{
+ int ret;
+ enum sync_metadata_status status;
+
+ assert(metadata);
+ assert(metadata->metadata_flag);
+ assert(ctx);
+
+ /*
+ * In UST, since we have to write the metadata from the cache packet
+ * by packet, we might need to start this procedure multiple times
+ * until all the metadata from the cache has been extracted.
+ */
+ do {
+ /*
+ * Steps :
+ * - Lock the metadata stream
+ * - Check if metadata stream node was deleted before locking.
+ * - if yes, release and return success
+ * - Check if new metadata is ready (flush + snapshot pos)
+ * - If nothing : release and return.
+ * - Lock the metadata_rdv_lock
+ * - Unlock the metadata stream
+ * - cond_wait on metadata_rdv to wait the wakeup from the
+ * metadata thread
+ * - Unlock the metadata_rdv_lock
+ */
+ pthread_mutex_lock(&metadata->lock);
+
+ /*
+ * There is a possibility that we were able to acquire a reference on the
+ * stream from the RCU hash table but between then and now, the node might
+ * have been deleted just before the lock is acquired. Thus, after locking,
+ * we make sure the metadata node has not been deleted which means that the
+ * buffers are closed.
+ *
+ * In that case, there is no need to sync the metadata hence returning a
+ * success return code.
+ */
+ ret = cds_lfht_is_node_deleted(&metadata->node.node);
+ if (ret) {
+ ret = 0;
+ goto end_unlock_mutex;
+ }
+
+ switch (ctx->type) {
+ case LTTNG_CONSUMER_KERNEL:
+ /*
+ * Empty the metadata cache and flush the current stream.
+ */
+ status = lttng_kconsumer_sync_metadata(metadata);
+ break;
+ case LTTNG_CONSUMER32_UST:
+ case LTTNG_CONSUMER64_UST:
+ /*
+ * Ask the sessiond if we have new metadata waiting and update the
+ * consumer metadata cache.
+ */
+ status = lttng_ustconsumer_sync_metadata(ctx, metadata);
+ break;
+ default:
+ abort();
+ }
+
+ switch (status) {
+ case SYNC_METADATA_STATUS_NEW_DATA:
+ break;
+ case SYNC_METADATA_STATUS_NO_DATA:
+ ret = 0;
+ goto end_unlock_mutex;
+ case SYNC_METADATA_STATUS_ERROR:
+ ret = -1;
+ goto end_unlock_mutex;
+ default:
+ abort();
+ }
+
+ /*
+ * At this point, new metadata have been flushed, so we wait on the
+ * rendez-vous point for the metadata thread to wake us up when it
+ * finishes consuming the metadata and continue execution.
+ */
+
+ pthread_mutex_lock(&metadata->metadata_rdv_lock);
+
+ /*
+ * Release metadata stream lock so the metadata thread can process it.
+ */
+ pthread_mutex_unlock(&metadata->lock);
+
+ /*
+ * Wait on the rendez-vous point. Once woken up, it means the metadata was
+ * consumed and thus synchronization is achieved.
+ */
+ pthread_cond_wait(&metadata->metadata_rdv, &metadata->metadata_rdv_lock);
+ pthread_mutex_unlock(&metadata->metadata_rdv_lock);
+ } while (status == SYNC_METADATA_STATUS_NEW_DATA);
+
+ /* Success */
+ return 0;
+
+end_unlock_mutex:
+ pthread_mutex_unlock(&metadata->lock);
+ return ret;
+}
+
+/*
+ * Synchronize the metadata using a given session ID. A successful acquisition
+ * of a metadata stream will trigger a request to the session daemon and a
+ * snapshot so the metadata thread can consume it.
+ *
+ * This function call is a rendez-vous point between the metadata thread and
+ * the data thread.