X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=567babadf177ade2912666c2a14ec0ee368ea7d2;hb=4c5691ba0756f7c0f4a604c814ee729e967da8e0;hp=ccf9b40386ab306b8130ea073b62d8a07bd343e7;hpb=f0867b598fe51e99257be877c5a4fe74e9a165d5;p=lttng-tools.git

diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c
index ccf9b4038..567babadf 100644
--- a/src/common/consumer/consumer.c
+++ b/src/common/consumer/consumer.c
@@ -323,6 +323,7 @@ static void free_relayd_rcu(struct rcu_head *head)
 	(void) relayd_close(&relayd->control_sock);
 	(void) relayd_close(&relayd->data_sock);
 
+	pthread_mutex_destroy(&relayd->ctrl_sock_mutex);
 	free(relayd);
 }
 
@@ -464,14 +465,13 @@ static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
  * If a local data context is available, notify the threads that the streams'
  * state have changed.
  */
-static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
-		struct lttng_consumer_local_data *ctx)
+void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd)
 {
 	uint64_t netidx;
 
 	assert(relayd);
 
-	DBG("Cleaning up relayd sockets");
+	DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx);
 
 	/* Save the net sequence index before destroying the object */
 	netidx = relayd->net_seq_idx;
@@ -491,10 +491,8 @@ static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
 	 * memory barrier ordering the updates of the end point status from the
 	 * read of this status which happens AFTER receiving this notify.
 	 */
-	if (ctx) {
-		notify_thread_lttng_pipe(ctx->consumer_data_pipe);
-		notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
-	}
+	notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe);
+	notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe);
 }
 
 /*
@@ -802,6 +800,8 @@ int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
 				stream->chan->tracefile_size, stream->chan->tracefile_count);
 		pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 		if (ret < 0) {
+			ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+			lttng_consumer_cleanup_relayd(relayd);
 			goto end;
 		}
 
@@ -843,6 +843,8 @@ int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
 		ret = relayd_streams_sent(&relayd->control_sock);
 		pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 		if (ret < 0) {
+			ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+			lttng_consumer_cleanup_relayd(relayd);
 			goto end;
 		}
 	} else {
@@ -1077,7 +1079,7 @@ int consumer_add_channel(struct lttng_consumer_channel *channel,
  */
 static int update_poll_array(struct lttng_consumer_local_data *ctx,
 		struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
-		struct lttng_ht *ht)
+		struct lttng_ht *ht, int *nb_inactive_fd)
 {
 	int i = 0;
 	struct lttng_ht_iter iter;
@@ -1089,6 +1091,7 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx,
 	assert(local_stream);
 
 	DBG("Updating poll fd array");
+	*nb_inactive_fd = 0;
 	rcu_read_lock();
 	cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
 		/*
@@ -1099,9 +1102,14 @@ static int update_poll_array(struct lttng_consumer_local_data *ctx,
 		 * just after the check. However, this is OK since the stream(s) will
 		 * be deleted once the thread is notified that the end point state has
 		 * changed where this function will be called back again.
+		 *
+		 * We track the number of inactive FDs because they still need to be
+		 * closed by the polling thread after a wakeup on the data_pipe or
+		 * metadata_pipe.
 		 */
 		if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
 				stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
+			(*nb_inactive_fd)++;
 			continue;
 		}
 		/*
@@ -1704,7 +1712,8 @@ write_error:
 	 * cleanup the relayd object and all associated streams.
 	 */
 	if (relayd && relayd_hang_up) {
-		cleanup_relayd(relayd, ctx);
+		ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+		lttng_consumer_cleanup_relayd(relayd);
 	}
 
 end:
@@ -1930,7 +1939,8 @@ write_error:
 	 * cleanup the relayd object and all associated streams.
 	 */
 	if (relayd && relayd_hang_up) {
-		cleanup_relayd(relayd, ctx);
+		ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+		lttng_consumer_cleanup_relayd(relayd);
 		/* Skip splice error so the consumer does not fail */
 		goto end;
 	}
@@ -2153,7 +2163,7 @@ int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
 
 	lttng_ht_add_unique_u64(ht, &stream->node);
 
-	lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
+	lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
 		&stream->node_channel_id);
 
 	/*
@@ -2457,6 +2467,8 @@ void *consumer_thread_data_poll(void *data)
 	struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
 	/* local view of consumer_data.fds_count */
 	int nb_fd = 0;
+	/* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
+	int nb_inactive_fd = 0;
 	struct lttng_consumer_local_data *ctx = data;
 	ssize_t len;
 
@@ -2513,7 +2525,7 @@ void *consumer_thread_data_poll(void *data)
 				goto end;
 			}
 			ret = update_poll_array(ctx, &pollfd, local_stream,
-					data_ht);
+					data_ht, &nb_inactive_fd);
 			if (ret < 0) {
 				ERR("Error in allocating pollfd or local_outfds");
 				lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
@@ -2526,7 +2538,8 @@ void *consumer_thread_data_poll(void *data)
 		pthread_mutex_unlock(&consumer_data.lock);
 
 		/* No FDs and consumer_quit, consumer_cleanup the thread */
-		if (nb_fd == 0 && CMM_LOAD_SHARED(consumer_quit) == 1) {
+		if (nb_fd == 0 && nb_inactive_fd == 0 &&
+				CMM_LOAD_SHARED(consumer_quit) == 1) {
 			err = 0;	/* All is OK */
 			goto end;
 		}
@@ -3360,7 +3373,7 @@ error:
  * This will create a relayd socket pair and add it to the relayd hash table.
  * The caller MUST acquire a RCU read side lock before calling it.
  */
-int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
+ void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
 		struct lttng_consumer_local_data *ctx, int sock,
 		struct pollfd *consumer_sockpoll,
 		struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
@@ -3513,10 +3526,11 @@ int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
 	 * Add relayd socket pair to consumer data hashtable. If object already
 	 * exists or on error, the function gracefully returns.
 	 */
+	relayd->ctx = ctx;
 	add_relayd(relayd);
 
 	/* All good! */
-	return 0;
+	return;
 
 error:
 	if (consumer_send_status_msg(sock, ret_code) < 0) {
@@ -3534,8 +3548,6 @@ error_nosignal:
 	if (relayd_created) {
 		free(relayd);
 	}
-
-	return ret;
 }
 
 /*
@@ -3641,6 +3653,8 @@ int consumer_data_pending(uint64_t id)
 		pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 		if (ret < 0) {
 			/* Communication error thus the relayd so no data pending. */
+			ERR("Relayd begin data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+			lttng_consumer_cleanup_relayd(relayd);
 			goto data_not_pending;
 		}
 	}
@@ -3682,6 +3696,13 @@ int consumer_data_pending(uint64_t id)
 						stream->relayd_stream_id,
 						stream->next_net_seq_num - 1);
 			}
+			if (ret < 0) {
+				ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+				lttng_consumer_cleanup_relayd(relayd);
+				pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
+				pthread_mutex_unlock(&stream->lock);
+				goto data_not_pending;
+			}
 			pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 			if (ret == 1) {
 				pthread_mutex_unlock(&stream->lock);
@@ -3700,6 +3721,8 @@ int consumer_data_pending(uint64_t id)
 				relayd->relayd_session_id, &is_data_inflight);
 		pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 		if (ret < 0) {
+			ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
+			lttng_consumer_cleanup_relayd(relayd);
 			goto data_not_pending;
 		}
 		if (is_data_inflight) {