X-Git-Url: http://git.efficios.com/?a=blobdiff_plain;f=src%2Fcommon%2Fconsumer%2Fconsumer.c;h=cb1414979c619a19f4eadd92a71088b7f83038f8;hb=b789125113ff140dc34d0bc3b9cec3967300e9cc;hp=b2feb3945c8a40f3c1476e4dba7d9b659b8f9f8b;hpb=cf0bcb51ea857687a353d2851e572dba6cc63cb0;p=lttng-tools.git

diff --git a/src/common/consumer/consumer.c b/src/common/consumer/consumer.c
index b2feb3945..cb1414979 100644
--- a/src/common/consumer/consumer.c
+++ b/src/common/consumer/consumer.c
@@ -67,6 +67,7 @@ struct consumer_channel_msg {
 	uint64_t key;				/* del */
 };
 
+/* Flag used to temporarily pause data consumption from testpoints. */
 int data_consumption_paused;
 
 /*
@@ -75,7 +76,7 @@ int data_consumption_paused;
  * Also updated by the signal handler (consumer_should_exit()). Read by the
  * polling threads.
  */
-volatile int consumer_quit;
+int consumer_quit;
 
 /*
  * Global hash table containing respectively metadata and data streams. The
@@ -322,6 +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);
 }
 
@@ -463,14 +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;
@@ -490,10 +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);
 }
 
 /*
@@ -801,6 +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;
 		}
 
@@ -842,6 +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 {
@@ -1076,7 +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;
@@ -1088,6 +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) {
 		/*
@@ -1098,9 +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;
 		}
 		/*
@@ -1226,7 +1235,7 @@ void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
 {
 	ssize_t ret;
 
-	consumer_quit = 1;
+	CMM_STORE_SHARED(consumer_quit, 1);
 	ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
 	if (ret < 1) {
 		PERROR("write consumer quit");
@@ -1703,7 +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:
@@ -1929,7 +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;
 	}
@@ -2152,7 +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);
 
 	/*
@@ -2456,6 +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;
 
@@ -2512,7 +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);
@@ -2525,7 +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 && consumer_quit == 1) {
+		if (nb_fd == 0 && nb_inactive_fd == 0 &&
+				CMM_LOAD_SHARED(consumer_quit) == 1) {
 			err = 0;	/* All is OK */
 			goto end;
 		}
@@ -2554,6 +2568,12 @@ void *consumer_thread_data_poll(void *data)
 			goto end;
 		}
 
+		if (caa_unlikely(data_consumption_paused)) {
+			DBG("Data consumption paused, sleeping...");
+			sleep(1);
+			goto restart;
+		}
+
 		/*
 		 * If the consumer_data_pipe triggered poll go directly to the
 		 * beginning of the loop to update the array. We want to prioritize
@@ -3196,7 +3216,7 @@ void *consumer_thread_sessiond_poll(void *data)
 			err = 0;
 			goto end;
 		}
-		if (consumer_quit) {
+		if (CMM_LOAD_SHARED(consumer_quit)) {
 			DBG("consumer_thread_receive_fds received quit from signal");
 			err = 0;	/* All is OK */
 			goto end;
@@ -3221,7 +3241,7 @@ end:
 	 * when all fds have hung up, the polling thread
 	 * can exit cleanly
 	 */
-	consumer_quit = 1;
+	CMM_STORE_SHARED(consumer_quit, 1);
 
 	/*
 	 * Notify the data poll thread to poll back again and test the
@@ -3353,7 +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,
@@ -3375,7 +3395,6 @@ int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
 		/* Not found. Allocate one. */
 		relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
 		if (relayd == NULL) {
-			ret = -ENOMEM;
 			ret_code = LTTCOMM_CONSUMERD_ENOMEM;
 			goto error;
 		} else {
@@ -3408,14 +3427,12 @@ int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
 	if (ret) {
 		/* Needing to exit in the middle of a command: error. */
 		lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
-		ret = -EINTR;
 		goto error_nosignal;
 	}
 
 	/* Get relayd socket from session daemon */
 	ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 	if (ret != sizeof(fd)) {
-		ret = -1;
 		fd = -1;	/* Just in case it gets set with an invalid value. */
 
 		/*
@@ -3489,7 +3506,6 @@ int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
 		break;
 	default:
 		ERR("Unknown relayd socket type (%d)", sock_type);
-		ret = -1;
 		ret_code = LTTCOMM_CONSUMERD_FATAL;
 		goto error;
 	}
@@ -3510,10 +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) {
@@ -3531,36 +3548,6 @@ error_nosignal:
 	if (relayd_created) {
 		free(relayd);
 	}
-
-	return ret;
-}
-
-/*
- * Try to lock the stream mutex.
- *
- * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
- */
-static int stream_try_lock(struct lttng_consumer_stream *stream)
-{
-	int ret;
-
-	assert(stream);
-
-	/*
-	 * Try to lock the stream mutex. On failure, we know that the stream is
-	 * being used else where hence there is data still being extracted.
-	 */
-	ret = pthread_mutex_trylock(&stream->lock);
-	if (ret) {
-		/* For both EBUSY and EINVAL error, the mutex is NOT locked. */
-		ret = 0;
-		goto end;
-	}
-
-	ret = 1;
-
-end:
-	return ret;
 }
 
 /*
@@ -3629,28 +3616,11 @@ int consumer_data_pending(uint64_t id)
 	/* Ease our life a bit */
 	ht = consumer_data.stream_list_ht;
 
-	relayd = find_relayd_by_session_id(id);
-	if (relayd) {
-		/* Send init command for data pending. */
-		pthread_mutex_lock(&relayd->ctrl_sock_mutex);
-		ret = relayd_begin_data_pending(&relayd->control_sock,
-				relayd->relayd_session_id);
-		pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
-		if (ret < 0) {
-			/* Communication error thus the relayd so no data pending. */
-			goto data_not_pending;
-		}
-	}
-
 	cds_lfht_for_each_entry_duplicate(ht->ht,
 			ht->hash_fct(&id, lttng_ht_seed),
 			ht->match_fct, &id,
 			&iter.iter, stream, node_session_id.node) {
-		/* If this call fails, the stream is being used hence data pending. */
-		ret = stream_try_lock(stream);
-		if (!ret) {
-			goto data_pending;
-		}
+		pthread_mutex_lock(&stream->lock);
 
 		/*
 		 * A removed node from the hash table indicates that the stream has
@@ -3668,9 +3638,27 @@ int consumer_data_pending(uint64_t id)
 			}
 		}
 
-		/* Relayd check */
-		if (relayd) {
-			pthread_mutex_lock(&relayd->ctrl_sock_mutex);
+		pthread_mutex_unlock(&stream->lock);
+	}
+
+	relayd = find_relayd_by_session_id(id);
+	if (relayd) {
+		unsigned int is_data_inflight = 0;
+
+		/* Send init command for data pending. */
+		pthread_mutex_lock(&relayd->ctrl_sock_mutex);
+		ret = relayd_begin_data_pending(&relayd->control_sock,
+				relayd->relayd_session_id);
+		if (ret < 0) {
+			pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
+			/* Communication error thus the relayd so no data pending. */
+			goto data_not_pending;
+		}
+
+		cds_lfht_for_each_entry_duplicate(ht->ht,
+				ht->hash_fct(&id, lttng_ht_seed),
+				ht->match_fct, &id,
+				&iter.iter, stream, node_session_id.node) {
 			if (stream->metadata_flag) {
 				ret = relayd_quiescent_control(&relayd->control_sock,
 						stream->relayd_stream_id);
@@ -3679,24 +3667,25 @@ int consumer_data_pending(uint64_t id)
 						stream->relayd_stream_id,
 						stream->next_net_seq_num - 1);
 			}
-			pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
+
 			if (ret == 1) {
-				pthread_mutex_unlock(&stream->lock);
+				pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 				goto data_pending;
+			} else 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);
+				goto data_not_pending;
 			}
 		}
-		pthread_mutex_unlock(&stream->lock);
-	}
 
-	if (relayd) {
-		unsigned int is_data_inflight = 0;
-
-		/* Send init command for data pending. */
-		pthread_mutex_lock(&relayd->ctrl_sock_mutex);
+		/* Send end command for data pending. */
 		ret = relayd_end_data_pending(&relayd->control_sock,
 				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) {