Fix: relayd: unbalanced RCU read lock/unlock

[lttng-tools.git] / src / bin / lttng-relayd / main.c

diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c
index ac30351522d159b744d3ee834fbb25cb5c6dd05b..6101fdb2043bf079f53eed3bdbb94cdf5ec26e25 100644 (file)
--- a/src/bin/lttng-relayd/main.c
+++ b/src/bin/lttng-relayd/main.c
@@ -1065,6 +1065,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
        stream->session_id = session->id;
        stream->index_fd = -1;
        stream->read_index_fd = -1;
+       stream->ctf_stream_id = -1ULL;
        lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
        pthread_mutex_init(&stream->lock, NULL);
 
@@ -1801,9 +1802,10 @@ int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
                DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
 
                /*
-                * Only flag a stream inactive when it has already received data.
+                * Only flag a stream inactive when it has already received data
+                * and no indexes are in flight.
                 */
-               if (stream->total_index_received > 0) {
+               if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
                        stream->beacon_ts_end = be64toh(index_info.timestamp_end);
                }
                ret = 0;
@@ -1820,9 +1822,13 @@ int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
                        goto end_rcu_unlock;
                }
                index_created = 1;
+               stream->indexes_in_flight++;
        }
 
        copy_index_control_data(index, &index_info);
+       if (stream->ctf_stream_id == -1ULL) {
+               stream->ctf_stream_id = be64toh(index_info.stream_id);
+       }
 
        if (index_created) {
                /*
@@ -1847,6 +1853,8 @@ int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
                        goto end_rcu_unlock;
                }
                stream->total_index_received++;
+               stream->indexes_in_flight--;
+               assert(stream->indexes_in_flight >= 0);
        }
 
 end_rcu_unlock:
@@ -2010,6 +2018,7 @@ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
                        goto error;
                }
                index_created = 1;
+               stream->indexes_in_flight++;
        }
 
        if (rotate_index || stream->index_fd < 0) {
@@ -2052,6 +2061,8 @@ static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
                        goto error;
                }
                stream->total_index_received++;
+               stream->indexes_in_flight--;
+               assert(stream->indexes_in_flight >= 0);
        }
 
 error:
@@ -2253,7 +2264,7 @@ static void destroy_connection(struct lttng_ht *relay_connections_ht,
        connection_delete(relay_connections_ht, conn);
 
        /* For the control socket, we try to destroy the session. */
-       if (conn->type == RELAY_CONTROL) {
+       if (conn->type == RELAY_CONTROL && conn->session) {
                destroy_session(conn->session, conn->sessions_ht);
        }
 
@@ -2275,6 +2286,7 @@ void *relay_thread_worker(void *data)
        struct lttcomm_relayd_hdr recv_hdr;
        struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
        struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
+       struct relay_index *index;
 
        DBG("[thread] Relay worker started");
 
@@ -2463,6 +2475,7 @@ restart:
 
                                if (revents & LPOLLIN) {
                                        if (conn->type != RELAY_DATA) {
+                                               rcu_read_unlock();
                                                continue;
                                        }
 
@@ -2506,6 +2519,14 @@ error:
        }
        rcu_read_unlock();
 error_poll_create:
+       rcu_read_lock();
+       cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
+                       index_n.node) {
+               health_code_update();
+               relay_index_delete(index);
+               relay_index_free_safe(index);
+       }
+       rcu_read_unlock();
        lttng_ht_destroy(indexes_ht);
 indexes_ht_error:
        lttng_ht_destroy(relay_connections_ht);