Fix: make sure no index is in flight before using inactivity beacons

[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 f6b23eb928a9137ae5eb939dc8c9efa4a892c81b..fa5a7db86c393640c15fe9e8b3bf5f652da98111 100644 (file)
--- a/src/bin/lttng-relayd/main.c
+++ b/src/bin/lttng-relayd/main.c
@@ -44,6 +44,7 @@
 #include <common/common.h>
 #include <common/compat/poll.h>
 #include <common/compat/socket.h>
+#include <common/compat/endian.h>
 #include <common/defaults.h>
 #include <common/daemonize.h>
 #include <common/futex.h>
@@ -250,7 +251,10 @@ int set_option(int opt, const char *arg, const char *optname)
                if (arg) {
                        lttng_opt_verbose = config_parse_value(arg);
                } else {
-                       lttng_opt_verbose += 1;
+                       /* Only 3 level of verbosity (-vvv). */
+                       if (lttng_opt_verbose < 3) {
+                               lttng_opt_verbose += 1;
+                       }
                }
                break;
        default:
@@ -382,8 +386,9 @@ int set_options(int argc, char **argv)
 
        /* assign default values */
        if (control_uri == NULL) {
-               ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
-                               DEFAULT_NETWORK_CONTROL_PORT);
+               ret = asprintf(&default_address,
+                       "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
+                       DEFAULT_NETWORK_CONTROL_PORT);
                if (ret < 0) {
                        PERROR("asprintf default data address");
                        goto exit;
@@ -397,8 +402,9 @@ int set_options(int argc, char **argv)
                }
        }
        if (data_uri == NULL) {
-               ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
-                               DEFAULT_NETWORK_DATA_PORT);
+               ret = asprintf(&default_address,
+                       "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
+                       DEFAULT_NETWORK_DATA_PORT);
                if (ret < 0) {
                        PERROR("asprintf default data address");
                        goto exit;
@@ -412,8 +418,9 @@ int set_options(int argc, char **argv)
                }
        }
        if (live_uri == NULL) {
-               ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
-                               DEFAULT_NETWORK_VIEWER_PORT);
+               ret = asprintf(&default_address,
+                       "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
+                       DEFAULT_NETWORK_VIEWER_PORT);
                if (ret < 0) {
                        PERROR("asprintf default viewer control address");
                        goto exit;
@@ -875,11 +882,12 @@ restart:
                                new_conn->sock = newsock;
 
                                /* Enqueue request for the dispatcher thread. */
-                               cds_wfq_enqueue(&relay_conn_queue.queue, &new_conn->qnode);
+                               cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
+                                                &new_conn->qnode);
 
                                /*
                                 * Wake the dispatch queue futex. Implicit memory barrier with
-                                * the exchange in cds_wfq_enqueue.
+                                * the exchange in cds_wfcq_enqueue.
                                 */
                                futex_nto1_wake(&relay_conn_queue.futex);
                        }
@@ -926,7 +934,7 @@ void *relay_thread_dispatcher(void *data)
 {
        int err = -1;
        ssize_t ret;
-       struct cds_wfq_node *node;
+       struct cds_wfcq_node *node;
        struct relay_connection *new_conn = NULL;
 
        DBG("[thread] Relay dispatcher started");
@@ -949,7 +957,8 @@ void *relay_thread_dispatcher(void *data)
                        health_code_update();
 
                        /* Dequeue commands */
-                       node = cds_wfq_dequeue_blocking(&relay_conn_queue.queue);
+                       node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
+                                                        &relay_conn_queue.tail);
                        if (node == NULL) {
                                DBG("Woken up but nothing in the relay command queue");
                                /* Continue thread execution */
@@ -1198,6 +1207,7 @@ 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);
 
@@ -1261,6 +1271,7 @@ int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
                        stream->stream_handle);
 
 end:
+       memset(&reply, 0, sizeof(reply));
        reply.handle = htobe64(stream->stream_handle);
        /* send the session id to the client or a negative return code on error */
        if (ret < 0) {
@@ -1342,6 +1353,7 @@ int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
 end_unlock:
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_UNK);
        } else {
@@ -1367,6 +1379,7 @@ void relay_unknown_command(struct relay_connection *conn)
        struct lttcomm_relayd_generic_reply reply;
        int ret;
 
+       memset(&reply, 0, sizeof(reply));
        reply.ret_code = htobe32(LTTNG_ERR_UNK);
        ret = conn->sock->ops->sendmsg(conn->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
@@ -1392,6 +1405,7 @@ int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
                ret = htobe32(LTTNG_ERR_UNK);
        }
 
+       memset(&reply, 0, sizeof(reply));
        reply.ret_code = ret;
        ret = conn->sock->ops->sendmsg(conn->sock, &reply,
                        sizeof(struct lttcomm_relayd_generic_reply), 0);
@@ -1553,6 +1567,7 @@ int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
                goto end;
        }
 
+       memset(&reply, 0, sizeof(reply));
        reply.major = RELAYD_VERSION_COMM_MAJOR;
        reply.minor = RELAYD_VERSION_COMM_MINOR;
 
@@ -1652,6 +1667,7 @@ int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
 end_unlock:
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        reply.ret_code = htobe32(ret);
        ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
@@ -1715,6 +1731,7 @@ int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
        }
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        reply.ret_code = htobe32(LTTNG_OK);
        ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
        if (ret < 0) {
@@ -1785,6 +1802,7 @@ int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
        }
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        /* All good, send back reply. */
        reply.ret_code = htobe32(LTTNG_OK);
 
@@ -1858,6 +1876,7 @@ int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
        }
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        /* All good, send back reply. */
        reply.ret_code = htobe32(is_data_inflight);
 
@@ -1925,9 +1944,10 @@ 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;
@@ -1944,9 +1964,13 @@ 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) {
                /*
@@ -1971,11 +1995,14 @@ 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:
        rcu_read_unlock();
 
+       memset(&reply, 0, sizeof(reply));
        if (ret < 0) {
                reply.ret_code = htobe32(LTTNG_ERR_UNK);
        } else {
@@ -2026,6 +2053,7 @@ int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
                uatomic_set(&conn->session->new_streams, 1);
        }
 
+       memset(&reply, 0, sizeof(reply));
        reply.ret_code = htobe32(LTTNG_OK);
        send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
        if (send_ret < 0) {
@@ -2132,6 +2160,7 @@ 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) {
@@ -2174,6 +2203,8 @@ 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:
@@ -2375,7 +2406,7 @@ 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);
        }
 
@@ -2740,7 +2771,7 @@ int main(int argc, char **argv)
        }
 
        /* Init relay command queue. */
-       cds_wfq_init(&relay_conn_queue.queue);
+       cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
 
        /* Set up max poll set size */
        lttng_poll_set_max_size();