Fix: relayd: live: dispose of zombie viewer metadata stream

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

diff --git a/src/bin/lttng-relayd/live.c b/src/bin/lttng-relayd/live.c
index 604d24aa6ac00f98cd6ee8e4b9bff4088b9b95f3..0355cf9d91f06cf95c62a84efd61ccb07fa29f60 100644 (file)
--- a/src/bin/lttng-relayd/live.c
+++ b/src/bin/lttng-relayd/live.c
@@ -374,7 +374,7 @@ static int make_viewer_streams(struct relay_session *relay_session,
                                 * chunk can be used safely.
                                 */
                                if ((relay_stream->ongoing_rotation.is_set ||
-                                                   relay_session->ongoing_rotation) &&
+                                               session_has_ongoing_rotation(relay_session)) &&
                                                relay_stream->trace_chunk) {
                                        viewer_stream_trace_chunk = lttng_trace_chunk_copy(
                                                        relay_stream->trace_chunk);
@@ -1170,7 +1170,7 @@ int viewer_get_new_streams(struct relay_connection *conn)
         * stream, because the chunk can be in an intermediate state
         * due to directory renaming.
         */
-       if (session->ongoing_rotation) {
+       if (session_has_ongoing_rotation(session)) {
                DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
                response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_NO_NEW);
                goto send_reply_unlock;
@@ -1180,7 +1180,12 @@ int viewer_get_new_streams(struct relay_connection *conn)
                        LTTNG_VIEWER_SEEK_BEGINNING, &nb_total, &nb_unsent,
                        &nb_created, &closed);
        if (ret < 0) {
-               goto error_unlock_session;
+               /*
+                * This is caused by an internal error; propagate the negative
+                * 'ret' to close the connection.
+                */
+               response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
+               goto send_reply_unlock;
        }
        send_streams = 1;
        response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
@@ -1235,10 +1240,6 @@ end_put_session:
        }
 error:
        return ret;
-error_unlock_session:
-       pthread_mutex_unlock(&session->lock);
-       session_put(session);
-       return ret;
 }
 
 /*
@@ -1320,7 +1321,7 @@ int viewer_attach_session(struct relay_connection *conn)
         * stream, because the chunk can be in an intermediate state
         * due to directory renaming.
         */
-       if (session->ongoing_rotation) {
+       if (session_has_ongoing_rotation(session)) {
                DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
                send_streams = 0;
                goto send_reply;
@@ -1653,7 +1654,7 @@ int viewer_get_next_index(struct relay_connection *conn)
                goto send_reply;
        }
 
-       if (rstream->trace->session->ongoing_rotation) {
+       if (session_has_ongoing_rotation(rstream->trace->session)) {
                /* Rotation is ongoing, try again later. */
                viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
                goto send_reply;
@@ -1983,6 +1984,7 @@ int viewer_get_metadata(struct relay_connection *conn)
        struct lttng_viewer_get_metadata request;
        struct lttng_viewer_metadata_packet reply;
        struct relay_viewer_stream *vstream = NULL;
+       bool dispose_of_stream = false;
 
        assert(conn);
 
@@ -2013,6 +2015,9 @@ int viewer_get_metadata(struct relay_connection *conn)
                reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
                goto send_reply;
        }
+
+       pthread_mutex_lock(&vstream->stream->trace->session->lock);
+       pthread_mutex_lock(&vstream->stream->trace->lock);
        pthread_mutex_lock(&vstream->stream->lock);
        if (!vstream->stream->is_metadata) {
                ERR("Invalid metadata stream");
@@ -2021,11 +2026,7 @@ int viewer_get_metadata(struct relay_connection *conn)
 
        if (vstream->metadata_sent >= vstream->stream->metadata_received) {
                /*
-                * The live viewers expect to receive a NO_NEW_METADATA
-                * status before a stream disappears, otherwise they abort the
-                * entire live connection when receiving an error status.
-                *
-                * Clear feature resets the metadata_sent to 0 until the
+                * Clear feature resets the metadata_received to 0 until the
                 * same metadata is received again.
                 */
                reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
@@ -2033,13 +2034,7 @@ int viewer_get_metadata(struct relay_connection *conn)
                 * The live viewer considers a closed 0 byte metadata stream as
                 * an error.
                 */
-               if (vstream->metadata_sent > 0) {
-                       if (vstream->stream->closed && vstream->stream->no_new_metadata_notified) {
-                               /* Release ownership for the viewer metadata stream. */
-                               viewer_stream_put(vstream);
-                       }
-                       vstream->stream->no_new_metadata_notified = true;
-               }
+               dispose_of_stream = vstream->metadata_sent > 0 && vstream->stream->closed;
                goto send_reply;
        }
 
@@ -2079,6 +2074,19 @@ int viewer_get_metadata(struct relay_connection *conn)
        len = vstream->stream->metadata_received - vstream->metadata_sent;
 
        if (!vstream->stream_file.trace_chunk) {
+               if (vstream->stream->trace->session->connection_closed) {
+                       /*
+                        * If the connection is closed, there is no way for the metadata stream
+                        * to ever transition back to an active chunk. As such, signal to the viewer
+                        * that there is no new metadata available.
+                        *
+                        * The stream can be disposed-of. On the next execution of this command,
+                        * the relay daemon will reply with an error status since the stream can't
+                        * be found.
+                        */
+                       dispose_of_stream = true;
+               }
+
                reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
                len = 0;
                goto send_reply;
@@ -2208,6 +2216,8 @@ send_reply:
        health_code_update();
        if (vstream) {
                pthread_mutex_unlock(&vstream->stream->lock);
+               pthread_mutex_unlock(&vstream->stream->trace->lock);
+               pthread_mutex_unlock(&vstream->stream->trace->session->lock);
        }
        ret = send_response(conn->sock, &reply, sizeof(reply));
        if (ret < 0) {
@@ -2232,7 +2242,22 @@ end_free:
 end:
        if (vstream) {
                viewer_stream_put(vstream);
+               if (dispose_of_stream) {
+                       /*
+                        * Trigger the destruction of the viewer stream
+                        * by releasing its global reference.
+                        *
+                        * The live viewers expect to receive a NO_NEW_METADATA
+                        * status before a stream disappears, otherwise they abort the
+                        * entire live connection when receiving an error status.
+                        *
+                        * On the next query for this stream, an error will be reported to the
+                        * client.
+                        */
+                       viewer_stream_put(vstream);
+               }
        }
+
        return ret;
 }