Fix: relayd: connection abruptly closed on viewer stream creation failure

[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 3eadbb1084da34ec07db25ceadae6a4fff4d759d..f1e440640aa465c9389f8061174e10657fa32a47 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);
@@ -1165,12 +1165,27 @@ int viewer_get_new_streams(struct relay_connection *conn)
         * the viewer's point of view.
         */
        pthread_mutex_lock(&session->lock);
+       /*
+        * If a session rotation is ongoing, do not attempt to open any
+        * stream, because the chunk can be in an intermediate state
+        * due to directory renaming.
+        */
+       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;
+       }
        ret = make_viewer_streams(session,
                        conn->viewer_session,
                        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;
        }
        send_streams = 1;
        response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
@@ -1225,10 +1240,6 @@ end_put_session:
        }
 error:
        return ret;
-error_unlock_session:
-       pthread_mutex_unlock(&session->lock);
-       session_put(session);
-       return ret;
 }
 
 /*
@@ -1305,6 +1316,17 @@ int viewer_attach_session(struct relay_connection *conn)
                goto send_reply;
        }
 
+       /*
+        * If a session rotation is ongoing, do not attempt to open any
+        * stream, because the chunk can be in an intermediate state
+        * due to directory renaming.
+        */
+       if (session_has_ongoing_rotation(session)) {
+               DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
+               send_streams = 0;
+               goto send_reply;
+       }
+
        ret = make_viewer_streams(session,
                        conn->viewer_session, seek_type,
                        &nb_streams, NULL, NULL, &closed);
@@ -1609,6 +1631,13 @@ int viewer_get_next_index(struct relay_connection *conn)
        metadata_viewer_stream =
                        ctf_trace_get_viewer_metadata_stream(ctf_trace);
 
+       /*
+        * Hold the session lock to protect against concurrent changes
+        * to the chunk files (e.g. rename done by clear), which are
+        * protected by the session ongoing rotation state. Those are
+        * synchronized with the session lock.
+        */
+       pthread_mutex_lock(&rstream->trace->session->lock);
        pthread_mutex_lock(&rstream->lock);
 
        /*
@@ -1625,7 +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;
@@ -1775,6 +1804,7 @@ int viewer_get_next_index(struct relay_connection *conn)
 send_reply:
        if (rstream) {
                pthread_mutex_unlock(&rstream->lock);
+               pthread_mutex_unlock(&rstream->trace->session->lock);
        }
 
        if (metadata_viewer_stream) {
@@ -1816,6 +1846,7 @@ end:
 
 error_put:
        pthread_mutex_unlock(&rstream->lock);
+       pthread_mutex_unlock(&rstream->trace->session->lock);
        if (metadata_viewer_stream) {
                viewer_stream_put(metadata_viewer_stream);
        }