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

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

diff --git a/src/bin/lttng-relayd/live.cpp b/src/bin/lttng-relayd/live.cpp
index 57b55587d2fb281cc46a4f9c13e08c9dadff9c1d..e0d446b2a1a4ee25cdc25539c870a87f52a1ef91 100644 (file)
--- a/src/bin/lttng-relayd/live.cpp
+++ b/src/bin/lttng-relayd/live.cpp
@@ -160,6 +160,24 @@ const char *lttng_viewer_attach_return_code_str(
        }
 };
 
+static
+const char *lttng_viewer_get_packet_return_code_str(
+               enum lttng_viewer_get_packet_return_code code)
+{
+       switch (code) {
+       case LTTNG_VIEWER_GET_PACKET_OK:
+               return "GET_PACKET_OK";
+       case LTTNG_VIEWER_GET_PACKET_RETRY:
+               return "GET_PACKET_RETRY";
+       case LTTNG_VIEWER_GET_PACKET_ERR:
+               return "GET_PACKET_ERR";
+       case LTTNG_VIEWER_GET_PACKET_EOF:
+               return "GET_PACKET_EOF";
+       default:
+               abort();
+       }
+};
+
 /*
  * Cleanup the daemon
  */
@@ -446,7 +464,7 @@ 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);
@@ -1238,7 +1256,7 @@ 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;
@@ -1248,7 +1266,12 @@ 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;
        }
        send_streams = 1;
        response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
@@ -1303,10 +1326,6 @@ end_put_session:
        }
 error:
        return ret;
-error_unlock_session:
-       pthread_mutex_unlock(&session->lock);
-       session_put(session);
-       return ret;
 }
 
 /*
@@ -1398,7 +1417,7 @@ 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;
@@ -1770,7 +1789,7 @@ 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 = LTTNG_VIEWER_INDEX_RETRY;
                DBG("Client requested index for stream id %" PRIu64" while a session rotation is ongoing, returning status=%s",
@@ -2069,6 +2088,7 @@ int viewer_get_packet(struct relay_connection *conn)
        uint32_t packet_data_len = 0;
        ssize_t read_len;
        uint64_t stream_id;
+       enum lttng_viewer_get_packet_return_code get_packet_status;
 
        health_code_update();
 
@@ -2085,9 +2105,10 @@ int viewer_get_packet(struct relay_connection *conn)
 
        vstream = viewer_stream_get_by_id(stream_id);
        if (!vstream) {
-               DBG("Client requested packet of unknown stream id %" PRIu64,
-                               stream_id);
-               reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
+               get_packet_status = LTTNG_VIEWER_GET_PACKET_ERR;
+               DBG("Client requested packet of unknown stream id %" PRIu64
+                       ", returning status=%s", stream_id,
+                       lttng_viewer_get_packet_return_code_str(get_packet_status));
                goto send_reply_nolock;
        } else {
                packet_data_len = be32toh(get_packet_info.len);
@@ -2096,8 +2117,9 @@ int viewer_get_packet(struct relay_connection *conn)
 
        reply = (char *) zmalloc(reply_size);
        if (!reply) {
-               PERROR("packet reply zmalloc");
-               reply_size = sizeof(reply_header);
+               get_packet_status = LTTNG_VIEWER_GET_PACKET_ERR;
+               PERROR("Falled to allocate reply, returning status=%s",
+                       lttng_viewer_get_packet_return_code_str(get_packet_status));
                goto error;
        }
 
@@ -2105,29 +2127,31 @@ int viewer_get_packet(struct relay_connection *conn)
        lseek_ret = fs_handle_seek(vstream->stream_file.handle,
                        be64toh(get_packet_info.offset), SEEK_SET);
        if (lseek_ret < 0) {
+               get_packet_status = LTTNG_VIEWER_GET_PACKET_ERR;
                PERROR("Failed to seek file system handle of viewer stream %" PRIu64
-                      " to offset %" PRIu64,
-                               stream_id,
-                               (uint64_t) be64toh(get_packet_info.offset));
+                      " to offset %" PRIu64", returning status=%s", stream_id,
+                       (uint64_t) be64toh(get_packet_info.offset),
+                       lttng_viewer_get_packet_return_code_str(get_packet_status));
                goto error;
        }
        read_len = fs_handle_read(vstream->stream_file.handle,
                        reply + sizeof(reply_header), packet_data_len);
        if (read_len < packet_data_len) {
+               get_packet_status = LTTNG_VIEWER_GET_PACKET_ERR;
                PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
-                      ", offset: %" PRIu64,
-                               stream_id,
-                               (uint64_t) be64toh(get_packet_info.offset));
+                      ", offset: %" PRIu64 ", returning status=%s", stream_id,
+                      (uint64_t) be64toh(get_packet_info.offset),
+                       lttng_viewer_get_packet_return_code_str(get_packet_status));
                goto error;
        }
-       reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
+
+       get_packet_status = LTTNG_VIEWER_GET_PACKET_OK;
        reply_header.len = htobe32(packet_data_len);
        goto send_reply;
 
 error:
        /* No payload to send on error. */
        reply_size = sizeof(reply_header);
-       reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
 
 send_reply:
        if (vstream) {
@@ -2137,6 +2161,7 @@ send_reply_nolock:
 
        health_code_update();
 
+       reply_header.status = htobe32(get_packet_status);
        if (reply) {
                memcpy(reply, &reply_header, sizeof(reply_header));
                ret = send_response(conn->sock, reply, reply_size);