Fix issue where the IPC server can fully consume a CPU core and prevent incoming connections (dotnet#102413)

Author: davmason

src/native/eventpipe/ds-ipc-pal-namedpipe.c

@@ -173,6 +173,25 @@ ds_ipc_free (DiagnosticsIpc *ipc)
 	ep_rt_object_free (ipc);
 }
 
+void
+ds_ipc_reset (DiagnosticsIpc *ipc)
+{
+	if (!ipc)
+		return;
+
+	if (ipc->pipe != INVALID_HANDLE_VALUE) {
+		CloseHandle (ipc->pipe);
+		ipc->pipe = INVALID_HANDLE_VALUE;
+	}
+
+	if (ipc->overlap.hEvent != INVALID_HANDLE_VALUE) {
+		CloseHandle (ipc->overlap.hEvent);
+		ipc->overlap.hEvent = INVALID_HANDLE_VALUE;
+	}
+
+	ipc->is_listening = false;
+}
+
 int32_t
 ds_ipc_poll (
 	DiagnosticsIpcPollHandle *poll_handles_data,
@@ -192,6 +211,10 @@ ds_ipc_poll (
 			// SERVER
 			EP_ASSERT (poll_handles_data [i].ipc->mode == DS_IPC_CONNECTION_MODE_LISTEN);
 			handles [i] = poll_handles_data [i].ipc->overlap.hEvent;
+			if (handles [i] == INVALID_HANDLE_VALUE) {
+				// Invalid handle, wait will fail. Signal error
+				poll_handles_data [i].events = DS_IPC_POLL_EVENTS_ERR;
+			}
 		} else {
 			// CLIENT
 			bool success = true;

src/native/eventpipe/ds-ipc-pal-socket.c

@@ -1064,6 +1064,11 @@ ds_ipc_free (DiagnosticsIpc *ipc)
 	ep_rt_object_free (ipc);
 }
 
+void
+ds_ipc_reset (DiagnosticsIpc *ipc)
+{
+}
+
 int32_t
 ds_ipc_poll (
 	DiagnosticsIpcPollHandle *poll_handles_data,

src/native/eventpipe/ds-ipc-pal.h

@@ -35,6 +35,9 @@ ds_ipc_alloc (
 void
 ds_ipc_free (DiagnosticsIpc *ipc);
 
+void
+ds_ipc_reset (DiagnosticsIpc *ipc);
+
 // Poll
 // Parameters:
 // - IpcPollHandle * poll_handles_data: Array of IpcPollHandles to poll

src/native/eventpipe/ds-ipc.c

@@ -431,6 +431,7 @@ ds_ipc_stream_factory_get_next_available_stream (ds_ipc_error_callback_func call
 					DS_LOG_INFO_2 ("ds_ipc_stream_factory_get_next_available_stream - NON :: Poll attempt: %d, connection %d had no events.", poll_attempts, connection_id);
 					break;
 				default:
+					ds_port_reset_vcall ((DiagnosticsPort *)ipc_poll_handle.user_data, callback);
 					DS_LOG_INFO_2 ("ds_ipc_stream_factory_get_next_available_stream - UNK :: Poll attempt: %d, connection %d had invalid PollEvent.", poll_attempts, connection_id);
 					saw_error = true;
 					break;
@@ -441,6 +442,12 @@ ds_ipc_stream_factory_get_next_available_stream (ds_ipc_error_callback_func call
 		}
 
 		if (!stream && saw_error) {
+			// Some errors can cause the poll to return instantly, we want to delay if we see an error to avoid
+			// runaway CPU usage.
+			if (poll_timeout_ms == DS_IPC_TIMEOUT_INFINITE)
+				poll_timeout_ms = DS_IPC_POLL_TIMEOUT_MAX_MS;
+			DS_LOG_DEBUG_1 ("ds_ipc_stream_factory_get_next_available_stream - Saw error, sleeping using timeout: %dms.", poll_timeout_ms);
+			ep_rt_thread_sleep ((uint64_t)poll_timeout_ms * NUM_NANOSECONDS_IN_1_MS);
 			_ds_current_port = NULL;
 			ep_raise_error ();
 		}
@@ -839,7 +846,9 @@ listen_port_reset (
 	ds_ipc_error_callback_func callback)
 {
 	EP_ASSERT (object != NULL);
-	return;
+	DiagnosticsListenPort *listen_port = (DiagnosticsListenPort *)object;
+	ds_ipc_reset (listen_port->port.ipc);
+	ds_ipc_listen (listen_port->port.ipc, callback);
 }
 
 static DiagnosticsPortVtable listen_port_vtable = {