fixed data race in core.Connection.Close().

The problem is triggered by the following events:
1. VPP stops.
2. core.Connection.healthCheckLoop() detects it and calls disconnectVPP(),
   which does close healthCheckDone channel for the first time.
   At this point Connection becomes unusable - re-entrance to
   connectLoop() will return immediately because of a closed
   healthCheckDone channel.
   But, the race is that connection may set c.vppConnected = 1
   before terminating the connect loop.
3. User calls Connection.Close(), who seed c.vppConnected = 1,
   and tries to close already closed channel, which leads to a panic.

This commit fixes this race by telling disconnectVPP() whether
it is a terminal stop triggered via Close(), or a temporary one
from connectLoop().

Change-Id: I555149da35ca3dc2b606bad59f2101266c0ef6b9
Signed-off-by: Sergey Elantsev <elantsev.s@yandex.ru>

1 files changed, 9 insertions, 5 deletions

diff --git a/core/connection.go b/core/connection.go
index 1bfcae5..f796f3d 100644
--- a/core/connection.go
+++ b/core/connection.go
@@ -216,14 +216,18 @@ func (c *Connection) Disconnect() {
 		return
 	}
 	if c.vppClient != nil {
-		c.disconnectVPP()
+		c.disconnectVPP(true)
 	}
 }
 
-// disconnectVPP disconnects from VPP in case it is connected.
-func (c *Connection) disconnectVPP() {
+// disconnectVPP disconnects from VPP in case it is connected. terminate tells
+// that disconnectVPP() was called from Close(), so healthCheckLoop() can be
+// terminated.
+func (c *Connection) disconnectVPP(terminate bool) {
 	if atomic.CompareAndSwapUint32(&c.vppConnected, 1, 0) {
-		close(c.healthCheckDone)
+		if terminate {
+			close(c.healthCheckDone)
+		}
 		log.Debug("Disconnecting from VPP..")
 
 		if err := c.vppClient.Disconnect(); err != nil {
@@ -383,7 +387,7 @@ HealthCheck:
 	}
 
 	// cleanup
-	c.disconnectVPP()
+	c.disconnectVPP(false)
 
 	// we are now disconnected, start connect loop
 	c.connectLoop()