Fix race conditions in overlay network driver

The overlay network driver is not properly using it's mutexes or
sync.Onces.  It made the classic mistake of not holding a lock through
various read-modify-write operations.  This can result in inconsistent
state storage leading to more catastrophic issues.

This patch attempts to maintain the previous semantics while holding the
driver lock through operations that are read-modify-write of the
driver's network state.

One example of this race would be if two goroutines tried to invoke
d.network() after the network ID was removed from the table.  Both would
try to reinstall it causing the "once" to get reinitialized twice
without any lock protection.  This could then lead to the "once" getting
invoked twice on the same network.  Furthermore, the changes to one of
these network structures gets effectively discarded.  It's also the
case, that because there would be two simultaneous instances of the
network, the various network Lock() invocations would be meaningless for
race prevention.

Signed-off-by: Chris Telfer <ctelfer@docker.com>

libnetwork/drivers/overlay/ov_network.go

@@ -203,6 +203,12 @@ func (d *driver) CreateNetwork(id string, option map[string]interface{}, nInfo d
 		n.subnets = append(n.subnets, s)
 	}
 
+	d.Lock()
+	defer d.Unlock()
+	if d.networks[n.id] != nil {
+		return fmt.Errorf("attempt to create overlay network %v that already exists", n.id)
+	}
+
 	if err := n.writeToStore(); err != nil {
 		return fmt.Errorf("failed to update data store for network %v: %v", n.id, err)
 	}
@@ -217,11 +223,13 @@ func (d *driver) CreateNetwork(id string, option map[string]interface{}, nInfo d
 
 	if nInfo != nil {
 		if err := nInfo.TableEventRegister(ovPeerTable, driverapi.EndpointObject); err != nil {
+			// XXX Undo writeToStore?  No method to so.  Why?
 			return err
 		}
 	}
 
-	d.addNetwork(n)
+	d.networks[id] = n
+
 	return nil
 }
 
@@ -235,7 +243,15 @@ func (d *driver) DeleteNetwork(nid string) error {
 		return err
 	}
 
-	n := d.network(nid)
+	d.Lock()
+	defer d.Unlock()
+
+	// This is similar to d.network(), but we need to keep holding the lock
+	// until we are done removing this network.
+	n, ok := d.networks[nid]
+	if !ok {
+		n = d.restoreNetworkFromStore(nid)
+	}
 	if n == nil {
 		return fmt.Errorf("could not find network with id %s", nid)
 	}
@@ -255,7 +271,7 @@ func (d *driver) DeleteNetwork(nid string) error {
 	}
 	// flush the peerDB entries
 	d.peerFlush(nid)
-	d.deleteNetwork(nid)
+	delete(d.networks, nid)
 
 	vnis, err := n.releaseVxlanID()
 	if err != nil {
@@ -805,32 +821,25 @@ func (n *network) watchMiss(nlSock *nl.NetlinkSocket, nsPath string) {
 	}
 }
 
-func (d *driver) addNetwork(n *network) {
-	d.Lock()
-	d.networks[n.id] = n
-	d.Unlock()
-}
-
-func (d *driver) deleteNetwork(nid string) {
-	d.Lock()
-	delete(d.networks, nid)
-	d.Unlock()
+// Restore a network from the store to the driver if it is present.
+// Must be called with the driver locked!
+func (d *driver) restoreNetworkFromStore(nid string) *network {
+	n := d.getNetworkFromStore(nid)
+	if n != nil {
+		n.driver = d
+		n.endpoints = endpointTable{}
+		n.once = &sync.Once{}
+		d.networks[nid] = n
+	}
+	return n
 }
 
 func (d *driver) network(nid string) *network {
 	d.Lock()
+	defer d.Unlock()
 	n, ok := d.networks[nid]
-	d.Unlock()
 	if !ok {
-		n = d.getNetworkFromStore(nid)
-		if n != nil {
-			n.driver = d
-			n.endpoints = endpointTable{}
-			n.once = &sync.Once{}
-			d.Lock()
-			d.networks[nid] = n
-			d.Unlock()
-		}
+		n = d.restoreNetworkFromStore(nid)
 	}
 
 	return n