Support dockerd and system restarts for ipvlan and macvlan networks

This commit carries forward the work done in
https://github.com/docker/libnetwork/pull/2295
and fixes two things
1. Allows macvlan and ipvlan to be restored properly
after dockerd or the system is restarted
2. Makes sure the refcount for the configOnly network
is not incremented for the above case so this network
can be deleted after all the associated ConfigFrom networks
are deleted

Addresses: https://github.com/docker/libnetwork/issues/1743

Signed-off-by: Arko Dasgupta <arko.dasgupta@docker.com>

libnetwork/controller.go

@@ -706,9 +706,10 @@ const overlayDSROptionString = "dsr"
 // are network specific and modeled in a generic way.
 func (c *controller) NewNetwork(networkType, name string, id string, options ...NetworkOption) (Network, error) {
 	var (
-		cap *driverapi.Capability
-		err error
-		t   *network
+		cap            *driverapi.Capability
+		err            error
+		t              *network
+		skipCfgEpCount bool
 	)
 
 	if id != "" {
@@ -802,7 +803,7 @@ func (c *controller) NewNetwork(networkType, name string, id string, options ...
 			return nil, types.InternalErrorf("Failed to apply configuration: %v", err)
 		}
 		defer func() {
-			if err == nil {
+			if err == nil && !skipCfgEpCount {
 				if err := t.getEpCnt().IncEndpointCnt(); err != nil {
 					logrus.Warnf("Failed to update reference count for configuration network %q on creation of network %q: %v",
 						t.Name(), network.Name(), err)
@@ -823,7 +824,13 @@ func (c *controller) NewNetwork(networkType, name string, id string, options ...
 
 	err = c.addNetwork(network)
 	if err != nil {
-		return nil, err
+		if strings.Contains(err.Error(), "restoring existing network") {
+			// This error can be ignored and set this boolean
+			// value to skip a refcount increment for configOnly networks
+			skipCfgEpCount = true
+		} else {
+			return nil, err
+		}
 	}
 	defer func() {
 		if err != nil {

libnetwork/drivers/ipvlan/ipvlan_network.go

@@ -60,10 +60,14 @@ func (d *driver) CreateNetwork(nid string, option map[string]interface{}, nInfo
 		// empty parent and --internal are handled the same. Set here to update k/v
 		config.Internal = true
 	}
-	err = d.createNetwork(config)
+	foundExisting, err := d.createNetwork(config)
 	if err != nil {
 		return err
 	}
+
+	if foundExisting {
+		return types.InternalMaskableErrorf("restoring existing network %s", config.ID)
+	}
 	// update persistent db, rollback on fail
 	err = d.storeUpdate(config)
 	if err != nil {
@@ -76,22 +80,34 @@ func (d *driver) CreateNetwork(nid string, option map[string]interface{}, nInfo
 }
 
 // createNetwork is used by new network callbacks and persistent network cache
-func (d *driver) createNetwork(config *configuration) error {
+func (d *driver) createNetwork(config *configuration) (bool, error) {
+	foundExisting := false
 	networkList := d.getNetworks()
 	for _, nw := range networkList {
 		if config.Parent == nw.config.Parent {
-			return fmt.Errorf("network %s is already using parent interface %s",
-				getDummyName(stringid.TruncateID(nw.config.ID)), config.Parent)
+			if config.ID != nw.config.ID {
+				return false, fmt.Errorf("network %s is already using parent interface %s",
+					getDummyName(stringid.TruncateID(nw.config.ID)), config.Parent)
+			}
+			logrus.Debugf("Create Network for the same ID %s\n", config.ID)
+			foundExisting = true
+			break
 		}
 	}
 	if !parentExists(config.Parent) {
 		// if the --internal flag is set, create a dummy link
 		if config.Internal {
-			err := createDummyLink(config.Parent, getDummyName(stringid.TruncateID(config.ID)))
-			if err != nil {
-				return err
+			if !dummyLinkExists(getDummyName(stringid.TruncateID(config.ID))) {
+				err := createDummyLink(config.Parent, getDummyName(stringid.TruncateID(config.ID)))
+				if err != nil {
+					return false, err
+				}
+				config.CreatedSlaveLink = true
+
+			} else {
+				logrus.Debugf("Dummy Link %s for ipvlan already exists", getDummyName(stringid.TruncateID(config.ID)))
 			}
-			config.CreatedSlaveLink = true
+
 			// notify the user in logs they have limited communications
 			if config.Parent == getDummyName(stringid.TruncateID(config.ID)) {
 				logrus.Debugf("Empty -o parent= and --internal flags limit communications to other containers inside of network: %s",
@@ -100,24 +116,31 @@ func (d *driver) createNetwork(config *configuration) error {
 		} else {
 			// if the subinterface parent_iface.vlan_id checks do not pass, return err.
 			//  a valid example is 'eth0.10' for a parent iface 'eth0' with a vlan id '10'
-			err := createVlanLink(config.Parent)
-			if err != nil {
-				return err
+			if !vlanLinkExists(config.Parent) {
+				err := createVlanLink(config.Parent)
+				if err != nil {
+					return false, err
+				}
+				// if driver created the networks slave link, record it for future deletion
+				config.CreatedSlaveLink = true
+			} else {
+				logrus.Debugf("Parent Sub Interface %s already Exists NetID %s", config.Parent, config.ID)
 			}
-			// if driver created the networks slave link, record it for future deletion
-			config.CreatedSlaveLink = true
+
 		}
 	}
-	n := &network{
-		id:        config.ID,
-		driver:    d,
-		endpoints: endpointTable{},
-		config:    config,
+	if !foundExisting {
+		n := &network{
+			id:        config.ID,
+			driver:    d,
+			endpoints: endpointTable{},
+			config:    config,
+		}
+		// add the network
+		d.addNetwork(n)
 	}
-	// add the *network
-	d.addNetwork(n)
 
-	return nil
+	return foundExisting, nil
 }
 
 // DeleteNetwork the network for the specified driver type

libnetwork/drivers/ipvlan/ipvlan_setup.go

@@ -70,6 +70,15 @@ func parentExists(ifaceStr string) bool {
 	return true
 }
 
+// vlanLinkExists checks if specified vlan link exists in the default namespace
+func vlanLinkExists(linkStr string) bool {
+	_, err := ns.NlHandle().LinkByName(linkStr)
+	if err != nil {
+		return false
+	}
+	return true
+}
+
 // createVlanLink parses sub-interfaces and vlan id for creation
 func createVlanLink(parentName string) error {
 	if strings.Contains(parentName, ".") {
@@ -156,6 +165,15 @@ func parseVlan(linkName string) (string, int, error) {
 	return parent, vidInt, nil
 }
 
+// dummyLinkExists checks if dummylink exists in the default namespace
+func dummyLinkExists(dummyName string) bool {
+	_, err := ns.NlHandle().LinkByName(dummyName)
+	if err != nil {
+		return false
+	}
+	return true
+}
+
 // createDummyLink creates a dummy0 parent link
 func createDummyLink(dummyName, truncNetID string) error {
 	// create a parent interface since one was not specified

libnetwork/drivers/ipvlan/ipvlan_store.go

@@ -55,7 +55,14 @@ func (d *driver) initStore(option map[string]interface{}) error {
 			return types.InternalErrorf("ipvlan driver failed to initialize data store: %v", err)
 		}
 
-		return d.populateNetworks()
+		err = d.populateNetworks()
+		if err != nil {
+			return err
+		}
+		err = d.populateEndpoints()
+		if err != nil {
+			return err
+		}
 	}
 
 	return nil
@@ -73,7 +80,7 @@ func (d *driver) populateNetworks() error {
 	}
 	for _, kvo := range kvol {
 		config := kvo.(*configuration)
-		if err = d.createNetwork(config); err != nil {
+		if _, err = d.createNetwork(config); err != nil {
 			logrus.Warnf("could not create ipvlan network for id %s from persistent state", config.ID)
 		}
 	}

libnetwork/drivers/macvlan/macvlan_network.go

@@ -64,10 +64,15 @@ func (d *driver) CreateNetwork(nid string, option map[string]interface{}, nInfo
 		// empty parent and --internal are handled the same. Set here to update k/v
 		config.Internal = true
 	}
-	err = d.createNetwork(config)
+	foundExisting, err := d.createNetwork(config)
 	if err != nil {
 		return err
 	}
+
+	if foundExisting {
+		return types.InternalMaskableErrorf("restoring existing network %s", config.ID)
+	}
+
 	// update persistent db, rollback on fail
 	err = d.storeUpdate(config)
 	if err != nil {
@@ -80,22 +85,32 @@ func (d *driver) CreateNetwork(nid string, option map[string]interface{}, nInfo
 }
 
 // createNetwork is used by new network callbacks and persistent network cache
-func (d *driver) createNetwork(config *configuration) error {
+func (d *driver) createNetwork(config *configuration) (bool, error) {
+	foundExisting := false
 	networkList := d.getNetworks()
 	for _, nw := range networkList {
 		if config.Parent == nw.config.Parent {
-			return fmt.Errorf("network %s is already using parent interface %s",
-				getDummyName(stringid.TruncateID(nw.config.ID)), config.Parent)
+			if config.ID != nw.config.ID {
+				return false, fmt.Errorf("network %s is already using parent interface %s",
+					getDummyName(stringid.TruncateID(nw.config.ID)), config.Parent)
+			}
+			logrus.Debugf("Create Network for the same ID %s\n", config.ID)
+			foundExisting = true
+			break
 		}
 	}
 	if !parentExists(config.Parent) {
 		// if the --internal flag is set, create a dummy link
 		if config.Internal {
-			err := createDummyLink(config.Parent, getDummyName(stringid.TruncateID(config.ID)))
-			if err != nil {
-				return err
+			if !dummyLinkExists(getDummyName(stringid.TruncateID(config.ID))) {
+				err := createDummyLink(config.Parent, getDummyName(stringid.TruncateID(config.ID)))
+				if err != nil {
+					return false, err
+				}
+				config.CreatedSlaveLink = true
+			} else {
+				logrus.Debugf("Dummy Link %s for Mac Vlan already exists", getDummyName(stringid.TruncateID(config.ID)))
 			}
-			config.CreatedSlaveLink = true
 			// notify the user in logs they have limited communications
 			if config.Parent == getDummyName(stringid.TruncateID(config.ID)) {
 				logrus.Debugf("Empty -o parent= and --internal flags limit communications to other containers inside of network: %s",
@@ -104,24 +119,33 @@ func (d *driver) createNetwork(config *configuration) error {
 		} else {
 			// if the subinterface parent_iface.vlan_id checks do not pass, return err.
 			//  a valid example is 'eth0.10' for a parent iface 'eth0' with a vlan id '10'
-			err := createVlanLink(config.Parent)
-			if err != nil {
-				return err
+
+			if !vlanLinkExists(config.Parent) {
+				// if the subinterface parent_iface.vlan_id checks do not pass, return err.
+				//  a valid example is 'eth0.10' for a parent iface 'eth0' with a vlan id '10'
+				err := createVlanLink(config.Parent)
+				if err != nil {
+					return false, err
+				}
+				// if driver created the networks slave link, record it for future deletion
+				config.CreatedSlaveLink = true
+			} else {
+				logrus.Debugf("Parent Sub Interface %s already Exists NetID %s", config.Parent, config.ID)
 			}
-			// if driver created the networks slave link, record it for future deletion
-			config.CreatedSlaveLink = true
 		}
 	}
-	n := &network{
-		id:        config.ID,
-		driver:    d,
-		endpoints: endpointTable{},
-		config:    config,
+	if !foundExisting {
+		n := &network{
+			id:        config.ID,
+			driver:    d,
+			endpoints: endpointTable{},
+			config:    config,
+		}
+		// add the network
+		d.addNetwork(n)
 	}
-	// add the *network
-	d.addNetwork(n)
 
-	return nil
+	return foundExisting, nil
 }
 
 // DeleteNetwork deletes the network for the specified driver type

libnetwork/drivers/macvlan/macvlan_setup.go

@@ -74,6 +74,15 @@ func parentExists(ifaceStr string) bool {
 	return true
 }
 
+// vlanLinkExists checks if specified vlan link exists in the default namespace
+func vlanLinkExists(linkStr string) bool {
+	_, err := ns.NlHandle().LinkByName(linkStr)
+	if err != nil {
+		return false
+	}
+	return true
+}
+
 // createVlanLink parses sub-interfaces and vlan id for creation
 func createVlanLink(parentName string) error {
 	if strings.Contains(parentName, ".") {
@@ -160,6 +169,15 @@ func parseVlan(linkName string) (string, int, error) {
 	return parent, vidInt, nil
 }
 
+// dummyLinkExists checks if dummylink exists in the default namespace
+func dummyLinkExists(dummyName string) bool {
+	_, err := ns.NlHandle().LinkByName(dummyName)
+	if err != nil {
+		return false
+	}
+	return true
+}
+
 // createDummyLink creates a dummy0 parent link
 func createDummyLink(dummyName, truncNetID string) error {
 	// create a parent interface since one was not specified

libnetwork/drivers/macvlan/macvlan_store.go

@@ -55,7 +55,15 @@ func (d *driver) initStore(option map[string]interface{}) error {
 			return types.InternalErrorf("macvlan driver failed to initialize data store: %v", err)
 		}
 
-		return d.populateNetworks()
+		err = d.populateNetworks()
+		if err != nil {
+			return err
+		}
+		err = d.populateEndpoints()
+		if err != nil {
+			return err
+		}
+
 	}
 
 	return nil
@@ -73,7 +81,7 @@ func (d *driver) populateNetworks() error {
 	}
 	for _, kvo := range kvol {
 		config := kvo.(*configuration)
-		if err = d.createNetwork(config); err != nil {
+		if _, err = d.createNetwork(config); err != nil {
 			logrus.Warnf("Could not create macvlan network for id %s from persistent state", config.ID)
 		}
 	}