1
0
Fork 0
mirror of https://github.com/moby/moby.git synced 2022-11-09 12:21:53 -05:00
moby--moby/libnetwork/drivers/overlay/overlay.go

392 lines
10 KiB
Go
Raw Normal View History

package overlay
//go:generate protoc -I.:../../Godeps/_workspace/src/github.com/gogo/protobuf --gogo_out=import_path=github.com/docker/libnetwork/drivers/overlay,Mgogoproto/gogo.proto=github.com/gogo/protobuf/gogoproto:. overlay.proto
import (
"context"
"fmt"
"net"
"sync"
"github.com/docker/libnetwork/datastore"
"github.com/docker/libnetwork/discoverapi"
"github.com/docker/libnetwork/driverapi"
"github.com/docker/libnetwork/idm"
"github.com/docker/libnetwork/netlabel"
"github.com/docker/libnetwork/osl"
"github.com/docker/libnetwork/types"
"github.com/hashicorp/serf/serf"
"github.com/sirupsen/logrus"
)
const (
networkType = "overlay"
vethPrefix = "veth"
vethLen = 7
vxlanIDStart = 256
vxlanIDEnd = (1 << 24) - 1
vxlanEncap = 50
secureOption = "encrypted"
)
var initVxlanIdm = make(chan (bool), 1)
type driver struct {
eventCh chan serf.Event
notifyCh chan ovNotify
exitCh chan chan struct{}
bindAddress string
advertiseAddress string
neighIP string
config map[string]interface{}
peerDb peerNetworkMap
secMap *encrMap
serfInstance *serf.Serf
networks networkTable
store datastore.DataStore
localStore datastore.DataStore
vxlanIdm *idm.Idm
initOS sync.Once
joinOnce sync.Once
localJoinOnce sync.Once
keys []*key
peerOpCh chan *peerOperation
peerOpCancel context.CancelFunc
sync.Mutex
}
// Init registers a new instance of overlay driver
func Init(dc driverapi.DriverCallback, config map[string]interface{}) error {
c := driverapi.Capability{
DataScope: datastore.GlobalScope,
ConnectivityScope: datastore.GlobalScope,
}
d := &driver{
networks: networkTable{},
peerDb: peerNetworkMap{
mp: map[string]*peerMap{},
},
secMap: &encrMap{nodes: map[string][]*spi{}},
config: config,
peerOpCh: make(chan *peerOperation),
}
// Launch the go routine for processing peer operations
ctx, cancel := context.WithCancel(context.Background())
d.peerOpCancel = cancel
go d.peerOpRoutine(ctx, d.peerOpCh)
if data, ok := config[netlabel.GlobalKVClient]; ok {
var err error
dsc, ok := data.(discoverapi.DatastoreConfigData)
if !ok {
return types.InternalErrorf("incorrect data in datastore configuration: %v", data)
}
d.store, err = datastore.NewDataStoreFromConfig(dsc)
if err != nil {
return types.InternalErrorf("failed to initialize data store: %v", err)
}
}
if data, ok := config[netlabel.LocalKVClient]; ok {
var err error
dsc, ok := data.(discoverapi.DatastoreConfigData)
if !ok {
return types.InternalErrorf("incorrect data in datastore configuration: %v", data)
}
d.localStore, err = datastore.NewDataStoreFromConfig(dsc)
if err != nil {
return types.InternalErrorf("failed to initialize local data store: %v", err)
}
}
if err := d.restoreEndpoints(); err != nil {
logrus.Warnf("Failure during overlay endpoints restore: %v", err)
}
return dc.RegisterDriver(networkType, d, c)
}
// Endpoints are stored in the local store. Restore them and reconstruct the overlay sandbox
func (d *driver) restoreEndpoints() error {
if d.localStore == nil {
logrus.Warn("Cannot restore overlay endpoints because local datastore is missing")
return nil
}
kvol, err := d.localStore.List(datastore.Key(overlayEndpointPrefix), &endpoint{})
if err != nil && err != datastore.ErrKeyNotFound {
return fmt.Errorf("failed to read overlay endpoint from store: %v", err)
}
if err == datastore.ErrKeyNotFound {
return nil
}
for _, kvo := range kvol {
ep := kvo.(*endpoint)
n := d.network(ep.nid)
if n == nil {
logrus.Debugf("Network (%.7s) not found for restored endpoint (%.7s)", ep.nid, ep.id)
logrus.Debugf("Deleting stale overlay endpoint (%.7s) from store", ep.id)
if err := d.deleteEndpointFromStore(ep); err != nil {
logrus.Debugf("Failed to delete stale overlay endpoint (%.7s) from store", ep.id)
}
continue
}
n.addEndpoint(ep)
s := n.getSubnetforIP(ep.addr)
if s == nil {
return fmt.Errorf("could not find subnet for endpoint %s", ep.id)
}
Refactor locking for join/leave to avoid race Instead of using "sync.Once" to determine whether to initialize a network sandbox or subnet sandbox, we use a traditional mutex + initialization boolean. This is because the initialization state isn't truly a once-and-done condition. Rather, libnetwork destroys network and subnet sandboxes when the last endpoint leaves them. The use of sync.Once in this kind of scenario requires, therefore, re-initializing the Once which is impoissible. So the approach that libnetwork currently takes is to use a pointer to a Once and redirect that pointer to a new Once on reset. This leads to nasty race conditions. In addition to refactoring the locking, this patch merges the functions joinSandbox(), and joinSubnetSandbox(). This makes the code both cleaner and it also holds the network and subnet locks through the series of read-modify-writes avoiding further potential races. This does reduce the potential parallelism which could be applied should there be many joins coming in on many different subnets in the same overlay network. However, this should be an extremely minor performance hit for a very obscure case. One important pattern in this commit is that it is crucial to avoid sending peerDB messages while holding a driver or network lock. The changes herein defer such (asynchronous) notifications until after release of such locks. This prevents deadlocks where the peerDB blocks acquiring said locks while the network method blocks trying to send to the peerDB's channel. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-05-04 14:33:00 -04:00
if err := n.joinSandbox(s, true, true); err != nil {
return fmt.Errorf("restore network sandbox failed: %v", err)
}
Ifaces := make(map[string][]osl.IfaceOption)
vethIfaceOption := make([]osl.IfaceOption, 1)
vethIfaceOption = append(vethIfaceOption, n.sbox.InterfaceOptions().Master(s.brName))
Ifaces["veth+veth"] = vethIfaceOption
err := n.sbox.Restore(Ifaces, nil, nil, nil)
if err != nil {
Refactor locking for join/leave to avoid race Instead of using "sync.Once" to determine whether to initialize a network sandbox or subnet sandbox, we use a traditional mutex + initialization boolean. This is because the initialization state isn't truly a once-and-done condition. Rather, libnetwork destroys network and subnet sandboxes when the last endpoint leaves them. The use of sync.Once in this kind of scenario requires, therefore, re-initializing the Once which is impoissible. So the approach that libnetwork currently takes is to use a pointer to a Once and redirect that pointer to a new Once on reset. This leads to nasty race conditions. In addition to refactoring the locking, this patch merges the functions joinSandbox(), and joinSubnetSandbox(). This makes the code both cleaner and it also holds the network and subnet locks through the series of read-modify-writes avoiding further potential races. This does reduce the potential parallelism which could be applied should there be many joins coming in on many different subnets in the same overlay network. However, this should be an extremely minor performance hit for a very obscure case. One important pattern in this commit is that it is crucial to avoid sending peerDB messages while holding a driver or network lock. The changes herein defer such (asynchronous) notifications until after release of such locks. This prevents deadlocks where the peerDB blocks acquiring said locks while the network method blocks trying to send to the peerDB's channel. Signed-off-by: Chris Telfer <ctelfer@docker.com>
2018-05-04 14:33:00 -04:00
n.leaveSandbox()
return fmt.Errorf("failed to restore overlay sandbox: %v", err)
}
d.peerAdd(ep.nid, ep.id, ep.addr.IP, ep.addr.Mask, ep.mac, net.ParseIP(d.advertiseAddress), false, false, true)
}
return nil
}
// Fini cleans up the driver resources
func Fini(drv driverapi.Driver) {
d := drv.(*driver)
// Notify the peer go routine to return
if d.peerOpCancel != nil {
d.peerOpCancel()
}
if d.exitCh != nil {
waitCh := make(chan struct{})
d.exitCh <- waitCh
<-waitCh
}
}
func (d *driver) configure() error {
// Apply OS specific kernel configs if needed
d.initOS.Do(applyOStweaks)
if d.store == nil {
return nil
}
if d.vxlanIdm == nil {
return d.initializeVxlanIdm()
}
return nil
}
func (d *driver) initializeVxlanIdm() error {
var err error
initVxlanIdm <- true
defer func() { <-initVxlanIdm }()
if d.vxlanIdm != nil {
return nil
}
d.vxlanIdm, err = idm.New(d.store, "vxlan-id", vxlanIDStart, vxlanIDEnd)
if err != nil {
return fmt.Errorf("failed to initialize vxlan id manager: %v", err)
}
return nil
}
func (d *driver) Type() string {
return networkType
}
func (d *driver) IsBuiltIn() bool {
return true
}
func validateSelf(node string) error {
advIP := net.ParseIP(node)
if advIP == nil {
return fmt.Errorf("invalid self address (%s)", node)
}
addrs, err := net.InterfaceAddrs()
if err != nil {
return fmt.Errorf("Unable to get interface addresses %v", err)
}
for _, addr := range addrs {
ip, _, err := net.ParseCIDR(addr.String())
if err == nil && ip.Equal(advIP) {
return nil
}
}
return fmt.Errorf("Multi-Host overlay networking requires cluster-advertise(%s) to be configured with a local ip-address that is reachable within the cluster", advIP.String())
}
func (d *driver) nodeJoin(advertiseAddress, bindAddress string, self bool) {
if self && !d.isSerfAlive() {
d.Lock()
d.advertiseAddress = advertiseAddress
d.bindAddress = bindAddress
d.Unlock()
// If containers are already running on this network update the
// advertise address in the peerDB
d.localJoinOnce.Do(func() {
d.peerDBUpdateSelf()
})
// If there is no cluster store there is no need to start serf.
if d.store != nil {
if err := validateSelf(advertiseAddress); err != nil {
logrus.Warn(err.Error())
}
err := d.serfInit()
if err != nil {
logrus.Errorf("initializing serf instance failed: %v", err)
d.Lock()
d.advertiseAddress = ""
d.bindAddress = ""
d.Unlock()
return
}
}
}
d.Lock()
if !self {
d.neighIP = advertiseAddress
}
neighIP := d.neighIP
d.Unlock()
if d.serfInstance != nil && neighIP != "" {
var err error
d.joinOnce.Do(func() {
err = d.serfJoin(neighIP)
if err == nil {
d.pushLocalDb()
}
})
if err != nil {
logrus.Errorf("joining serf neighbor %s failed: %v", advertiseAddress, err)
d.Lock()
d.joinOnce = sync.Once{}
d.Unlock()
return
}
}
}
func (d *driver) pushLocalEndpointEvent(action, nid, eid string) {
n := d.network(nid)
if n == nil {
logrus.Debugf("Error pushing local endpoint event for network %s", nid)
return
}
ep := n.endpoint(eid)
if ep == nil {
logrus.Debugf("Error pushing local endpoint event for ep %s / %s", nid, eid)
return
}
if !d.isSerfAlive() {
return
}
d.notifyCh <- ovNotify{
action: "join",
nw: n,
ep: ep,
}
}
// DiscoverNew is a notification for a new discovery event, such as a new node joining a cluster
func (d *driver) DiscoverNew(dType discoverapi.DiscoveryType, data interface{}) error {
var err error
switch dType {
case discoverapi.NodeDiscovery:
nodeData, ok := data.(discoverapi.NodeDiscoveryData)
if !ok || nodeData.Address == "" {
return fmt.Errorf("invalid discovery data")
}
d.nodeJoin(nodeData.Address, nodeData.BindAddress, nodeData.Self)
case discoverapi.DatastoreConfig:
if d.store != nil {
return types.ForbiddenErrorf("cannot accept datastore configuration: Overlay driver has a datastore configured already")
}
dsc, ok := data.(discoverapi.DatastoreConfigData)
if !ok {
return types.InternalErrorf("incorrect data in datastore configuration: %v", data)
}
d.store, err = datastore.NewDataStoreFromConfig(dsc)
if err != nil {
return types.InternalErrorf("failed to initialize data store: %v", err)
}
case discoverapi.EncryptionKeysConfig:
encrData, ok := data.(discoverapi.DriverEncryptionConfig)
if !ok {
return fmt.Errorf("invalid encryption key notification data")
}
keys := make([]*key, 0, len(encrData.Keys))
for i := 0; i < len(encrData.Keys); i++ {
k := &key{
value: encrData.Keys[i],
tag: uint32(encrData.Tags[i]),
}
keys = append(keys, k)
}
if err := d.setKeys(keys); err != nil {
logrus.Warn(err)
}
case discoverapi.EncryptionKeysUpdate:
var newKey, delKey, priKey *key
encrData, ok := data.(discoverapi.DriverEncryptionUpdate)
if !ok {
return fmt.Errorf("invalid encryption key notification data")
}
if encrData.Key != nil {
newKey = &key{
value: encrData.Key,
tag: uint32(encrData.Tag),
}
}
if encrData.Primary != nil {
priKey = &key{
value: encrData.Primary,
tag: uint32(encrData.PrimaryTag),
}
}
if encrData.Prune != nil {
delKey = &key{
value: encrData.Prune,
tag: uint32(encrData.PruneTag),
}
}
if err := d.updateKeys(newKey, priKey, delKey); err != nil {
logrus.Warn(err)
}
default:
}
return nil
}
// DiscoverDelete is a notification for a discovery delete event, such as a node leaving a cluster
func (d *driver) DiscoverDelete(dType discoverapi.DiscoveryType, data interface{}) error {
return nil
}