github.com/khulnasoft-lab/khulnasoft@v26.0.1-0.20240328202558-330a6f959fe0+incompatible/libnetwork/controller.go (about)

     1  /*
     2  Package libnetwork provides the basic functionality and extension points to
     3  create network namespaces and allocate interfaces for containers to use.
     4  
     5  	networkType := "bridge"
     6  
     7  	// Create a new controller instance
     8  	driverOptions := options.Generic{}
     9  	genericOption := make(map[string]interface{})
    10  	genericOption[netlabel.GenericData] = driverOptions
    11  	controller, err := libnetwork.New(config.OptionDriverConfig(networkType, genericOption))
    12  	if err != nil {
    13  		return
    14  	}
    15  
    16  	// Create a network for containers to join.
    17  	// NewNetwork accepts Variadic optional arguments that libnetwork and Drivers can make use of
    18  	network, err := controller.NewNetwork(networkType, "network1", "")
    19  	if err != nil {
    20  		return
    21  	}
    22  
    23  	// For each new container: allocate IP and interfaces. The returned network
    24  	// settings will be used for container infos (inspect and such), as well as
    25  	// iptables rules for port publishing. This info is contained or accessible
    26  	// from the returned endpoint.
    27  	ep, err := network.CreateEndpoint("Endpoint1")
    28  	if err != nil {
    29  		return
    30  	}
    31  
    32  	// Create the sandbox for the container.
    33  	// NewSandbox accepts Variadic optional arguments which libnetwork can use.
    34  	sbx, err := controller.NewSandbox("container1",
    35  		libnetwork.OptionHostname("test"),
    36  		libnetwork.OptionDomainname("example.com"))
    37  
    38  	// A sandbox can join the endpoint via the join api.
    39  	err = ep.Join(sbx)
    40  	if err != nil {
    41  		return
    42  	}
    43  */
    44  package libnetwork
    45  
    46  import (
    47  	"context"
    48  	"fmt"
    49  	"net"
    50  	"path/filepath"
    51  	"runtime"
    52  	"strings"
    53  	"sync"
    54  	"time"
    55  
    56  	"github.com/containerd/log"
    57  	"github.com/docker/docker/libnetwork/cluster"
    58  	"github.com/docker/docker/libnetwork/config"
    59  	"github.com/docker/docker/libnetwork/datastore"
    60  	"github.com/docker/docker/libnetwork/diagnostic"
    61  	"github.com/docker/docker/libnetwork/discoverapi"
    62  	"github.com/docker/docker/libnetwork/driverapi"
    63  	remotedriver "github.com/docker/docker/libnetwork/drivers/remote"
    64  	"github.com/docker/docker/libnetwork/drvregistry"
    65  	"github.com/docker/docker/libnetwork/ipamapi"
    66  	"github.com/docker/docker/libnetwork/netlabel"
    67  	"github.com/docker/docker/libnetwork/osl"
    68  	"github.com/docker/docker/libnetwork/scope"
    69  	"github.com/docker/docker/libnetwork/types"
    70  	"github.com/docker/docker/pkg/plugingetter"
    71  	"github.com/docker/docker/pkg/plugins"
    72  	"github.com/docker/docker/pkg/stringid"
    73  	"github.com/moby/locker"
    74  	"github.com/pkg/errors"
    75  )
    76  
    77  // NetworkWalker is a client provided function which will be used to walk the Networks.
    78  // When the function returns true, the walk will stop.
    79  type NetworkWalker func(nw *Network) bool
    80  
    81  // Controller manages networks.
    82  type Controller struct {
    83  	id               string
    84  	drvRegistry      drvregistry.Networks
    85  	ipamRegistry     drvregistry.IPAMs
    86  	sandboxes        map[string]*Sandbox
    87  	cfg              *config.Config
    88  	store            *datastore.Store
    89  	extKeyListener   net.Listener
    90  	svcRecords       map[string]*svcInfo
    91  	serviceBindings  map[serviceKey]*service
    92  	ingressSandbox   *Sandbox
    93  	agent            *nwAgent
    94  	networkLocker    *locker.Locker
    95  	agentInitDone    chan struct{}
    96  	agentStopDone    chan struct{}
    97  	keys             []*types.EncryptionKey
    98  	DiagnosticServer *diagnostic.Server
    99  	mu               sync.Mutex
   100  
   101  	// FIXME(thaJeztah): defOsSbox is always nil on non-Linux: move these fields to Linux-only files.
   102  	defOsSboxOnce sync.Once
   103  	defOsSbox     *osl.Namespace
   104  }
   105  
   106  // New creates a new instance of network controller.
   107  func New(cfgOptions ...config.Option) (*Controller, error) {
   108  	c := &Controller{
   109  		id:               stringid.GenerateRandomID(),
   110  		cfg:              config.New(cfgOptions...),
   111  		sandboxes:        map[string]*Sandbox{},
   112  		svcRecords:       make(map[string]*svcInfo),
   113  		serviceBindings:  make(map[serviceKey]*service),
   114  		agentInitDone:    make(chan struct{}),
   115  		networkLocker:    locker.New(),
   116  		DiagnosticServer: diagnostic.New(),
   117  	}
   118  
   119  	if err := c.initStores(); err != nil {
   120  		return nil, err
   121  	}
   122  
   123  	c.drvRegistry.Notify = c
   124  
   125  	// External plugins don't need config passed through daemon. They can
   126  	// bootstrap themselves.
   127  	if err := remotedriver.Register(&c.drvRegistry, c.cfg.PluginGetter); err != nil {
   128  		return nil, err
   129  	}
   130  
   131  	if err := registerNetworkDrivers(&c.drvRegistry, c.makeDriverConfig); err != nil {
   132  		return nil, err
   133  	}
   134  
   135  	if err := initIPAMDrivers(&c.ipamRegistry, c.cfg.PluginGetter, c.cfg.DefaultAddressPool); err != nil {
   136  		return nil, err
   137  	}
   138  
   139  	c.WalkNetworks(func(nw *Network) bool {
   140  		if n := nw; n.hasSpecialDriver() && !n.ConfigOnly() {
   141  			if err := n.getController().addNetwork(n); err != nil {
   142  				log.G(context.TODO()).Warnf("Failed to populate network %q with driver %q", nw.Name(), nw.Type())
   143  			}
   144  		}
   145  		return false
   146  	})
   147  
   148  	// Reserve pools first before doing cleanup. Otherwise the
   149  	// cleanups of endpoint/network and sandbox below will
   150  	// generate many unnecessary warnings
   151  	c.reservePools()
   152  
   153  	// Cleanup resources
   154  	if err := c.sandboxCleanup(c.cfg.ActiveSandboxes); err != nil {
   155  		log.G(context.TODO()).WithError(err).Error("error during sandbox cleanup")
   156  	}
   157  	if err := c.cleanupLocalEndpoints(); err != nil {
   158  		log.G(context.TODO()).WithError(err).Warnf("error during endpoint cleanup")
   159  	}
   160  	c.networkCleanup()
   161  
   162  	if err := c.startExternalKeyListener(); err != nil {
   163  		return nil, err
   164  	}
   165  
   166  	setupArrangeUserFilterRule(c)
   167  	return c, nil
   168  }
   169  
   170  // SetClusterProvider sets the cluster provider.
   171  func (c *Controller) SetClusterProvider(provider cluster.Provider) {
   172  	var sameProvider bool
   173  	c.mu.Lock()
   174  	// Avoids to spawn multiple goroutine for the same cluster provider
   175  	if c.cfg.ClusterProvider == provider {
   176  		// If the cluster provider is already set, there is already a go routine spawned
   177  		// that is listening for events, so nothing to do here
   178  		sameProvider = true
   179  	} else {
   180  		c.cfg.ClusterProvider = provider
   181  	}
   182  	c.mu.Unlock()
   183  
   184  	if provider == nil || sameProvider {
   185  		return
   186  	}
   187  	// We don't want to spawn a new go routine if the previous one did not exit yet
   188  	c.AgentStopWait()
   189  	go c.clusterAgentInit()
   190  }
   191  
   192  // SetKeys configures the encryption key for gossip and overlay data path.
   193  func (c *Controller) SetKeys(keys []*types.EncryptionKey) error {
   194  	// libnetwork side of agent depends on the keys. On the first receipt of
   195  	// keys setup the agent. For subsequent key set handle the key change
   196  	subsysKeys := make(map[string]int)
   197  	for _, key := range keys {
   198  		if key.Subsystem != subsysGossip &&
   199  			key.Subsystem != subsysIPSec {
   200  			return fmt.Errorf("key received for unrecognized subsystem")
   201  		}
   202  		subsysKeys[key.Subsystem]++
   203  	}
   204  	for s, count := range subsysKeys {
   205  		if count != keyringSize {
   206  			return fmt.Errorf("incorrect number of keys for subsystem %v", s)
   207  		}
   208  	}
   209  
   210  	if c.getAgent() == nil {
   211  		c.mu.Lock()
   212  		c.keys = keys
   213  		c.mu.Unlock()
   214  		return nil
   215  	}
   216  	return c.handleKeyChange(keys)
   217  }
   218  
   219  func (c *Controller) getAgent() *nwAgent {
   220  	c.mu.Lock()
   221  	defer c.mu.Unlock()
   222  	return c.agent
   223  }
   224  
   225  func (c *Controller) clusterAgentInit() {
   226  	clusterProvider := c.cfg.ClusterProvider
   227  	var keysAvailable bool
   228  	for {
   229  		eventType := <-clusterProvider.ListenClusterEvents()
   230  		// The events: EventSocketChange, EventNodeReady and EventNetworkKeysAvailable are not ordered
   231  		// when all the condition for the agent initialization are met then proceed with it
   232  		switch eventType {
   233  		case cluster.EventNetworkKeysAvailable:
   234  			// Validates that the keys are actually available before starting the initialization
   235  			// This will handle old spurious messages left on the channel
   236  			c.mu.Lock()
   237  			keysAvailable = c.keys != nil
   238  			c.mu.Unlock()
   239  			fallthrough
   240  		case cluster.EventSocketChange, cluster.EventNodeReady:
   241  			if keysAvailable && c.isSwarmNode() {
   242  				c.agentOperationStart()
   243  				if err := c.agentSetup(clusterProvider); err != nil {
   244  					c.agentStopComplete()
   245  				} else {
   246  					c.agentInitComplete()
   247  				}
   248  			}
   249  		case cluster.EventNodeLeave:
   250  			c.agentOperationStart()
   251  			c.mu.Lock()
   252  			c.keys = nil
   253  			c.mu.Unlock()
   254  
   255  			// We are leaving the cluster. Make sure we
   256  			// close the gossip so that we stop all
   257  			// incoming gossip updates before cleaning up
   258  			// any remaining service bindings. But before
   259  			// deleting the networks since the networks
   260  			// should still be present when cleaning up
   261  			// service bindings
   262  			c.agentClose()
   263  			c.cleanupServiceDiscovery("")
   264  			c.cleanupServiceBindings("")
   265  
   266  			c.agentStopComplete()
   267  
   268  			return
   269  		}
   270  	}
   271  }
   272  
   273  // AgentInitWait waits for agent initialization to be completed in the controller.
   274  func (c *Controller) AgentInitWait() {
   275  	c.mu.Lock()
   276  	agentInitDone := c.agentInitDone
   277  	c.mu.Unlock()
   278  
   279  	if agentInitDone != nil {
   280  		<-agentInitDone
   281  	}
   282  }
   283  
   284  // AgentStopWait waits for the Agent stop to be completed in the controller.
   285  func (c *Controller) AgentStopWait() {
   286  	c.mu.Lock()
   287  	agentStopDone := c.agentStopDone
   288  	c.mu.Unlock()
   289  	if agentStopDone != nil {
   290  		<-agentStopDone
   291  	}
   292  }
   293  
   294  // agentOperationStart marks the start of an Agent Init or Agent Stop
   295  func (c *Controller) agentOperationStart() {
   296  	c.mu.Lock()
   297  	if c.agentInitDone == nil {
   298  		c.agentInitDone = make(chan struct{})
   299  	}
   300  	if c.agentStopDone == nil {
   301  		c.agentStopDone = make(chan struct{})
   302  	}
   303  	c.mu.Unlock()
   304  }
   305  
   306  // agentInitComplete notifies the successful completion of the Agent initialization
   307  func (c *Controller) agentInitComplete() {
   308  	c.mu.Lock()
   309  	if c.agentInitDone != nil {
   310  		close(c.agentInitDone)
   311  		c.agentInitDone = nil
   312  	}
   313  	c.mu.Unlock()
   314  }
   315  
   316  // agentStopComplete notifies the successful completion of the Agent stop
   317  func (c *Controller) agentStopComplete() {
   318  	c.mu.Lock()
   319  	if c.agentStopDone != nil {
   320  		close(c.agentStopDone)
   321  		c.agentStopDone = nil
   322  	}
   323  	c.mu.Unlock()
   324  }
   325  
   326  func (c *Controller) makeDriverConfig(ntype string) map[string]interface{} {
   327  	if c.cfg == nil {
   328  		return nil
   329  	}
   330  
   331  	cfg := map[string]interface{}{}
   332  	for _, label := range c.cfg.Labels {
   333  		key, val, _ := strings.Cut(label, "=")
   334  		if !strings.HasPrefix(key, netlabel.DriverPrefix+"."+ntype) {
   335  			continue
   336  		}
   337  
   338  		cfg[key] = val
   339  	}
   340  
   341  	// Merge in the existing config for this driver.
   342  	for k, v := range c.cfg.DriverConfig(ntype) {
   343  		cfg[k] = v
   344  	}
   345  
   346  	if c.cfg.Scope.IsValid() {
   347  		cfg[netlabel.LocalKVClient] = c.store
   348  	}
   349  
   350  	return cfg
   351  }
   352  
   353  // ID returns the controller's unique identity.
   354  func (c *Controller) ID() string {
   355  	return c.id
   356  }
   357  
   358  // BuiltinDrivers returns the list of builtin network drivers.
   359  func (c *Controller) BuiltinDrivers() []string {
   360  	drivers := []string{}
   361  	c.drvRegistry.WalkDrivers(func(name string, driver driverapi.Driver, capability driverapi.Capability) bool {
   362  		if driver.IsBuiltIn() {
   363  			drivers = append(drivers, name)
   364  		}
   365  		return false
   366  	})
   367  	return drivers
   368  }
   369  
   370  // BuiltinIPAMDrivers returns the list of builtin ipam drivers.
   371  func (c *Controller) BuiltinIPAMDrivers() []string {
   372  	drivers := []string{}
   373  	c.ipamRegistry.WalkIPAMs(func(name string, driver ipamapi.Ipam, _ *ipamapi.Capability) bool {
   374  		if driver.IsBuiltIn() {
   375  			drivers = append(drivers, name)
   376  		}
   377  		return false
   378  	})
   379  	return drivers
   380  }
   381  
   382  func (c *Controller) processNodeDiscovery(nodes []net.IP, add bool) {
   383  	c.drvRegistry.WalkDrivers(func(name string, driver driverapi.Driver, capability driverapi.Capability) bool {
   384  		if d, ok := driver.(discoverapi.Discover); ok {
   385  			c.pushNodeDiscovery(d, capability, nodes, add)
   386  		}
   387  		return false
   388  	})
   389  }
   390  
   391  func (c *Controller) pushNodeDiscovery(d discoverapi.Discover, capability driverapi.Capability, nodes []net.IP, add bool) {
   392  	var self net.IP
   393  	// try swarm-mode config
   394  	if agent := c.getAgent(); agent != nil {
   395  		self = net.ParseIP(agent.advertiseAddr)
   396  	}
   397  
   398  	if d == nil || capability.ConnectivityScope != scope.Global || nodes == nil {
   399  		return
   400  	}
   401  
   402  	for _, node := range nodes {
   403  		nodeData := discoverapi.NodeDiscoveryData{Address: node.String(), Self: node.Equal(self)}
   404  		var err error
   405  		if add {
   406  			err = d.DiscoverNew(discoverapi.NodeDiscovery, nodeData)
   407  		} else {
   408  			err = d.DiscoverDelete(discoverapi.NodeDiscovery, nodeData)
   409  		}
   410  		if err != nil {
   411  			log.G(context.TODO()).Debugf("discovery notification error: %v", err)
   412  		}
   413  	}
   414  }
   415  
   416  // Config returns the bootup configuration for the controller.
   417  func (c *Controller) Config() config.Config {
   418  	c.mu.Lock()
   419  	defer c.mu.Unlock()
   420  	if c.cfg == nil {
   421  		return config.Config{}
   422  	}
   423  	return *c.cfg
   424  }
   425  
   426  func (c *Controller) isManager() bool {
   427  	c.mu.Lock()
   428  	defer c.mu.Unlock()
   429  	if c.cfg == nil || c.cfg.ClusterProvider == nil {
   430  		return false
   431  	}
   432  	return c.cfg.ClusterProvider.IsManager()
   433  }
   434  
   435  func (c *Controller) isAgent() bool {
   436  	c.mu.Lock()
   437  	defer c.mu.Unlock()
   438  	if c.cfg == nil || c.cfg.ClusterProvider == nil {
   439  		return false
   440  	}
   441  	return c.cfg.ClusterProvider.IsAgent()
   442  }
   443  
   444  func (c *Controller) isSwarmNode() bool {
   445  	return c.isManager() || c.isAgent()
   446  }
   447  
   448  func (c *Controller) GetPluginGetter() plugingetter.PluginGetter {
   449  	return c.cfg.PluginGetter
   450  }
   451  
   452  func (c *Controller) RegisterDriver(networkType string, driver driverapi.Driver, capability driverapi.Capability) error {
   453  	if d, ok := driver.(discoverapi.Discover); ok {
   454  		c.agentDriverNotify(d)
   455  	}
   456  	return nil
   457  }
   458  
   459  // XXX  This should be made driver agnostic.  See comment below.
   460  const overlayDSROptionString = "dsr"
   461  
   462  // NewNetwork creates a new network of the specified network type. The options
   463  // are network specific and modeled in a generic way.
   464  func (c *Controller) NewNetwork(networkType, name string, id string, options ...NetworkOption) (_ *Network, retErr error) {
   465  	if id != "" {
   466  		c.networkLocker.Lock(id)
   467  		defer c.networkLocker.Unlock(id) //nolint:errcheck
   468  
   469  		if _, err := c.NetworkByID(id); err == nil {
   470  			return nil, NetworkNameError(id)
   471  		}
   472  	}
   473  
   474  	if strings.TrimSpace(name) == "" {
   475  		return nil, ErrInvalidName(name)
   476  	}
   477  
   478  	// Make sure two concurrent calls to this method won't create conflicting
   479  	// networks, otherwise libnetwork will end up in an invalid state.
   480  	if name != "" {
   481  		c.networkLocker.Lock(name)
   482  		defer c.networkLocker.Unlock(name)
   483  
   484  		if _, err := c.NetworkByName(name); err == nil {
   485  			return nil, NetworkNameError(name)
   486  		}
   487  	}
   488  
   489  	if id == "" {
   490  		id = stringid.GenerateRandomID()
   491  	}
   492  
   493  	defaultIpam := defaultIpamForNetworkType(networkType)
   494  	// Construct the network object
   495  	nw := &Network{
   496  		name:             name,
   497  		networkType:      networkType,
   498  		generic:          map[string]interface{}{netlabel.GenericData: make(map[string]string)},
   499  		ipamType:         defaultIpam,
   500  		id:               id,
   501  		created:          time.Now(),
   502  		ctrlr:            c,
   503  		persist:          true,
   504  		drvOnce:          &sync.Once{},
   505  		loadBalancerMode: loadBalancerModeDefault,
   506  	}
   507  
   508  	nw.processOptions(options...)
   509  	if err := nw.validateConfiguration(); err != nil {
   510  		return nil, err
   511  	}
   512  
   513  	// These variables must be defined here, as declaration would otherwise
   514  	// be skipped by the "goto addToStore"
   515  	var (
   516  		caps driverapi.Capability
   517  		err  error
   518  
   519  		skipCfgEpCount bool
   520  	)
   521  
   522  	// Reset network types, force local scope and skip allocation and
   523  	// plumbing for configuration networks. Reset of the config-only
   524  	// network drivers is needed so that this special network is not
   525  	// usable by old engine versions.
   526  	if nw.configOnly {
   527  		nw.scope = scope.Local
   528  		nw.networkType = "null"
   529  		goto addToStore
   530  	}
   531  
   532  	_, caps, err = nw.resolveDriver(nw.networkType, true)
   533  	if err != nil {
   534  		return nil, err
   535  	}
   536  
   537  	if nw.scope == scope.Local && caps.DataScope == scope.Global {
   538  		return nil, types.ForbiddenErrorf("cannot downgrade network scope for %s networks", networkType)
   539  	}
   540  	if nw.ingress && caps.DataScope != scope.Global {
   541  		return nil, types.ForbiddenErrorf("Ingress network can only be global scope network")
   542  	}
   543  
   544  	// At this point the network scope is still unknown if not set by user
   545  	if (caps.DataScope == scope.Global || nw.scope == scope.Swarm) &&
   546  		c.isSwarmNode() && !nw.dynamic {
   547  		if c.isManager() {
   548  			// For non-distributed controlled environment, globalscoped non-dynamic networks are redirected to Manager
   549  			return nil, ManagerRedirectError(name)
   550  		}
   551  		return nil, types.ForbiddenErrorf("Cannot create a multi-host network from a worker node. Please create the network from a manager node.")
   552  	}
   553  
   554  	if nw.scope == scope.Swarm && !c.isSwarmNode() {
   555  		return nil, types.ForbiddenErrorf("cannot create a swarm scoped network when swarm is not active")
   556  	}
   557  
   558  	// Make sure we have a driver available for this network type
   559  	// before we allocate anything.
   560  	if _, err := nw.driver(true); err != nil {
   561  		return nil, err
   562  	}
   563  
   564  	// From this point on, we need the network specific configuration,
   565  	// which may come from a configuration-only network
   566  	if nw.configFrom != "" {
   567  		configNetwork, err := c.getConfigNetwork(nw.configFrom)
   568  		if err != nil {
   569  			return nil, types.NotFoundErrorf("configuration network %q does not exist", nw.configFrom)
   570  		}
   571  		if err := configNetwork.applyConfigurationTo(nw); err != nil {
   572  			return nil, types.InternalErrorf("Failed to apply configuration: %v", err)
   573  		}
   574  		nw.generic[netlabel.Internal] = nw.internal
   575  		defer func() {
   576  			if retErr == nil && !skipCfgEpCount {
   577  				if err := configNetwork.getEpCnt().IncEndpointCnt(); err != nil {
   578  					log.G(context.TODO()).Warnf("Failed to update reference count for configuration network %q on creation of network %q: %v", configNetwork.Name(), nw.name, err)
   579  				}
   580  			}
   581  		}()
   582  	}
   583  
   584  	if err := nw.ipamAllocate(); err != nil {
   585  		return nil, err
   586  	}
   587  	defer func() {
   588  		if retErr != nil {
   589  			nw.ipamRelease()
   590  		}
   591  	}()
   592  
   593  	// Note from thaJeztah to future code visitors, or "future self".
   594  	//
   595  	// This code was previously assigning the error to the global "err"
   596  	// variable (before it was renamed to "retErr"), but in case of a
   597  	// "MaskableError" did not *return* the error:
   598  	// https://github.com/moby/moby/blob/b325dcbff60a04cedbe40eb627465fc7379d05bf/libnetwork/controller.go#L566-L573
   599  	//
   600  	// Depending on code paths further down, that meant that this error
   601  	// was either overwritten by other errors (and thus not handled in
   602  	// defer statements) or handled (if no other code was overwriting it.
   603  	//
   604  	// I suspect this was a bug (but possible without effect), but it could
   605  	// have been intentional. This logic is confusing at least, and even
   606  	// more so combined with the handling in defer statements that check for
   607  	// both the "err" return AND "skipCfgEpCount":
   608  	// https://github.com/moby/moby/blob/b325dcbff60a04cedbe40eb627465fc7379d05bf/libnetwork/controller.go#L586-L602
   609  	//
   610  	// To save future visitors some time to dig up history:
   611  	//
   612  	// - config-only networks were added in 25082206df465d1c11dd1276a65b4a1dc701bd43
   613  	// - the special error-handling and "skipCfgEpcoung" was added in ddd22a819867faa0cd7d12b0c3fad1099ac3eb26
   614  	// - and updated in 87b082f3659f9ec245ab15d781e6bfffced0af83 to don't use string-matching
   615  	//
   616  	// To cut a long story short: if this broke anything, you know who to blame :)
   617  	if err := c.addNetwork(nw); err != nil {
   618  		if _, ok := err.(types.MaskableError); ok { //nolint:gosimple
   619  			// This error can be ignored and set this boolean
   620  			// value to skip a refcount increment for configOnly networks
   621  			skipCfgEpCount = true
   622  		} else {
   623  			return nil, err
   624  		}
   625  	}
   626  	defer func() {
   627  		if retErr != nil {
   628  			if err := nw.deleteNetwork(); err != nil {
   629  				log.G(context.TODO()).Warnf("couldn't roll back driver network on network %s creation failure: %v", nw.name, retErr)
   630  			}
   631  		}
   632  	}()
   633  
   634  	// XXX If the driver type is "overlay" check the options for DSR
   635  	// being set.  If so, set the network's load balancing mode to DSR.
   636  	// This should really be done in a network option, but due to
   637  	// time pressure to get this in without adding changes to moby,
   638  	// swarm and CLI, it is being implemented as a driver-specific
   639  	// option.  Unfortunately, drivers can't influence the core
   640  	// "libnetwork.Network" data type.  Hence we need this hack code
   641  	// to implement in this manner.
   642  	if gval, ok := nw.generic[netlabel.GenericData]; ok && nw.networkType == "overlay" {
   643  		optMap := gval.(map[string]string)
   644  		if _, ok := optMap[overlayDSROptionString]; ok {
   645  			nw.loadBalancerMode = loadBalancerModeDSR
   646  		}
   647  	}
   648  
   649  addToStore:
   650  	// First store the endpoint count, then the network. To avoid to
   651  	// end up with a datastore containing a network and not an epCnt,
   652  	// in case of an ungraceful shutdown during this function call.
   653  	epCnt := &endpointCnt{n: nw}
   654  	if err := c.updateToStore(epCnt); err != nil {
   655  		return nil, err
   656  	}
   657  	defer func() {
   658  		if retErr != nil {
   659  			if err := c.deleteFromStore(epCnt); err != nil {
   660  				log.G(context.TODO()).Warnf("could not rollback from store, epCnt %v on failure (%v): %v", epCnt, retErr, err)
   661  			}
   662  		}
   663  	}()
   664  
   665  	nw.epCnt = epCnt
   666  	if err := c.updateToStore(nw); err != nil {
   667  		return nil, err
   668  	}
   669  	defer func() {
   670  		if retErr != nil {
   671  			if err := c.deleteFromStore(nw); err != nil {
   672  				log.G(context.TODO()).Warnf("could not rollback from store, network %v on failure (%v): %v", nw, retErr, err)
   673  			}
   674  		}
   675  	}()
   676  
   677  	if nw.configOnly {
   678  		return nw, nil
   679  	}
   680  
   681  	joinCluster(nw)
   682  	defer func() {
   683  		if retErr != nil {
   684  			nw.cancelDriverWatches()
   685  			if err := nw.leaveCluster(); err != nil {
   686  				log.G(context.TODO()).Warnf("Failed to leave agent cluster on network %s on failure (%v): %v", nw.name, retErr, err)
   687  			}
   688  		}
   689  	}()
   690  
   691  	if nw.hasLoadBalancerEndpoint() {
   692  		if err := nw.createLoadBalancerSandbox(); err != nil {
   693  			return nil, err
   694  		}
   695  	}
   696  
   697  	if c.isSwarmNode() {
   698  		c.mu.Lock()
   699  		arrangeIngressFilterRule()
   700  		c.mu.Unlock()
   701  	}
   702  
   703  	// Sets up the DOCKER-USER chain for each iptables version (IPv4, IPv6)
   704  	// that's enabled in the controller's configuration.
   705  	for _, ipVersion := range c.enabledIptablesVersions() {
   706  		if err := setupUserChain(ipVersion); err != nil {
   707  			log.G(context.TODO()).WithError(err).Warnf("Controller.NewNetwork %s:", name)
   708  		}
   709  	}
   710  
   711  	return nw, nil
   712  }
   713  
   714  var joinCluster NetworkWalker = func(nw *Network) bool {
   715  	if nw.configOnly {
   716  		return false
   717  	}
   718  	if err := nw.joinCluster(); err != nil {
   719  		log.G(context.TODO()).Errorf("Failed to join network %s (%s) into agent cluster: %v", nw.Name(), nw.ID(), err)
   720  	}
   721  	nw.addDriverWatches()
   722  	return false
   723  }
   724  
   725  func (c *Controller) reservePools() {
   726  	networks, err := c.getNetworks()
   727  	if err != nil {
   728  		log.G(context.TODO()).Warnf("Could not retrieve networks from local store during ipam allocation for existing networks: %v", err)
   729  		return
   730  	}
   731  
   732  	for _, n := range networks {
   733  		if n.configOnly {
   734  			continue
   735  		}
   736  		if !doReplayPoolReserve(n) {
   737  			continue
   738  		}
   739  		// Construct pseudo configs for the auto IP case
   740  		autoIPv4 := (len(n.ipamV4Config) == 0 || (len(n.ipamV4Config) == 1 && n.ipamV4Config[0].PreferredPool == "")) && len(n.ipamV4Info) > 0
   741  		autoIPv6 := (len(n.ipamV6Config) == 0 || (len(n.ipamV6Config) == 1 && n.ipamV6Config[0].PreferredPool == "")) && len(n.ipamV6Info) > 0
   742  		if autoIPv4 {
   743  			n.ipamV4Config = []*IpamConf{{PreferredPool: n.ipamV4Info[0].Pool.String()}}
   744  		}
   745  		if n.enableIPv6 && autoIPv6 {
   746  			n.ipamV6Config = []*IpamConf{{PreferredPool: n.ipamV6Info[0].Pool.String()}}
   747  		}
   748  		// Account current network gateways
   749  		for i, cfg := range n.ipamV4Config {
   750  			if cfg.Gateway == "" && n.ipamV4Info[i].Gateway != nil {
   751  				cfg.Gateway = n.ipamV4Info[i].Gateway.IP.String()
   752  			}
   753  		}
   754  		if n.enableIPv6 {
   755  			for i, cfg := range n.ipamV6Config {
   756  				if cfg.Gateway == "" && n.ipamV6Info[i].Gateway != nil {
   757  					cfg.Gateway = n.ipamV6Info[i].Gateway.IP.String()
   758  				}
   759  			}
   760  		}
   761  		// Reserve pools
   762  		if err := n.ipamAllocate(); err != nil {
   763  			log.G(context.TODO()).Warnf("Failed to allocate ipam pool(s) for network %q (%s): %v", n.Name(), n.ID(), err)
   764  		}
   765  		// Reserve existing endpoints' addresses
   766  		ipam, _, err := n.getController().getIPAMDriver(n.ipamType)
   767  		if err != nil {
   768  			log.G(context.TODO()).Warnf("Failed to retrieve ipam driver for network %q (%s) during address reservation", n.Name(), n.ID())
   769  			continue
   770  		}
   771  		epl, err := n.getEndpointsFromStore()
   772  		if err != nil {
   773  			log.G(context.TODO()).Warnf("Failed to retrieve list of current endpoints on network %q (%s)", n.Name(), n.ID())
   774  			continue
   775  		}
   776  		for _, ep := range epl {
   777  			if ep.Iface() == nil {
   778  				log.G(context.TODO()).Warnf("endpoint interface is empty for %q (%s)", ep.Name(), ep.ID())
   779  				continue
   780  			}
   781  			if err := ep.assignAddress(ipam, true, ep.Iface().AddressIPv6() != nil); err != nil {
   782  				log.G(context.TODO()).Warnf("Failed to reserve current address for endpoint %q (%s) on network %q (%s)",
   783  					ep.Name(), ep.ID(), n.Name(), n.ID())
   784  			}
   785  		}
   786  	}
   787  }
   788  
   789  func doReplayPoolReserve(n *Network) bool {
   790  	_, caps, err := n.getController().getIPAMDriver(n.ipamType)
   791  	if err != nil {
   792  		log.G(context.TODO()).Warnf("Failed to retrieve ipam driver for network %q (%s): %v", n.Name(), n.ID(), err)
   793  		return false
   794  	}
   795  	return caps.RequiresRequestReplay
   796  }
   797  
   798  func (c *Controller) addNetwork(n *Network) error {
   799  	d, err := n.driver(true)
   800  	if err != nil {
   801  		return err
   802  	}
   803  
   804  	// Create the network
   805  	if err := d.CreateNetwork(n.id, n.generic, n, n.getIPData(4), n.getIPData(6)); err != nil {
   806  		return err
   807  	}
   808  
   809  	n.startResolver()
   810  
   811  	return nil
   812  }
   813  
   814  // Networks returns the list of Network(s) managed by this controller.
   815  func (c *Controller) Networks(ctx context.Context) []*Network {
   816  	var list []*Network
   817  
   818  	for _, n := range c.getNetworksFromStore(ctx) {
   819  		if n.inDelete {
   820  			continue
   821  		}
   822  		list = append(list, n)
   823  	}
   824  
   825  	return list
   826  }
   827  
   828  // WalkNetworks uses the provided function to walk the Network(s) managed by this controller.
   829  func (c *Controller) WalkNetworks(walker NetworkWalker) {
   830  	for _, n := range c.Networks(context.TODO()) {
   831  		if walker(n) {
   832  			return
   833  		}
   834  	}
   835  }
   836  
   837  // NetworkByName returns the Network which has the passed name.
   838  // If not found, the error [ErrNoSuchNetwork] is returned.
   839  func (c *Controller) NetworkByName(name string) (*Network, error) {
   840  	if name == "" {
   841  		return nil, ErrInvalidName(name)
   842  	}
   843  	var n *Network
   844  
   845  	c.WalkNetworks(func(current *Network) bool {
   846  		if current.Name() == name {
   847  			n = current
   848  			return true
   849  		}
   850  		return false
   851  	})
   852  
   853  	if n == nil {
   854  		return nil, ErrNoSuchNetwork(name)
   855  	}
   856  
   857  	return n, nil
   858  }
   859  
   860  // NetworkByID returns the Network which has the passed id.
   861  // If not found, the error [ErrNoSuchNetwork] is returned.
   862  func (c *Controller) NetworkByID(id string) (*Network, error) {
   863  	if id == "" {
   864  		return nil, ErrInvalidID(id)
   865  	}
   866  	return c.getNetworkFromStore(id)
   867  }
   868  
   869  // NewSandbox creates a new sandbox for containerID.
   870  func (c *Controller) NewSandbox(containerID string, options ...SandboxOption) (_ *Sandbox, retErr error) {
   871  	if containerID == "" {
   872  		return nil, types.InvalidParameterErrorf("invalid container ID")
   873  	}
   874  
   875  	var sb *Sandbox
   876  	c.mu.Lock()
   877  	for _, s := range c.sandboxes {
   878  		if s.containerID == containerID {
   879  			// If not a stub, then we already have a complete sandbox.
   880  			if !s.isStub {
   881  				sbID := s.ID()
   882  				c.mu.Unlock()
   883  				return nil, types.ForbiddenErrorf("container %s is already present in sandbox %s", containerID, sbID)
   884  			}
   885  
   886  			// We already have a stub sandbox from the
   887  			// store. Make use of it so that we don't lose
   888  			// the endpoints from store but reset the
   889  			// isStub flag.
   890  			sb = s
   891  			sb.isStub = false
   892  			break
   893  		}
   894  	}
   895  	c.mu.Unlock()
   896  
   897  	// Create sandbox and process options first. Key generation depends on an option
   898  	if sb == nil {
   899  		// TODO(thaJeztah): given that a "containerID" must be unique in the list of sandboxes, is there any reason we're not using containerID as sandbox ID on non-Windows?
   900  		sandboxID := containerID
   901  		if runtime.GOOS != "windows" {
   902  			sandboxID = stringid.GenerateRandomID()
   903  		}
   904  		sb = &Sandbox{
   905  			id:                 sandboxID,
   906  			containerID:        containerID,
   907  			endpoints:          []*Endpoint{},
   908  			epPriority:         map[string]int{},
   909  			populatedEndpoints: map[string]struct{}{},
   910  			config:             containerConfig{},
   911  			controller:         c,
   912  			extDNS:             []extDNSEntry{},
   913  		}
   914  	}
   915  
   916  	sb.processOptions(options...)
   917  
   918  	c.mu.Lock()
   919  	if sb.ingress && c.ingressSandbox != nil {
   920  		c.mu.Unlock()
   921  		return nil, types.ForbiddenErrorf("ingress sandbox already present")
   922  	}
   923  
   924  	if sb.ingress {
   925  		c.ingressSandbox = sb
   926  		sb.config.hostsPath = filepath.Join(c.cfg.DataDir, "/network/files/hosts")
   927  		sb.config.resolvConfPath = filepath.Join(c.cfg.DataDir, "/network/files/resolv.conf")
   928  		sb.id = "ingress_sbox"
   929  	} else if sb.loadBalancerNID != "" {
   930  		sb.id = "lb_" + sb.loadBalancerNID
   931  	}
   932  	c.mu.Unlock()
   933  
   934  	defer func() {
   935  		if retErr != nil {
   936  			c.mu.Lock()
   937  			if sb.ingress {
   938  				c.ingressSandbox = nil
   939  			}
   940  			c.mu.Unlock()
   941  		}
   942  	}()
   943  
   944  	if err := sb.setupResolutionFiles(); err != nil {
   945  		return nil, err
   946  	}
   947  	if err := c.setupOSLSandbox(sb); err != nil {
   948  		return nil, err
   949  	}
   950  
   951  	c.mu.Lock()
   952  	c.sandboxes[sb.id] = sb
   953  	c.mu.Unlock()
   954  	defer func() {
   955  		if retErr != nil {
   956  			c.mu.Lock()
   957  			delete(c.sandboxes, sb.id)
   958  			c.mu.Unlock()
   959  		}
   960  	}()
   961  
   962  	if err := sb.storeUpdate(); err != nil {
   963  		return nil, fmt.Errorf("failed to update the store state of sandbox: %v", err)
   964  	}
   965  
   966  	return sb, nil
   967  }
   968  
   969  // GetSandbox returns the Sandbox which has the passed id.
   970  //
   971  // It returns an [ErrInvalidID] when passing an invalid ID, or an
   972  // [types.NotFoundError] if no Sandbox was found for the container.
   973  func (c *Controller) GetSandbox(containerID string) (*Sandbox, error) {
   974  	if containerID == "" {
   975  		return nil, ErrInvalidID("id is empty")
   976  	}
   977  	c.mu.Lock()
   978  	defer c.mu.Unlock()
   979  	if runtime.GOOS == "windows" {
   980  		// fast-path for Windows, which uses the container ID as sandbox ID.
   981  		if sb := c.sandboxes[containerID]; sb != nil && !sb.isStub {
   982  			return sb, nil
   983  		}
   984  	} else {
   985  		for _, sb := range c.sandboxes {
   986  			if sb.containerID == containerID && !sb.isStub {
   987  				return sb, nil
   988  			}
   989  		}
   990  	}
   991  
   992  	return nil, types.NotFoundErrorf("network sandbox for container %s not found", containerID)
   993  }
   994  
   995  // SandboxByID returns the Sandbox which has the passed id.
   996  // If not found, a [types.NotFoundError] is returned.
   997  func (c *Controller) SandboxByID(id string) (*Sandbox, error) {
   998  	if id == "" {
   999  		return nil, ErrInvalidID(id)
  1000  	}
  1001  	c.mu.Lock()
  1002  	s, ok := c.sandboxes[id]
  1003  	c.mu.Unlock()
  1004  	if !ok {
  1005  		return nil, types.NotFoundErrorf("sandbox %s not found", id)
  1006  	}
  1007  	return s, nil
  1008  }
  1009  
  1010  // SandboxDestroy destroys a sandbox given a container ID.
  1011  func (c *Controller) SandboxDestroy(id string) error {
  1012  	var sb *Sandbox
  1013  	c.mu.Lock()
  1014  	for _, s := range c.sandboxes {
  1015  		if s.containerID == id {
  1016  			sb = s
  1017  			break
  1018  		}
  1019  	}
  1020  	c.mu.Unlock()
  1021  
  1022  	// It is not an error if sandbox is not available
  1023  	if sb == nil {
  1024  		return nil
  1025  	}
  1026  
  1027  	return sb.Delete()
  1028  }
  1029  
  1030  func (c *Controller) loadDriver(networkType string) error {
  1031  	var err error
  1032  
  1033  	if pg := c.GetPluginGetter(); pg != nil {
  1034  		_, err = pg.Get(networkType, driverapi.NetworkPluginEndpointType, plugingetter.Lookup)
  1035  	} else {
  1036  		_, err = plugins.Get(networkType, driverapi.NetworkPluginEndpointType)
  1037  	}
  1038  
  1039  	if err != nil {
  1040  		if errors.Cause(err) == plugins.ErrNotFound {
  1041  			return types.NotFoundErrorf(err.Error())
  1042  		}
  1043  		return err
  1044  	}
  1045  
  1046  	return nil
  1047  }
  1048  
  1049  func (c *Controller) loadIPAMDriver(name string) error {
  1050  	var err error
  1051  
  1052  	if pg := c.GetPluginGetter(); pg != nil {
  1053  		_, err = pg.Get(name, ipamapi.PluginEndpointType, plugingetter.Lookup)
  1054  	} else {
  1055  		_, err = plugins.Get(name, ipamapi.PluginEndpointType)
  1056  	}
  1057  
  1058  	if err != nil {
  1059  		if errors.Cause(err) == plugins.ErrNotFound {
  1060  			return types.NotFoundErrorf(err.Error())
  1061  		}
  1062  		return err
  1063  	}
  1064  
  1065  	return nil
  1066  }
  1067  
  1068  func (c *Controller) getIPAMDriver(name string) (ipamapi.Ipam, *ipamapi.Capability, error) {
  1069  	id, caps := c.ipamRegistry.IPAM(name)
  1070  	if id == nil {
  1071  		// Might be a plugin name. Try loading it
  1072  		if err := c.loadIPAMDriver(name); err != nil {
  1073  			return nil, nil, err
  1074  		}
  1075  
  1076  		// Now that we resolved the plugin, try again looking up the registry
  1077  		id, caps = c.ipamRegistry.IPAM(name)
  1078  		if id == nil {
  1079  			return nil, nil, types.InvalidParameterErrorf("invalid ipam driver: %q", name)
  1080  		}
  1081  	}
  1082  
  1083  	return id, caps, nil
  1084  }
  1085  
  1086  // Stop stops the network controller.
  1087  func (c *Controller) Stop() {
  1088  	c.closeStores()
  1089  	c.stopExternalKeyListener()
  1090  	osl.GC()
  1091  }
  1092  
  1093  // StartDiagnostic starts the network diagnostic server listening on port.
  1094  func (c *Controller) StartDiagnostic(port int) {
  1095  	c.mu.Lock()
  1096  	if !c.DiagnosticServer.IsDiagnosticEnabled() {
  1097  		c.DiagnosticServer.EnableDiagnostic("127.0.0.1", port)
  1098  	}
  1099  	c.mu.Unlock()
  1100  }
  1101  
  1102  // StopDiagnostic stops the network diagnostic server.
  1103  func (c *Controller) StopDiagnostic() {
  1104  	c.mu.Lock()
  1105  	if c.DiagnosticServer.IsDiagnosticEnabled() {
  1106  		c.DiagnosticServer.DisableDiagnostic()
  1107  	}
  1108  	c.mu.Unlock()
  1109  }
  1110  
  1111  // IsDiagnosticEnabled returns true if the diagnostic server is running.
  1112  func (c *Controller) IsDiagnosticEnabled() bool {
  1113  	c.mu.Lock()
  1114  	defer c.mu.Unlock()
  1115  	return c.DiagnosticServer.IsDiagnosticEnabled()
  1116  }