moby/libnetwork/sandbox.go
Rob Murray 6c68be24a2 Windows DNS resolver forwarding
Make the internal DNS resolver for Windows containers forward requests
to upsteam DNS servers when it cannot respond itself, rather than
returning SERVFAIL.

Windows containers are normally configured with the internal resolver
first for service discovery (container name lookup), then external
resolvers from '--dns' or the host's networking configuration.

When a tool like ping gets a SERVFAIL from the internal resolver, it
tries the other nameservers. But, nslookup does not, and with this
change it does not need to.

The internal resolver learns external server addresses from the
container's HNSEndpoint configuration, so it will use the same DNS
servers as processes in the container.

The internal resolver for Windows containers listens on the network's
gateway address, and each container may have a different set of external
DNS servers. So, the resolver uses the source address of the DNS request
to select external resolvers.

On Windows, daemon.json feature option 'windows-no-dns-proxy' can be used
to prevent the internal resolver from forwarding requests (restoring the
old behaviour).

Signed-off-by: Rob Murray <rob.murray@docker.com>
2024-04-16 18:57:28 +01:00

742 lines
18 KiB
Go

package libnetwork
import (
"context"
"encoding/json"
"fmt"
"net"
"sort"
"strings"
"sync"
"github.com/containerd/log"
"github.com/docker/docker/libnetwork/etchosts"
"github.com/docker/docker/libnetwork/osl"
"github.com/docker/docker/libnetwork/types"
"go.opentelemetry.io/otel"
"go.opentelemetry.io/otel/attribute"
"go.opentelemetry.io/otel/trace"
)
// SandboxOption is an option setter function type used to pass various options to
// NewNetContainer method. The various setter functions of type SandboxOption are
// provided by libnetwork, they look like ContainerOptionXXXX(...)
type SandboxOption func(sb *Sandbox)
func (sb *Sandbox) processOptions(options ...SandboxOption) {
for _, opt := range options {
if opt != nil {
opt(sb)
}
}
}
// Sandbox provides the control over the network container entity.
// It is a one to one mapping with the container.
type Sandbox struct {
id string
containerID string
config containerConfig
extDNS []extDNSEntry
osSbox *osl.Namespace
controller *Controller
resolver *Resolver
resolverOnce sync.Once
endpoints []*Endpoint
epPriority map[string]int
populatedEndpoints map[string]struct{}
joinLeaveDone chan struct{}
dbIndex uint64
dbExists bool
isStub bool
inDelete bool
ingress bool
ndotsSet bool
oslTypes []osl.SandboxType // slice of properties of this sandbox
loadBalancerNID string // NID that this SB is a load balancer for
mu sync.Mutex
// This mutex is used to serialize service related operation for an endpoint
// The lock is here because the endpoint is saved into the store so is not unique
service sync.Mutex
}
// These are the container configs used to customize container /etc/hosts file.
type hostsPathConfig struct {
hostName string
domainName string
hostsPath string
originHostsPath string
extraHosts []extraHost
parentUpdates []parentUpdate
}
type parentUpdate struct {
cid string
name string
ip string
}
type extraHost struct {
name string
IP string
}
// These are the container configs used to customize container /etc/resolv.conf file.
type resolvConfPathConfig struct {
resolvConfPath string
originResolvConfPath string
resolvConfHashFile string
dnsList []string
dnsSearchList []string
dnsOptionsList []string
}
type containerConfig struct {
containerConfigOS //nolint:nolintlint,unused // only populated on windows
hostsPathConfig
resolvConfPathConfig
generic map[string]interface{}
useDefaultSandBox bool
useExternalKey bool
exposedPorts []types.TransportPort
}
// ID returns the ID of the sandbox.
func (sb *Sandbox) ID() string {
return sb.id
}
// ContainerID returns the container id associated to this sandbox.
func (sb *Sandbox) ContainerID() string {
return sb.containerID
}
// Key returns the sandbox's key.
func (sb *Sandbox) Key() string {
if sb.config.useDefaultSandBox {
return osl.GenerateKey("default")
}
return osl.GenerateKey(sb.id)
}
// Labels returns the sandbox's labels.
func (sb *Sandbox) Labels() map[string]interface{} {
sb.mu.Lock()
defer sb.mu.Unlock()
opts := make(map[string]interface{}, len(sb.config.generic))
for k, v := range sb.config.generic {
opts[k] = v
}
return opts
}
// Delete destroys this container after detaching it from all connected endpoints.
func (sb *Sandbox) Delete() error {
return sb.delete(false)
}
func (sb *Sandbox) delete(force bool) error {
sb.mu.Lock()
if sb.inDelete {
sb.mu.Unlock()
return types.ForbiddenErrorf("another sandbox delete in progress")
}
// Set the inDelete flag. This will ensure that we don't
// update the store until we have completed all the endpoint
// leaves and deletes. And when endpoint leaves and deletes
// are completed then we can finally delete the sandbox object
// altogether from the data store. If the daemon exits
// ungracefully in the middle of a sandbox delete this way we
// will have all the references to the endpoints in the
// sandbox so that we can clean them up when we restart
sb.inDelete = true
sb.mu.Unlock()
c := sb.controller
// Detach from all endpoints
retain := false
for _, ep := range sb.Endpoints() {
// gw network endpoint detach and removal are automatic
if ep.endpointInGWNetwork() && !force {
continue
}
// Retain the sanbdox if we can't obtain the network from store.
if _, err := c.getNetworkFromStore(ep.getNetwork().ID()); err != nil {
if !c.isSwarmNode() {
retain = true
}
log.G(context.TODO()).Warnf("Failed getting network for ep %s during sandbox %s delete: %v", ep.ID(), sb.ID(), err)
continue
}
if !force {
if err := ep.Leave(sb); err != nil {
log.G(context.TODO()).Warnf("Failed detaching sandbox %s from endpoint %s: %v\n", sb.ID(), ep.ID(), err)
}
}
if err := ep.Delete(force); err != nil {
log.G(context.TODO()).Warnf("Failed deleting endpoint %s: %v\n", ep.ID(), err)
}
}
if retain {
sb.mu.Lock()
sb.inDelete = false
sb.mu.Unlock()
return fmt.Errorf("could not cleanup all the endpoints in container %s / sandbox %s", sb.containerID, sb.id)
}
// Container is going away. Path cache in etchosts is most
// likely not required any more. Drop it.
etchosts.Drop(sb.config.hostsPath)
if sb.resolver != nil {
sb.resolver.Stop()
}
if sb.osSbox != nil && !sb.config.useDefaultSandBox {
if err := sb.osSbox.Destroy(); err != nil {
log.G(context.TODO()).WithError(err).Warn("error destroying network sandbox")
}
}
if err := sb.storeDelete(); err != nil {
log.G(context.TODO()).Warnf("Failed to delete sandbox %s from store: %v", sb.ID(), err)
}
c.mu.Lock()
if sb.ingress {
c.ingressSandbox = nil
}
delete(c.sandboxes, sb.ID())
c.mu.Unlock()
return nil
}
// Rename changes the name of all attached Endpoints.
func (sb *Sandbox) Rename(name string) error {
var err error
for _, ep := range sb.Endpoints() {
if ep.endpointInGWNetwork() {
continue
}
oldName := ep.Name()
lEp := ep
if err = ep.rename(name); err != nil {
break
}
defer func() {
if err != nil {
if err2 := lEp.rename(oldName); err2 != nil {
log.G(context.TODO()).WithField("old", oldName).WithField("origError", err).WithError(err2).Error("error renaming sandbox")
}
}
}()
}
return err
}
// Refresh leaves all the endpoints, resets and re-applies the options,
// re-joins all the endpoints without destroying the osl sandbox
func (sb *Sandbox) Refresh(options ...SandboxOption) error {
// Store connected endpoints
epList := sb.Endpoints()
// Detach from all endpoints
for _, ep := range epList {
if err := ep.Leave(sb); err != nil {
log.G(context.TODO()).Warnf("Failed detaching sandbox %s from endpoint %s: %v\n", sb.ID(), ep.ID(), err)
}
}
// Re-apply options
sb.config = containerConfig{}
sb.processOptions(options...)
// Setup discovery files
if err := sb.setupResolutionFiles(); err != nil {
return err
}
// Re-connect to all endpoints
for _, ep := range epList {
if err := ep.Join(sb); err != nil {
log.G(context.TODO()).Warnf("Failed attach sandbox %s to endpoint %s: %v\n", sb.ID(), ep.ID(), err)
}
}
return nil
}
func (sb *Sandbox) MarshalJSON() ([]byte, error) {
sb.mu.Lock()
defer sb.mu.Unlock()
// We are just interested in the container ID. This can be expanded to include all of containerInfo if there is a need
return json.Marshal(sb.id)
}
func (sb *Sandbox) UnmarshalJSON(b []byte) (err error) {
sb.mu.Lock()
defer sb.mu.Unlock()
var id string
if err := json.Unmarshal(b, &id); err != nil {
return err
}
sb.id = id
return nil
}
// Endpoints returns all the endpoints connected to the sandbox.
func (sb *Sandbox) Endpoints() []*Endpoint {
sb.mu.Lock()
defer sb.mu.Unlock()
eps := make([]*Endpoint, len(sb.endpoints))
copy(eps, sb.endpoints)
return eps
}
func (sb *Sandbox) addEndpoint(ep *Endpoint) {
sb.mu.Lock()
defer sb.mu.Unlock()
l := len(sb.endpoints)
i := sort.Search(l, func(j int) bool {
return ep.Less(sb.endpoints[j])
})
sb.endpoints = append(sb.endpoints, nil)
copy(sb.endpoints[i+1:], sb.endpoints[i:])
sb.endpoints[i] = ep
}
func (sb *Sandbox) removeEndpoint(ep *Endpoint) {
sb.mu.Lock()
defer sb.mu.Unlock()
sb.removeEndpointRaw(ep)
}
func (sb *Sandbox) removeEndpointRaw(ep *Endpoint) {
for i, e := range sb.endpoints {
if e == ep {
sb.endpoints = append(sb.endpoints[:i], sb.endpoints[i+1:]...)
return
}
}
}
func (sb *Sandbox) GetEndpoint(id string) *Endpoint {
sb.mu.Lock()
defer sb.mu.Unlock()
for _, ep := range sb.endpoints {
if ep.id == id {
return ep
}
}
return nil
}
func (sb *Sandbox) HandleQueryResp(name string, ip net.IP) {
for _, ep := range sb.Endpoints() {
n := ep.getNetwork()
n.HandleQueryResp(name, ip)
}
}
func (sb *Sandbox) ResolveIP(ctx context.Context, ip string) string {
var svc string
log.G(ctx).Debugf("IP To resolve %v", ip)
for _, ep := range sb.Endpoints() {
n := ep.getNetwork()
svc = n.ResolveIP(ctx, ip)
if len(svc) != 0 {
return svc
}
}
return svc
}
// ResolveService returns all the backend details about the containers or hosts
// backing a service. Its purpose is to satisfy an SRV query.
func (sb *Sandbox) ResolveService(ctx context.Context, name string) ([]*net.SRV, []net.IP) {
log.G(ctx).Debugf("Service name To resolve: %v", name)
// There are DNS implementations that allow SRV queries for names not in
// the format defined by RFC 2782. Hence specific validations checks are
// not done
if parts := strings.SplitN(name, ".", 3); len(parts) < 3 {
return nil, nil
}
for _, ep := range sb.Endpoints() {
n := ep.getNetwork()
srv, ip := n.ResolveService(ctx, name)
if len(srv) > 0 {
return srv, ip
}
}
return nil, nil
}
func (sb *Sandbox) ResolveName(ctx context.Context, name string, ipType int) ([]net.IP, bool) {
// Embedded server owns the docker network domain. Resolution should work
// for both container_name and container_name.network_name
// We allow '.' in service name and network name. For a name a.b.c.d the
// following have to tried;
// {a.b.c.d in the networks container is connected to}
// {a.b.c in network d},
// {a.b in network c.d},
// {a in network b.c.d},
log.G(ctx).Debugf("Name To resolve: %v", name)
name = strings.TrimSuffix(name, ".")
reqName := []string{name}
networkName := []string{""}
if strings.Contains(name, ".") {
var i int
dup := name
for {
if i = strings.LastIndex(dup, "."); i == -1 {
break
}
networkName = append(networkName, name[i+1:])
reqName = append(reqName, name[:i])
dup = dup[:i]
}
}
epList := sb.Endpoints()
// In swarm mode, services with exposed ports are connected to user overlay
// network, ingress network and docker_gwbridge networks. Name resolution
// should prioritize returning the VIP/IPs on user overlay network.
//
// Re-order the endpoints based on the network-type they're attached to;
//
// 1. dynamic networks (user overlay networks)
// 2. ingress network(s)
// 3. local networks ("docker_gwbridge")
if sb.controller.isSwarmNode() {
sort.Sort(ByNetworkType(epList))
}
for i := 0; i < len(reqName); i++ {
// First check for local container alias
ip, ipv6Miss := sb.resolveName(ctx, reqName[i], networkName[i], epList, true, ipType)
if ip != nil {
return ip, false
}
if ipv6Miss {
return ip, ipv6Miss
}
// Resolve the actual container name
ip, ipv6Miss = sb.resolveName(ctx, reqName[i], networkName[i], epList, false, ipType)
if ip != nil {
return ip, false
}
if ipv6Miss {
return ip, ipv6Miss
}
}
return nil, false
}
func (sb *Sandbox) resolveName(ctx context.Context, nameOrAlias string, networkName string, epList []*Endpoint, lookupAlias bool, ipType int) (_ []net.IP, ipv6Miss bool) {
ctx, span := otel.Tracer("").Start(ctx, "Sandbox.resolveName", trace.WithAttributes(
attribute.String("libnet.resolver.name-or-alias", nameOrAlias),
attribute.String("libnet.network.name", networkName),
attribute.Bool("libnet.resolver.alias-lookup", lookupAlias),
attribute.Int("libnet.resolver.ip-family", ipType)))
defer span.End()
for _, ep := range epList {
if lookupAlias && len(ep.aliases) == 0 {
continue
}
nw := ep.getNetwork()
if networkName != "" && networkName != nw.Name() {
continue
}
name := nameOrAlias
if lookupAlias {
ep.mu.Lock()
alias, ok := ep.aliases[nameOrAlias]
ep.mu.Unlock()
if !ok {
continue
}
name = alias
} else {
// If it is a regular lookup and if the requested name is an alias
// don't perform a svc lookup for this endpoint.
ep.mu.Lock()
_, ok := ep.aliases[nameOrAlias]
ep.mu.Unlock()
if ok {
continue
}
}
ip, miss := nw.ResolveName(ctx, name, ipType)
if ip != nil {
return ip, false
}
if miss {
ipv6Miss = miss
}
}
return nil, ipv6Miss
}
// hasExternalAccess returns true if any of sb's Endpoints appear to have external
// network access.
func (sb *Sandbox) hasExternalAccess() bool {
for _, ep := range sb.Endpoints() {
nw := ep.getNetwork()
if nw.Internal() || nw.Type() == "null" || nw.Type() == "host" {
continue
}
if ep.hasGatewayOrDefaultRoute() {
return true
}
}
return false
}
// EnableService makes a managed container's service available by adding the
// endpoint to the service load balancer and service discovery.
func (sb *Sandbox) EnableService() (err error) {
log.G(context.TODO()).Debugf("EnableService %s START", sb.containerID)
defer func() {
if err != nil {
if err2 := sb.DisableService(); err2 != nil {
log.G(context.TODO()).WithError(err2).WithField("origError", err).Error("Error while disabling service after original error")
}
}
}()
for _, ep := range sb.Endpoints() {
if !ep.isServiceEnabled() {
if err := ep.addServiceInfoToCluster(sb); err != nil {
return fmt.Errorf("could not update state for endpoint %s into cluster: %v", ep.Name(), err)
}
ep.enableService()
}
}
log.G(context.TODO()).Debugf("EnableService %s DONE", sb.containerID)
return nil
}
// DisableService removes a managed container's endpoints from the load balancer
// and service discovery.
func (sb *Sandbox) DisableService() (err error) {
log.G(context.TODO()).Debugf("DisableService %s START", sb.containerID)
failedEps := []string{}
defer func() {
if len(failedEps) > 0 {
err = fmt.Errorf("failed to disable service on sandbox:%s, for endpoints %s", sb.ID(), strings.Join(failedEps, ","))
}
}()
for _, ep := range sb.Endpoints() {
if ep.isServiceEnabled() {
if err := ep.deleteServiceInfoFromCluster(sb, false, "DisableService"); err != nil {
failedEps = append(failedEps, ep.Name())
log.G(context.TODO()).Warnf("failed update state for endpoint %s into cluster: %v", ep.Name(), err)
}
ep.disableService()
}
}
log.G(context.TODO()).Debugf("DisableService %s DONE", sb.containerID)
return nil
}
func (sb *Sandbox) clearNetworkResources(origEp *Endpoint) error {
ep := sb.GetEndpoint(origEp.id)
if ep == nil {
return fmt.Errorf("could not find the sandbox endpoint data for endpoint %s",
origEp.id)
}
sb.mu.Lock()
osSbox := sb.osSbox
inDelete := sb.inDelete
sb.mu.Unlock()
if osSbox != nil {
releaseOSSboxResources(osSbox, ep)
}
sb.mu.Lock()
delete(sb.populatedEndpoints, ep.ID())
if len(sb.endpoints) == 0 {
// sb.endpoints should never be empty and this is unexpected error condition
// We log an error message to note this down for debugging purposes.
log.G(context.TODO()).Errorf("No endpoints in sandbox while trying to remove endpoint %s", ep.Name())
sb.mu.Unlock()
return nil
}
var (
gwepBefore, gwepAfter *Endpoint
index = -1
)
for i, e := range sb.endpoints {
if e == ep {
index = i
}
if len(e.Gateway()) > 0 && gwepBefore == nil {
gwepBefore = e
}
if index != -1 && gwepBefore != nil {
break
}
}
if index == -1 {
log.G(context.TODO()).Warnf("Endpoint %s has already been deleted", ep.Name())
sb.mu.Unlock()
return nil
}
sb.removeEndpointRaw(ep)
for _, e := range sb.endpoints {
if len(e.Gateway()) > 0 {
gwepAfter = e
break
}
}
delete(sb.epPriority, ep.ID())
sb.mu.Unlock()
if gwepAfter != nil && gwepBefore != gwepAfter {
if err := sb.updateGateway(gwepAfter); err != nil {
return err
}
}
// Only update the store if we did not come here as part of
// sandbox delete. If we came here as part of delete then do
// not bother updating the store. The sandbox object will be
// deleted anyway
if !inDelete {
return sb.storeUpdate()
}
return nil
}
// joinLeaveStart waits to ensure there are no joins or leaves in progress and
// marks this join/leave in progress without race
func (sb *Sandbox) joinLeaveStart() {
sb.mu.Lock()
defer sb.mu.Unlock()
for sb.joinLeaveDone != nil {
joinLeaveDone := sb.joinLeaveDone
sb.mu.Unlock()
<-joinLeaveDone
sb.mu.Lock()
}
sb.joinLeaveDone = make(chan struct{})
}
// joinLeaveEnd marks the end of this join/leave operation and
// signals the same without race to other join and leave waiters
func (sb *Sandbox) joinLeaveEnd() {
sb.mu.Lock()
defer sb.mu.Unlock()
if sb.joinLeaveDone != nil {
close(sb.joinLeaveDone)
sb.joinLeaveDone = nil
}
}
// <=> Returns true if a < b, false if a > b and advances to next level if a == b
// epi.prio <=> epj.prio # 2 < 1
// epi.gw <=> epj.gw # non-gw < gw
// epi.internal <=> epj.internal # non-internal < internal
// epi.joininfo <=> epj.joininfo # ipv6 < ipv4
// epi.name <=> epj.name # bar < foo
func (epi *Endpoint) Less(epj *Endpoint) bool {
var prioi, prioj int
sbi, _ := epi.getSandbox()
sbj, _ := epj.getSandbox()
// Prio defaults to 0
if sbi != nil {
prioi = sbi.epPriority[epi.ID()]
}
if sbj != nil {
prioj = sbj.epPriority[epj.ID()]
}
if prioi != prioj {
return prioi > prioj
}
gwi := epi.endpointInGWNetwork()
gwj := epj.endpointInGWNetwork()
if gwi != gwj {
return gwj
}
inti := epi.getNetwork().Internal()
intj := epj.getNetwork().Internal()
if inti != intj {
return intj
}
jii := 0
if epi.joinInfo != nil {
if epi.joinInfo.gw != nil {
jii = jii + 1
}
if epi.joinInfo.gw6 != nil {
jii = jii + 2
}
}
jij := 0
if epj.joinInfo != nil {
if epj.joinInfo.gw != nil {
jij = jij + 1
}
if epj.joinInfo.gw6 != nil {
jij = jij + 2
}
}
if jii != jij {
return jii > jij
}
return epi.network.Name() < epj.network.Name()
}
func (sb *Sandbox) NdotsSet() bool {
return sb.ndotsSet
}