feat: add host dns support for resolving member addrs

Closes #8330

Signed-off-by: Dmitriy Matrenichev <dmitry.matrenichev@siderolabs.com>

Author: Dmitriy Matrenichev

api/resource/definitions/network/network.proto

@@ -106,6 +106,7 @@ message HostDNSConfigSpec {
   bool enabled = 1;
   repeated common.NetIPPort listen_addresses = 2;
   common.NetIP service_host_dns_address = 3;
+  bool resolve_member_names = 4;
 }
 
 // HostnameSpecSpec describes node hostname.

hack/release.toml

@@ -56,6 +56,16 @@ machine:
 ```
 
 Please note that on running cluster you will have to kill CoreDNS pods for this change to apply.
+
+If you want to can also enable the resolving of member addresses through their host and node names:
+
+```yaml
+machine:
+  features:
+    hostDNS:
+      enabled: true
+      resolveMemberNames: true
+```
 """
 
     [notes.secureboot-image]

internal/app/machined/pkg/controllers/network/dns_resolve_cache.go

@@ -10,27 +10,33 @@ import (
 	"fmt"
 	"net"
 	"net/netip"
+	"slices"
+	"strings"
 	"sync"
 	"time"
 
 	"github.com/coredns/coredns/plugin/pkg/proxy"
 	"github.com/cosi-project/runtime/pkg/controller"
 	"github.com/cosi-project/runtime/pkg/safe"
 	"github.com/cosi-project/runtime/pkg/state"
+	dnssrv "github.com/miekg/dns"
 	"github.com/siderolabs/gen/optional"
 	"github.com/siderolabs/gen/pair"
 	"go.uber.org/zap"
 
 	"github.com/siderolabs/talos/internal/pkg/dns"
+	"github.com/siderolabs/talos/pkg/machinery/resources/cluster"
 	"github.com/siderolabs/talos/pkg/machinery/resources/network"
 )
 
 // DNSResolveCacheController starts dns server on both udp and tcp ports based on finalized network configuration.
 type DNSResolveCacheController struct {
+	State  state.State
 	Logger *zap.Logger
 
 	mx          sync.Mutex
 	handler     *dns.Handler
+	nodeHandler *dns.NodeHandler
 	cache       *dns.Cache
 	runners     map[runnerConfig]pair.Pair[func(), <-chan struct{}]
 	reconcile   chan struct{}
@@ -115,6 +121,8 @@ func (ctrl *DNSResolveCacheController) Run(ctx context.Context, r controller.Run
 			continue
 		}
 
+		ctrl.nodeHandler.SetEnabled(cfg.TypedSpec().ResolveMemberNames)
+
 		touchedRunners := make(map[runnerConfig]struct{}, len(ctrl.runners))
 
 		for _, addr := range cfg.TypedSpec().ListenAddresses {
@@ -191,7 +199,8 @@ func (ctrl *DNSResolveCacheController) init(ctx context.Context) {
 
 	ctrl.originalCtx = ctx
 	ctrl.handler = dns.NewHandler(ctrl.Logger)
-	ctrl.cache = dns.NewCache(ctrl.handler, ctrl.Logger)
+	ctrl.nodeHandler = dns.NewNodeHandler(ctrl.handler, &stateMapper{state: ctrl.State}, ctrl.Logger)
+	ctrl.cache = dns.NewCache(ctrl.nodeHandler, ctrl.Logger)
 	ctrl.runners = map[runnerConfig]pair.Pair[func(), <-chan struct{}]{}
 	ctrl.reconcile = make(chan struct{}, 1)
 
@@ -288,3 +297,49 @@ func newDNSRunner(cfg runnerConfig, cache *dns.Cache, logger *zap.Logger) (*dns.
 
 	return dns.NewServer(serverOpts), nil
 }
+
+type stateMapper struct {
+	state state.State
+}
+
+func (s *stateMapper) ResolveAddr(ctx context.Context, qType uint16, name string) []netip.Addr {
+	name = strings.TrimRight(name, ".")
+
+	list, err := safe.ReaderListAll[*cluster.Member](ctx, s.state)
+	if err != nil {
+		return nil
+	}
+
+	elem, ok := list.Find(func(res *cluster.Member) bool {
+		return fqdnMatch(name, res.TypedSpec().Hostname) || fqdnMatch(name, res.Metadata().ID())
+	})
+	if !ok {
+		return nil
+	}
+
+	result := slices.DeleteFunc(slices.Clone(elem.TypedSpec().Addresses), func(addr netip.Addr) bool {
+		return !((qType == dnssrv.TypeA && addr.Is4()) || (qType == dnssrv.TypeAAAA && addr.Is6()))
+	})
+
+	if len(result) == 0 {
+		return nil
+	}
+
+	return result
+}
+
+func fqdnMatch(what, where string) bool {
+	what = strings.TrimRight(what, ".")
+	where = strings.TrimRight(where, ".")
+
+	if what == where {
+		return true
+	}
+
+	first, _, found := strings.Cut(where, ".")
+	if !found {
+		return false
+	}
+
+	return what == first
+}

internal/app/machined/pkg/controllers/network/dns_resolve_cache_test.go

@@ -5,6 +5,7 @@
 package network_test
 
 import (
+	"errors"
 	"net"
 	"net/netip"
 	"sync"
@@ -23,6 +24,8 @@ import (
 
 	"github.com/siderolabs/talos/internal/app/machined/pkg/controllers/ctest"
 	netctrl "github.com/siderolabs/talos/internal/app/machined/pkg/controllers/network"
+	"github.com/siderolabs/talos/pkg/machinery/config/machine"
+	"github.com/siderolabs/talos/pkg/machinery/resources/cluster"
 	"github.com/siderolabs/talos/pkg/machinery/resources/network"
 )
 
@@ -141,6 +144,108 @@ func (suite *DNSServer) TestSetupStartStop() {
 	rtestutils.AssertLength[*network.DNSUpstream](suite.Ctx(), suite.T(), suite.State(), len(dnsSlice))
 }
 
+func (suite *DNSServer) TestResolveMembers() {
+	port := must.Value(getDynamicPort())(suite.T())
+
+	const (
+		id  = "talos-default-controlplane-1"
+		id2 = "foo.example.com."
+	)
+
+	member := cluster.NewMember(cluster.NamespaceName, id)
+	*member.TypedSpec() = cluster.MemberSpec{
+		NodeID: id,
+		Addresses: []netip.Addr{
+			netip.MustParseAddr("172.20.0.2"),
+		},
+		Hostname:        id,
+		MachineType:     machine.TypeControlPlane,
+		OperatingSystem: "Talos dev",
+		ControlPlane:    nil,
+	}
+
+	suite.Require().NoError(suite.State().Create(suite.Ctx(), member))
+
+	member = cluster.NewMember(cluster.NamespaceName, id2)
+	*member.TypedSpec() = cluster.MemberSpec{
+		NodeID: id2,
+		Addresses: []netip.Addr{
+			netip.MustParseAddr("172.20.0.3"),
+		},
+		Hostname:        id2,
+		MachineType:     machine.TypeWorker,
+		OperatingSystem: "Talos dev",
+		ControlPlane:    nil,
+	}
+
+	suite.Require().NoError(suite.State().Create(suite.Ctx(), member))
+
+	cfg := network.NewHostDNSConfig(network.HostDNSConfigID)
+	cfg.TypedSpec().Enabled = true
+	cfg.TypedSpec().ListenAddresses = makeAddrs(port)
+	cfg.TypedSpec().ResolveMemberNames = true
+	suite.Require().NoError(suite.State().Create(suite.Ctx(), cfg))
+
+	rtestutils.AssertResources(suite.Ctx(), suite.T(), suite.State(),
+		expectedDNSRunners(port),
+		func(r *network.DNSResolveCache, assert *assert.Assertions) {
+			assert.Equal("running", r.TypedSpec().Status)
+		},
+	)
+
+	suite.Require().NoError(retry.Constant(3*time.Second, retry.WithUnits(100*time.Millisecond)).Retry(func() error {
+		exchange, err := dns.Exchange(
+			&dns.Msg{
+				MsgHdr: dns.MsgHdr{Id: dns.Id(), RecursionDesired: true},
+				Question: []dns.Question{
+					{Name: dns.Fqdn(id), Qtype: dns.TypeA, Qclass: dns.ClassINET},
+				},
+			},
+			"127.0.0.53:"+port,
+		)
+		if err != nil {
+			return retry.ExpectedError(err)
+		}
+
+		if exchange.Rcode != dns.RcodeSuccess {
+			return retry.ExpectedErrorf("expected rcode %d, got %d for %q", dns.RcodeSuccess, exchange.Rcode, id)
+		}
+
+		proper := dns.Fqdn(id)
+
+		if exchange.Answer[0].Header().Name != proper {
+			return retry.ExpectedErrorf("expected answer name %q, got %q", proper, exchange.Answer[0].Header().Name)
+		}
+
+		return nil
+	}))
+
+	suite.Require().NoError(retry.Constant(3*time.Second, retry.WithUnits(100*time.Millisecond)).Retry(func() error {
+		exchange, err := dns.Exchange(
+			&dns.Msg{
+				MsgHdr: dns.MsgHdr{Id: dns.Id(), RecursionDesired: true},
+				Question: []dns.Question{
+					{Name: dns.Fqdn("foo"), Qtype: dns.TypeA, Qclass: dns.ClassINET},
+				},
+			},
+			"127.0.0.53:"+port,
+		)
+		if err != nil {
+			return retry.ExpectedError(err)
+		}
+
+		if exchange.Rcode != dns.RcodeSuccess {
+			return retry.ExpectedErrorf("expected rcode %d, got %d for %q", dns.RcodeSuccess, exchange.Rcode, id2)
+		}
+
+		if !exchange.Answer[0].(*dns.A).A.Equal(net.ParseIP("172.20.0.3")) {
+			return retry.ExpectedError(errors.New("unexpected ip"))
+		}
+
+		return nil
+	}))
+}
+
 func TestDNSServer(t *testing.T) {
 	suite.Run(t, &DNSServer{
 		DefaultSuite: ctest.DefaultSuite{
@@ -149,6 +254,7 @@ func TestDNSServer(t *testing.T) {
 				suite.Require().NoError(suite.Runtime().RegisterController(&netctrl.DNSUpstreamController{}))
 				suite.Require().NoError(suite.Runtime().RegisterController(&netctrl.DNSResolveCacheController{
 					Logger: zaptest.NewLogger(t),
+					State:  suite.State(),
 				}))
 			},
 		},

internal/app/machined/pkg/controllers/network/hostdns_config.go

@@ -98,6 +98,7 @@ func (ctrl *HostDNSConfigController) Run(ctx context.Context, r controller.Runti
 			}
 
 			res.TypedSpec().Enabled = cfgProvider.Machine().Features().HostDNS().Enabled()
+			res.TypedSpec().ResolveMemberNames = cfgProvider.Machine().Features().HostDNS().ResolveMemberNames()
 
 			if cfgProvider.Machine().Features().HostDNS().ForwardKubeDNSToHost() {
 				serviceCIDRStr := cfgProvider.Cluster().Network().ServiceCIDRs()[0]

internal/app/machined/pkg/runtime/v1alpha2/v1alpha2_controller.go

@@ -194,6 +194,7 @@ func (ctrl *Controller) Run(ctx context.Context, drainer *runtime.Drainer) error
 		&network.AddressStatusController{},
 		&network.DeviceConfigController{},
 		&network.DNSResolveCacheController{
+			State:  ctrl.v1alpha1Runtime.State().V1Alpha2().Resources(),
 			Logger: dnsCacheLogger,
 		},
 		&network.DNSUpstreamController{},