[proxy] Wildcard certificate support (#5583)

2026-04-18 08:16:39 +00:00 · 2026-03-12 16:00:28 +01:00
parent 8f389fef19
commit c545689448
5 changed files with 455 additions and 22 deletions
--- a/proxy/internal/acme/manager.go
+++ b/proxy/internal/acme/manager.go
@@ -11,6 +11,8 @@ import (
 	"fmt"
 	"math/rand/v2"
 	"net"
+	"os"
+	"path/filepath"
 	"slices"
 	"strings"
 	"sync"
@@ -20,6 +22,7 @@ import (
 	"golang.org/x/crypto/acme"
 	"golang.org/x/crypto/acme/autocert"

+	"github.com/netbirdio/netbird/proxy/internal/certwatch"
 	"github.com/netbirdio/netbird/shared/management/domain"
 )

@@ -49,6 +52,34 @@ type metricsRecorder interface {
 	RecordCertificateIssuance(duration time.Duration)
 }

+// wildcardEntry maps a domain suffix (e.g. ".example.com") to a certwatch
+// watcher that hot-reloads the corresponding wildcard certificate from disk.
+type wildcardEntry struct {
+	suffix  string // e.g. ".example.com"
+	pattern string // e.g. "*.example.com"
+	watcher *certwatch.Watcher
+}
+
+// ManagerConfig holds the configuration values for the ACME certificate manager.
+type ManagerConfig struct {
+	// CertDir is the directory used for caching ACME certificates.
+	CertDir string
+	// ACMEURL is the ACME directory URL (e.g. Let's Encrypt).
+	ACMEURL string
+	// EABKID and EABHMACKey are optional External Account Binding credentials
+	// required by some CAs (e.g. ZeroSSL). EABHMACKey is the base64
+	// URL-encoded string provided by the CA.
+	EABKID     string
+	EABHMACKey string
+	// LockMethod controls the cross-replica coordination strategy.
+	LockMethod CertLockMethod
+	// WildcardDir is an optional path to a directory containing wildcard
+	// certificate pairs (<name>.crt / <name>.key). Wildcard patterns are
+	// extracted from the certificates' SAN lists. Domains matching a
+	// wildcard are served from disk; all others go through ACME.
+	WildcardDir string
+}
+
 // Manager wraps autocert.Manager with domain tracking and cross-replica
 // coordination via a pluggable locking strategy. The locker prevents
 // duplicate ACME requests when multiple replicas share a certificate cache.
@@ -60,54 +91,182 @@ type Manager struct {
 	mu      sync.RWMutex
 	domains map[domain.Domain]*domainInfo

+	// wildcards holds all loaded wildcard certificates, keyed by suffix.
+	wildcards []wildcardEntry
+
 	certNotifier certificateNotifier
 	logger       *log.Logger
 	metrics      metricsRecorder
 }

-// NewManager creates a new ACME certificate manager. The certDir is used
-// for caching certificates. The lockMethod controls cross-replica
-// coordination strategy (see CertLockMethod constants).
-// eabKID and eabHMACKey are optional External Account Binding credentials
-// required for some CAs like ZeroSSL. The eabHMACKey should be the base64
-// URL-encoded string provided by the CA.
-func NewManager(certDir, acmeURL, eabKID, eabHMACKey string, notifier certificateNotifier, logger *log.Logger, lockMethod CertLockMethod, metrics metricsRecorder) *Manager {
+// NewManager creates a new ACME certificate manager.
+func NewManager(cfg ManagerConfig, notifier certificateNotifier, logger *log.Logger, metrics metricsRecorder) (*Manager, error) {
 	if logger == nil {
 		logger = log.StandardLogger()
 	}
 	mgr := &Manager{
-		certDir:      certDir,
-		locker:       newCertLocker(lockMethod, certDir, logger),
+		certDir:      cfg.CertDir,
+		locker:       newCertLocker(cfg.LockMethod, cfg.CertDir, logger),
 		domains:      make(map[domain.Domain]*domainInfo),
 		certNotifier: notifier,
 		logger:       logger,
 		metrics:      metrics,
 	}

+	if cfg.WildcardDir != "" {
+		entries, err := loadWildcardDir(cfg.WildcardDir, logger)
+		if err != nil {
+			return nil, fmt.Errorf("load wildcard certificates from %q: %w", cfg.WildcardDir, err)
+		}
+		mgr.wildcards = entries
+	}
+
 	var eab *acme.ExternalAccountBinding
-	if eabKID != "" && eabHMACKey != "" {
-		decodedKey, err := base64.RawURLEncoding.DecodeString(eabHMACKey)
+	if cfg.EABKID != "" && cfg.EABHMACKey != "" {
+		decodedKey, err := base64.RawURLEncoding.DecodeString(cfg.EABHMACKey)
 		if err != nil {
 			logger.Errorf("failed to decode EAB HMAC key: %v", err)
 		} else {
 			eab = &acme.ExternalAccountBinding{
-				KID: eabKID,
+				KID: cfg.EABKID,
 				Key: decodedKey,
 			}
-			logger.Infof("configured External Account Binding with KID: %s", eabKID)
+			logger.Infof("configured External Account Binding with KID: %s", cfg.EABKID)
 		}
 	}

 	mgr.Manager = &autocert.Manager{
 		Prompt:                 autocert.AcceptTOS,
 		HostPolicy:             mgr.hostPolicy,
-		Cache:                  autocert.DirCache(certDir),
+		Cache:                  autocert.DirCache(cfg.CertDir),
 		ExternalAccountBinding: eab,
 		Client: &acme.Client{
-			DirectoryURL: acmeURL,
+			DirectoryURL: cfg.ACMEURL,
 		},
 	}
-	return mgr
+	return mgr, nil
+}
+
+// WatchWildcards starts watching all wildcard certificate files for changes.
+// It blocks until ctx is cancelled. It is a no-op if no wildcards are loaded.
+func (mgr *Manager) WatchWildcards(ctx context.Context) {
+	if len(mgr.wildcards) == 0 {
+		return
+	}
+	seen := make(map[*certwatch.Watcher]struct{})
+	var wg sync.WaitGroup
+	for i := range mgr.wildcards {
+		w := mgr.wildcards[i].watcher
+		if _, ok := seen[w]; ok {
+			continue
+		}
+		seen[w] = struct{}{}
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			w.Watch(ctx)
+		}()
+	}
+	wg.Wait()
+}
+
+// loadWildcardDir scans dir for .crt files, pairs each with a matching .key
+// file, loads them, and extracts wildcard SANs (*.example.com) to build
+// the suffix lookup entries.
+func loadWildcardDir(dir string, logger *log.Logger) ([]wildcardEntry, error) {
+	crtFiles, err := filepath.Glob(filepath.Join(dir, "*.crt"))
+	if err != nil {
+		return nil, fmt.Errorf("glob certificate files: %w", err)
+	}
+
+	if len(crtFiles) == 0 {
+		return nil, fmt.Errorf("no .crt files found in %s", dir)
+	}
+
+	var entries []wildcardEntry
+
+	for _, crtPath := range crtFiles {
+		base := strings.TrimSuffix(filepath.Base(crtPath), ".crt")
+		keyPath := filepath.Join(dir, base+".key")
+		if _, err := os.Stat(keyPath); err != nil {
+			logger.Warnf("skipping %s: no matching key file %s", crtPath, keyPath)
+			continue
+		}
+
+		watcher, err := certwatch.NewWatcher(crtPath, keyPath, logger)
+		if err != nil {
+			logger.Warnf("skipping %s: %v", crtPath, err)
+			continue
+		}
+
+		leaf := watcher.Leaf()
+		if leaf == nil {
+			logger.Warnf("skipping %s: no parsed leaf certificate", crtPath)
+			continue
+		}
+
+		for _, san := range leaf.DNSNames {
+			suffix, ok := parseWildcard(san)
+			if !ok {
+				continue
+			}
+			entries = append(entries, wildcardEntry{
+				suffix:  suffix,
+				pattern: san,
+				watcher: watcher,
+			})
+			logger.Infof("wildcard certificate loaded: %s (from %s)", san, filepath.Base(crtPath))
+		}
+	}
+
+	if len(entries) == 0 {
+		return nil, fmt.Errorf("no wildcard SANs (*.example.com) found in certificates in %s", dir)
+	}
+
+	return entries, nil
+}
+
+// parseWildcard validates a wildcard domain pattern like "*.example.com"
+// and returns the suffix ".example.com" for matching.
+func parseWildcard(pattern string) (suffix string, ok bool) {
+	if !strings.HasPrefix(pattern, "*.") {
+		return "", false
+	}
+	parent := pattern[1:] // ".example.com"
+	if strings.Count(parent, ".") < 1 {
+		return "", false
+	}
+	return strings.ToLower(parent), true
+}
+
+// findWildcardEntry returns the wildcard entry that covers host, or nil.
+func (mgr *Manager) findWildcardEntry(host string) *wildcardEntry {
+	if len(mgr.wildcards) == 0 {
+		return nil
+	}
+	host = strings.ToLower(host)
+	for i := range mgr.wildcards {
+		e := &mgr.wildcards[i]
+		if !strings.HasSuffix(host, e.suffix) {
+			continue
+		}
+		// Single-level match: prefix before suffix must have no dots.
+		prefix := strings.TrimSuffix(host, e.suffix)
+		if len(prefix) > 0 && !strings.Contains(prefix, ".") {
+			return e
+		}
+	}
+	return nil
+}
+
+// WildcardPatterns returns the wildcard patterns that are currently loaded.
+func (mgr *Manager) WildcardPatterns() []string {
+	patterns := make([]string, len(mgr.wildcards))
+	for i, e := range mgr.wildcards {
+		patterns[i] = e.pattern
+	}
+	slices.Sort(patterns)
+	return patterns
 }

 func (mgr *Manager) hostPolicy(_ context.Context, host string) error {
@@ -123,8 +282,39 @@ func (mgr *Manager) hostPolicy(_ context.Context, host string) error {
 	return nil
 }

-// AddDomain registers a domain for ACME certificate prefetching.
-func (mgr *Manager) AddDomain(d domain.Domain, accountID, serviceID string) {
+// GetCertificate returns the TLS certificate for the given ClientHello.
+// If the requested domain matches a loaded wildcard, the static wildcard
+// certificate is returned. Otherwise, the ACME autocert manager handles
+// the request.
+func (mgr *Manager) GetCertificate(hello *tls.ClientHelloInfo) (*tls.Certificate, error) {
+	if e := mgr.findWildcardEntry(hello.ServerName); e != nil {
+		return e.watcher.GetCertificate(hello)
+	}
+	return mgr.Manager.GetCertificate(hello)
+}
+
+// AddDomain registers a domain for certificate management. Domains that
+// match a loaded wildcard are marked ready immediately (they use the
+// static wildcard certificate) and the method returns true. All other
+// domains go through ACME prefetch and the method returns false.
+//
+// When AddDomain returns true the caller is responsible for sending any
+// certificate-ready notifications after the surrounding operation (e.g.
+// mapping update) has committed successfully.
+func (mgr *Manager) AddDomain(d domain.Domain, accountID, serviceID string) (wildcardHit bool) {
+	name := d.PunycodeString()
+	if e := mgr.findWildcardEntry(name); e != nil {
+		mgr.mu.Lock()
+		mgr.domains[d] = &domainInfo{
+			accountID: accountID,
+			serviceID: serviceID,
+			state:     domainReady,
+		}
+		mgr.mu.Unlock()
+		mgr.logger.Debugf("domain %q matches wildcard %q, using static certificate", name, e.pattern)
+		return true
+	}
+
 	mgr.mu.Lock()
 	mgr.domains[d] = &domainInfo{
 		accountID: accountID,
@@ -134,6 +324,7 @@ func (mgr *Manager) AddDomain(d domain.Domain, accountID, serviceID string) {
 	mgr.mu.Unlock()

 	go mgr.prefetchCertificate(d)
+	return false
 }

 // prefetchCertificate proactively triggers certificate generation for a domain.
--- a/proxy/internal/acme/manager_test.go
+++ b/proxy/internal/acme/manager_test.go
@@ -2,6 +2,16 @@ package acme

 import (
 	"context"
+	"crypto/ecdsa"
+	"crypto/elliptic"
+	"crypto/rand"
+	"crypto/tls"
+	"crypto/x509"
+	"crypto/x509/pkix"
+	"encoding/pem"
+	"math/big"
+	"os"
+	"path/filepath"
 	"testing"
 	"time"

@@ -10,7 +20,8 @@ import (
 )

 func TestHostPolicy(t *testing.T) {
-	mgr := NewManager(t.TempDir(), "https://acme.example.com/directory", "", "", nil, nil, "", nil)
+	mgr, err := NewManager(ManagerConfig{CertDir: t.TempDir(), ACMEURL: "https://acme.example.com/directory"}, nil, nil, nil)
+	require.NoError(t, err)
 	mgr.AddDomain("example.com", "acc1", "rp1")

 	// Wait for the background prefetch goroutine to finish so the temp dir
@@ -70,7 +81,8 @@ func TestHostPolicy(t *testing.T) {
 }

 func TestDomainStates(t *testing.T) {
-	mgr := NewManager(t.TempDir(), "https://acme.example.com/directory", "", "", nil, nil, "", nil)
+	mgr, err := NewManager(ManagerConfig{CertDir: t.TempDir(), ACMEURL: "https://acme.example.com/directory"}, nil, nil, nil)
+	require.NoError(t, err)

 	assert.Equal(t, 0, mgr.PendingCerts(), "initially zero")
 	assert.Equal(t, 0, mgr.TotalDomains(), "initially zero domains")
@@ -100,3 +112,193 @@ func TestDomainStates(t *testing.T) {
 	assert.Contains(t, failed, "b.example.com")
 	assert.Empty(t, mgr.ReadyDomains())
 }
+
+func TestParseWildcard(t *testing.T) {
+	tests := []struct {
+		pattern    string
+		wantSuffix string
+		wantOK     bool
+	}{
+		{"*.example.com", ".example.com", true},
+		{"*.foo.example.com", ".foo.example.com", true},
+		{"*.COM", ".com", true},       // single-label TLD
+		{"example.com", "", false},    // no wildcard prefix
+		{"*example.com", "", false},   // missing dot
+		{"**.example.com", "", false}, // double star
+		{"", "", false},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.pattern, func(t *testing.T) {
+			suffix, ok := parseWildcard(tc.pattern)
+			assert.Equal(t, tc.wantOK, ok)
+			if ok {
+				assert.Equal(t, tc.wantSuffix, suffix)
+			}
+		})
+	}
+}
+
+func TestMatchesWildcard(t *testing.T) {
+	wcDir := t.TempDir()
+	generateSelfSignedCert(t, wcDir, "example", "*.example.com")
+
+	acmeDir := t.TempDir()
+	mgr, err := NewManager(ManagerConfig{CertDir: acmeDir, ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.NoError(t, err)
+
+	tests := []struct {
+		host  string
+		match bool
+	}{
+		{"foo.example.com", true},
+		{"bar.example.com", true},
+		{"FOO.Example.COM", true},      // case insensitive
+		{"example.com", false},         // bare parent
+		{"sub.foo.example.com", false}, // multi-level
+		{"notexample.com", false},
+		{"", false},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.host, func(t *testing.T) {
+			assert.Equal(t, tc.match, mgr.findWildcardEntry(tc.host) != nil)
+		})
+	}
+}
+
+// generateSelfSignedCert creates a temporary self-signed certificate and key
+// for testing purposes. The baseName controls the output filenames:
+// <baseName>.crt and <baseName>.key.
+func generateSelfSignedCert(t *testing.T, dir, baseName string, dnsNames ...string) {
+	t.Helper()
+
+	key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
+	require.NoError(t, err)
+
+	template := &x509.Certificate{
+		SerialNumber: big.NewInt(1),
+		Subject:      pkix.Name{CommonName: dnsNames[0]},
+		DNSNames:     dnsNames,
+		NotBefore:    time.Now().Add(-time.Hour),
+		NotAfter:     time.Now().Add(24 * time.Hour),
+	}
+
+	certDER, err := x509.CreateCertificate(rand.Reader, template, template, &key.PublicKey, key)
+	require.NoError(t, err)
+
+	certFile, err := os.Create(filepath.Join(dir, baseName+".crt"))
+	require.NoError(t, err)
+	require.NoError(t, pem.Encode(certFile, &pem.Block{Type: "CERTIFICATE", Bytes: certDER}))
+	require.NoError(t, certFile.Close())
+
+	keyDER, err := x509.MarshalECPrivateKey(key)
+	require.NoError(t, err)
+	keyFile, err := os.Create(filepath.Join(dir, baseName+".key"))
+	require.NoError(t, err)
+	require.NoError(t, pem.Encode(keyFile, &pem.Block{Type: "EC PRIVATE KEY", Bytes: keyDER}))
+	require.NoError(t, keyFile.Close())
+}
+
+func TestWildcardAddDomainSkipsACME(t *testing.T) {
+	wcDir := t.TempDir()
+	generateSelfSignedCert(t, wcDir, "example", "*.example.com")
+
+	acmeDir := t.TempDir()
+	mgr, err := NewManager(ManagerConfig{CertDir: acmeDir, ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.NoError(t, err)
+
+	// Add a wildcard-matching domain — should be immediately ready.
+	mgr.AddDomain("foo.example.com", "acc1", "svc1")
+	assert.Equal(t, 0, mgr.PendingCerts(), "wildcard domain should not be pending")
+	assert.Equal(t, []string{"foo.example.com"}, mgr.ReadyDomains())
+
+	// Add a non-wildcard domain — should go through ACME (pending then failed).
+	mgr.AddDomain("other.net", "acc2", "svc2")
+	assert.Equal(t, 2, mgr.TotalDomains())
+
+	// Wait for the ACME prefetch to fail.
+	assert.Eventually(t, func() bool {
+		return mgr.PendingCerts() == 0
+	}, 30*time.Second, 100*time.Millisecond)
+
+	assert.Equal(t, []string{"foo.example.com"}, mgr.ReadyDomains())
+	assert.Contains(t, mgr.FailedDomains(), "other.net")
+}
+
+func TestWildcardGetCertificate(t *testing.T) {
+	wcDir := t.TempDir()
+	generateSelfSignedCert(t, wcDir, "example", "*.example.com")
+
+	acmeDir := t.TempDir()
+	mgr, err := NewManager(ManagerConfig{CertDir: acmeDir, ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.NoError(t, err)
+
+	mgr.AddDomain("foo.example.com", "acc1", "svc1")
+
+	// GetCertificate for a wildcard-matching domain should return the static cert.
+	cert, err := mgr.GetCertificate(&tls.ClientHelloInfo{ServerName: "foo.example.com"})
+	require.NoError(t, err)
+	require.NotNil(t, cert)
+	assert.Contains(t, cert.Leaf.DNSNames, "*.example.com")
+}
+
+func TestMultipleWildcards(t *testing.T) {
+	wcDir := t.TempDir()
+	generateSelfSignedCert(t, wcDir, "example", "*.example.com")
+	generateSelfSignedCert(t, wcDir, "other", "*.other.org")
+
+	acmeDir := t.TempDir()
+	mgr, err := NewManager(ManagerConfig{CertDir: acmeDir, ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.NoError(t, err)
+
+	assert.ElementsMatch(t, []string{"*.example.com", "*.other.org"}, mgr.WildcardPatterns())
+
+	// Both wildcards should resolve.
+	mgr.AddDomain("foo.example.com", "acc1", "svc1")
+	mgr.AddDomain("bar.other.org", "acc2", "svc2")
+
+	assert.Equal(t, 0, mgr.PendingCerts())
+	assert.ElementsMatch(t, []string{"foo.example.com", "bar.other.org"}, mgr.ReadyDomains())
+
+	// GetCertificate routes to the correct cert.
+	cert1, err := mgr.GetCertificate(&tls.ClientHelloInfo{ServerName: "foo.example.com"})
+	require.NoError(t, err)
+	assert.Contains(t, cert1.Leaf.DNSNames, "*.example.com")
+
+	cert2, err := mgr.GetCertificate(&tls.ClientHelloInfo{ServerName: "bar.other.org"})
+	require.NoError(t, err)
+	assert.Contains(t, cert2.Leaf.DNSNames, "*.other.org")
+
+	// Non-matching domain falls through to ACME.
+	mgr.AddDomain("custom.net", "acc3", "svc3")
+	assert.Eventually(t, func() bool {
+		return mgr.PendingCerts() == 0
+	}, 30*time.Second, 100*time.Millisecond)
+	assert.Contains(t, mgr.FailedDomains(), "custom.net")
+}
+
+func TestWildcardDirEmpty(t *testing.T) {
+	wcDir := t.TempDir()
+	// Empty directory — no .crt files.
+	_, err := NewManager(ManagerConfig{CertDir: t.TempDir(), ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "no .crt files found")
+}
+
+func TestWildcardDirNonWildcardCert(t *testing.T) {
+	wcDir := t.TempDir()
+	// Certificate without a wildcard SAN.
+	generateSelfSignedCert(t, wcDir, "plain", "plain.example.com")
+
+	_, err := NewManager(ManagerConfig{CertDir: t.TempDir(), ACMEURL: "https://acme.example.com/directory", WildcardDir: wcDir}, nil, nil, nil)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "no wildcard SANs")
+}
+
+func TestNoWildcardDir(t *testing.T) {
+	// Empty string means no wildcard dir — pure ACME mode.
+	mgr, err := NewManager(ManagerConfig{CertDir: t.TempDir(), ACMEURL: "https://acme.example.com/directory"}, nil, nil, nil)
+	require.NoError(t, err)
+	assert.Empty(t, mgr.WildcardPatterns())
+}