add doh support

.gitignore

@@ -3,3 +3,4 @@
 .idea/**
 .gocache/**
 tslink
+config.toml

USAGE.md

@@ -14,6 +14,14 @@
 | `ephemeral` | bool | 否 | `true` | 节点是否临时节点，离开 Tailnet 后自动删除 |
 | `accept_routes` | bool | 否 | `true` | 是否接受其他节点发布的子网路由 |
 
+### `[dns]` — DNS 解析配置
+
+| 字段 | 类型 | 必填 | 默认值 | 说明 |
+|------|------|------|--------|------|
+| `doh_servers` | []string | 否 | `[]` | DNS-over-HTTPS（RFC 8484）回落解析器地址列表，须为 `http(s)://` URL |
+
+`dst_addr` 中的域名默认走 Tailnet 自身的 DNS 解析器（支持 MagicDNS 与 split-DNS）。当该解析器无法解析目标（例如宿主机本身没有可用的系统 DNS，或目标不在 Tailnet 的 split-DNS 路由内）时，会**依次**尝试 `doh_servers` 中配置的 DoH 端点解析公网域名。留空则关闭此回落。仅对原始域名发起 DoH 查询（Tailnet 内部 MagicDNS 名称无法通过 DoH 解析）。
+
 ### `[[forward.<name>]]` — 转发规则（Tailscale → 本地）
 
 将 Tailscale 上的流量转发到本地服务。`<name>` 为自定义标签名。
@@ -89,6 +97,10 @@ hostname = ""                           # 可选，留空使用本机主机名
 ephemeral = true                        # 可选，临时节点
 accept_routes = true                    # 可选，接受子网路由
 
+[dns]
+# 可选，Tailnet DNS 无法解析时回落到 DoH 解析公网域名；留空关闭
+doh_servers = ["https://cloudflare-dns.com/dns-query", "https://dns.google/dns-query"]
+
 # 示例1: 将 Tailnet 上 8080 端口的请求转发到本地 9090
 [[forward.web]]
 protocol = "tcp"

config.example.toml

@@ -5,6 +5,13 @@ hostname = ""  # leave blank to use machine name
 ephemeral = true
 accept_routes = true
 
+[dns]
+# Fallback DNS-over-HTTPS resolvers (RFC 8484), tried only when the tailnet
+# resolver can't resolve a destination (e.g. the host has no working system DNS,
+# or the name is outside the tailnet's split-DNS routes). Leave empty to disable.
+doh_servers = []
+# doh_servers = ["https://cloudflare-dns.com/dns-query", "https://dns.google/dns-query"]
+
 [[forward.web]]  # you -> others
 protocol = "tcp"
 tailscale_port = 8080

core/config.go

@@ -7,6 +7,7 @@ import (
 	"io"
 	"net"
 	"net/http"
+	"net/url"
 	"os"
 	"strings"
 	"time"
@@ -41,6 +42,13 @@ type Core struct {
 	AcceptRoutes bool   `toml:"accept_routes"`
 }
 
+// DNS holds resolver options. DoHServers are DNS-over-HTTPS endpoints (RFC 8484)
+// queried as a fallback when the tailnet resolver cannot resolve a dial
+// destination. An empty list disables the fallback.
+type DNS struct {
+	DoHServers []string `toml:"doh_servers"`
+}
+
 func (r ConnectRule) LANEnabled() bool {
 	if r.LanEnable != nil {
 		return *r.LanEnable
@@ -67,6 +75,7 @@ func (r ConnectRule) BindIP() string {
 
 type Config struct {
 	Core    Core                     `toml:"core"`
+	DNS     DNS                      `toml:"dns"`
 	Forward map[string][]ForwardRule `toml:"forward"`
 	Connect map[string][]ConnectRule `toml:"connect"`
 }
@@ -94,6 +103,13 @@ func (cfg *Config) Validate() error {
 		errs = append(errs, errors.New("core.auth_key is required"))
 	}
 
+	for i, server := range cfg.DNS.DoHServers {
+		u, err := url.Parse(strings.TrimSpace(server))
+		if err != nil || (u.Scheme != "http" && u.Scheme != "https") || u.Host == "" {
+			errs = append(errs, fmt.Errorf("dns.doh_servers[%d] must be a valid http(s) URL", i))
+		}
+	}
+
 	usedForwardListeners := make(map[string]string)
 	usedConnectListeners := make(map[string]string)
 

core/config_test.go

@@ -51,6 +51,7 @@ func TestConfigValidateAcceptsValidConfig(t *testing.T) {
 
 	cfg := Config{
 		Core: Core{AuthKey: "tskey-auth-example"},
+		DNS:  DNS{DoHServers: []string{"https://cloudflare-dns.com/dns-query"}},
 		Forward: map[string][]ForwardRule{
 			"web": {
 				{Protocol: "tcp", TailscalePort: 8080, LocalAddr: "127.0.0.1:9090"},
@@ -106,6 +107,29 @@ func TestConfigValidateRejectsInvalidConfig(t *testing.T) {
 	}
 }
 
+func TestConfigValidateRejectsInvalidDoHServer(t *testing.T) {
+	t.Parallel()
+
+	cfg := Config{
+		Core: Core{AuthKey: "tskey-auth-example"},
+		DNS:  DNS{DoHServers: []string{"https://ok.example/dns-query", "not a url", "ftp://wrong.example"}},
+	}
+
+	err := cfg.Validate()
+	if err == nil {
+		t.Fatal("Validate() returned nil, want error")
+	}
+
+	for _, want := range []string{
+		"dns.doh_servers[1] must be a valid http(s) URL",
+		"dns.doh_servers[2] must be a valid http(s) URL",
+	} {
+		if !strings.Contains(err.Error(), want) {
+			t.Fatalf("Validate() error %q does not contain %q", err.Error(), want)
+		}
+	}
+}
+
 func boolPtr(v bool) *bool {
 	return &v
 }

core/dns.go

@@ -117,11 +117,6 @@ func resolveDialAddr(ctx context.Context, srv *tsnet.Server, addr string) (strin
 		return addr, nil // already ip:port, nothing to resolve
 	}
 
-	dnsMgr, ok := srv.Sys().DNSManager.GetOK()
-	if !ok {
-		return addr, errors.New("DNS manager not available")
-	}
-
 	// Names to try, in order. A bare single-label name additionally gets the
 	// MagicDNS suffix appended so short tailnet hostnames still resolve; a name
 	// that already contains a dot (an FQDN, including split-DNS suffixes) is
@@ -132,6 +127,7 @@ func resolveDialAddr(ctx context.Context, srv *tsnet.Server, addr string) (strin
 	}
 
 	var lastErr error
+	if dnsMgr, ok := srv.Sys().DNSManager.GetOK(); ok {
 		for _, name := range candidates {
 			ip, err := resolveHostViaResolver(ctx, dnsMgr, name)
 			if err != nil {
@@ -140,17 +136,61 @@ func resolveDialAddr(ctx context.Context, srv *tsnet.Server, addr string) (strin
 			}
 			return net.JoinHostPort(ip.String(), port), nil
 		}
-	return addr, fmt.Errorf("resolve %q via tailnet DNS: %w", host, lastErr)
+	} else {
+		lastErr = errors.New("DNS manager not available")
 	}
 
+	// Fallback: resolve public names via DNS-over-HTTPS when the tailnet
+	// resolver couldn't (no working system DNS on the host, or a name outside
+	// the tailnet's split-DNS routes). Only the original host is queried — DoH
+	// can't resolve tailnet-internal MagicDNS names.
+	if dohEnabled() {
+		if ip, derr := resolveHostViaDoH(ctx, host); derr == nil {
+			return net.JoinHostPort(ip.String(), port), nil
+		} else {
+			lastErr = fmt.Errorf("tailnet dns: %v; doh: %w", lastErr, derr)
+		}
+	}
+
+	return addr, fmt.Errorf("resolve %q: %w", host, lastErr)
+}
+
+// dnsExchange sends a single DNS question and returns the first address answer,
+// or a CNAME target if one is present instead. It abstracts the transport so the
+// Tailscale resolver and the DNS-over-HTTPS fallback (see doh.go) can share the
+// CNAME-chasing logic in resolveHostChase.
+type dnsExchange func(ctx context.Context, name dnsmessage.Name, qType dnsmessage.Type) (netip.Addr, string, error)
+
 // resolveHostViaResolver resolves a hostname to a netip.Addr using the
 // Tailscale DNS resolver. It queries A then AAAA records and follows CNAME
 // chains (up to 8 levels deep).
 func resolveHostViaResolver(ctx context.Context, resolver *dns.Manager, host string) (netip.Addr, error) {
-	return resolveHostWithDepth(ctx, resolver, host, 0)
+	return resolveHostChase(ctx, host, 0, tailnetExchange(resolver))
 }
 
-func resolveHostWithDepth(ctx context.Context, r *dns.Manager, host string, depth int) (netip.Addr, error) {
+// tailnetExchange returns a dnsExchange backed by the Tailscale DNS resolver.
+func tailnetExchange(r *dns.Manager) dnsExchange {
+	return func(ctx context.Context, name dnsmessage.Name, qType dnsmessage.Type) (netip.Addr, string, error) {
+		queryBytes, err := buildDNSQuery(name, qType)
+		if err != nil {
+			return netip.Addr{}, "", err
+		}
+		qctx, cancel := context.WithTimeout(ctx, 5*time.Second)
+		defer cancel()
+		respBytes, err := r.Query(qctx, queryBytes, "udp", netip.AddrPort{})
+		if err != nil {
+			return netip.Addr{}, "", fmt.Errorf("DNS resolution failed for %s: %w", strings.TrimSuffix(name.String(), "."), err)
+		}
+		return parseDNSAnswer(respBytes)
+	}
+}
+
+// resolveHostChase resolves host to an address by issuing A then AAAA questions
+// through exchange and following CNAME chains (up to maxCNAMEChase levels deep).
+// A first (MagicDNS hands out an IPv4 for tailnet peers), then AAAA so IPv6-only
+// split-DNS hosts still resolve; a CNAME seen in either answer is chased once no
+// address record is found.
+func resolveHostChase(ctx context.Context, host string, depth int, exchange dnsExchange) (netip.Addr, error) {
 	const maxCNAMEChase = 8
 	if depth > maxCNAMEChase {
 		return netip.Addr{}, fmt.Errorf("CNAME chain too deep for %s", host)
@@ -161,12 +201,9 @@ func resolveHostWithDepth(ctx context.Context, r *dns.Manager, host string, dept
 		return netip.Addr{}, fmt.Errorf("invalid hostname %s: %w", host, err)
 	}
 
-	// Query A first (MagicDNS hands out an IPv4 for tailnet peers), then AAAA so
-	// IPv6-only split-DNS hosts still resolve. A CNAME seen in either answer is
-	// chased once no address record is found.
 	var cnameTarget string
 	for _, qType := range []dnsmessage.Type{dnsmessage.TypeA, dnsmessage.TypeAAAA} {
-		ip, cname, err := queryResolver(ctx, r, name, qType)
+		ip, cname, err := exchange(ctx, name, qType)
 		if err != nil {
 			return netip.Addr{}, err
 		}
@@ -178,18 +215,15 @@ func resolveHostWithDepth(ctx context.Context, r *dns.Manager, host string, dept
 		}
 	}
 
-	// Follow CNAME if no direct address record was found.
 	if cnameTarget != "" {
-		return resolveHostWithDepth(ctx, r, cnameTarget, depth+1)
+		return resolveHostChase(ctx, cnameTarget, depth+1, exchange)
 	}
 
 	return netip.Addr{}, fmt.Errorf("no A/AAAA record found for %s", host)
 }
 
-// queryResolver sends a single question of the given type to the Tailscale DNS
-// resolver and returns the first address answer, or a CNAME target if one is
-// present instead.
-func queryResolver(ctx context.Context, r *dns.Manager, name dnsmessage.Name, qType dnsmessage.Type) (netip.Addr, string, error) {
+// buildDNSQuery packs a single-question DNS query message for name/qType.
+func buildDNSQuery(name dnsmessage.Name, qType dnsmessage.Type) ([]byte, error) {
 	msg := dnsmessage.Message{
 		Header: dnsmessage.Header{RecursionDesired: true},
 		Questions: []dnsmessage.Question{
@@ -198,16 +232,14 @@ func queryResolver(ctx context.Context, r *dns.Manager, name dnsmessage.Name, qT
 	}
 	queryBytes, err := msg.Pack()
 	if err != nil {
-		return netip.Addr{}, "", fmt.Errorf("failed to pack DNS query: %w", err)
+		return nil, fmt.Errorf("failed to pack DNS query: %w", err)
 	}
-
-	qctx, cancel := context.WithTimeout(ctx, 5*time.Second)
-	defer cancel()
-	respBytes, err := r.Query(qctx, queryBytes, "udp", netip.AddrPort{})
-	if err != nil {
-		return netip.Addr{}, "", fmt.Errorf("DNS resolution failed for %s: %w", strings.TrimSuffix(name.String(), "."), err)
+	return queryBytes, nil
 }
 
+// parseDNSAnswer unpacks a DNS response and returns the first A/AAAA address, or
+// a CNAME target if one is present instead of an address record.
+func parseDNSAnswer(respBytes []byte) (netip.Addr, string, error) {
 	var resp dnsmessage.Message
 	if err := resp.Unpack(respBytes); err != nil {
 		return netip.Addr{}, "", fmt.Errorf("failed to unpack DNS response: %w", err)

core/doh.go

@@ -0,0 +1,110 @@
+package core
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/netip"
+	"sync"
+	"time"
+
+	"golang.org/x/net/dns/dnsmessage"
+)
+
+// dohContentType is the RFC 8484 media type for binary DNS messages over HTTPS.
+const dohContentType = "application/dns-message"
+
+// maxDoHResponse bounds how much of a DoH response body we read, guarding against
+// a hostile or misbehaving endpoint streaming an unbounded body.
+const maxDoHResponse = 64 << 10 // 64 KiB
+
+// Configured DNS-over-HTTPS endpoints, set once at startup from the config file
+// (see SetDoHServers). Mirrors the magicDNSSuffix package-global pattern in
+// dns.go so the resolver need not thread config through every call.
+var (
+	dohMu      sync.RWMutex
+	dohServers []string
+	dohClient  = &http.Client{Timeout: 10 * time.Second}
+)
+
+// SetDoHServers records the DNS-over-HTTPS fallback endpoints.
+func SetDoHServers(servers []string) {
+	dohMu.Lock()
+	defer dohMu.Unlock()
+	dohServers = append([]string(nil), servers...)
+}
+
+// dohEnabled reports whether any DoH fallback endpoint is configured.
+func dohEnabled() bool {
+	dohMu.RLock()
+	defer dohMu.RUnlock()
+	return len(dohServers) > 0
+}
+
+// getDoHServers returns a copy of the configured DoH endpoints.
+func getDoHServers() []string {
+	dohMu.RLock()
+	defer dohMu.RUnlock()
+	return append([]string(nil), dohServers...)
+}
+
+// resolveHostViaDoH resolves host through the configured DoH endpoints.
+func resolveHostViaDoH(ctx context.Context, host string) (netip.Addr, error) {
+	return resolveViaDoHServers(ctx, getDoHServers(), host)
+}
+
+// resolveViaDoHServers tries each server in order, returning the first address
+// that resolves. It takes the server list explicitly so it can be exercised in
+// tests without touching package globals.
+func resolveViaDoHServers(ctx context.Context, servers []string, host string) (netip.Addr, error) {
+	if len(servers) == 0 {
+		return netip.Addr{}, errors.New("no DoH servers configured")
+	}
+	var lastErr error
+	for _, server := range servers {
+		ip, err := resolveHostChase(ctx, host, 0, dohExchange(server))
+		if err != nil {
+			lastErr = fmt.Errorf("doh %s: %w", server, err)
+			continue
+		}
+		return ip, nil
+	}
+	return netip.Addr{}, lastErr
+}
+
+// dohExchange returns a dnsExchange that resolves a single question against a
+// single DoH endpoint using the RFC 8484 binary wire format over HTTPS POST.
+func dohExchange(server string) dnsExchange {
+	return func(ctx context.Context, name dnsmessage.Name, qType dnsmessage.Type) (netip.Addr, string, error) {
+		query, err := buildDNSQuery(name, qType)
+		if err != nil {
+			return netip.Addr{}, "", err
+		}
+
+		req, err := http.NewRequestWithContext(ctx, http.MethodPost, server, bytes.NewReader(query))
+		if err != nil {
+			return netip.Addr{}, "", fmt.Errorf("build DoH request: %w", err)
+		}
+		req.Header.Set("Content-Type", dohContentType)
+		req.Header.Set("Accept", dohContentType)
+
+		resp, err := dohClient.Do(req)
+		if err != nil {
+			return netip.Addr{}, "", fmt.Errorf("DoH request to %s failed: %w", server, err)
+		}
+		defer resp.Body.Close()
+
+		if resp.StatusCode != http.StatusOK {
+			return netip.Addr{}, "", fmt.Errorf("DoH request to %s: unexpected status %d", server, resp.StatusCode)
+		}
+
+		respBytes, err := io.ReadAll(io.LimitReader(resp.Body, maxDoHResponse))
+		if err != nil {
+			return netip.Addr{}, "", fmt.Errorf("read DoH response from %s: %w", server, err)
+		}
+		return parseDNSAnswer(respBytes)
+	}
+}

core/doh_test.go

@@ -0,0 +1,96 @@
+package core
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"net/netip"
+	"testing"
+
+	"golang.org/x/net/dns/dnsmessage"
+)
+
+func TestResolveViaDoHServers(t *testing.T) {
+	t.Parallel()
+
+	want := netip.AddrFrom4([4]byte{93, 184, 216, 34})
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodPost {
+			t.Errorf("method = %s, want POST", r.Method)
+		}
+		if ct := r.Header.Get("Content-Type"); ct != dohContentType {
+			t.Errorf("Content-Type = %q, want %q", ct, dohContentType)
+		}
+		w.Header().Set("Content-Type", dohContentType)
+		_, _ = w.Write(packAResponse(t, "example.com.", [4]byte{93, 184, 216, 34}))
+	}))
+	defer srv.Close()
+
+	got, err := resolveViaDoHServers(context.Background(), []string{srv.URL}, "example.com")
+	if err != nil {
+		t.Fatalf("resolveViaDoHServers() error: %v", err)
+	}
+	if got != want {
+		t.Fatalf("resolveViaDoHServers() = %v, want %v", got, want)
+	}
+}
+
+func TestResolveViaDoHServersFallsThrough(t *testing.T) {
+	t.Parallel()
+
+	// First endpoint errors; the second answers. Confirms the loop advances past
+	// a failing server instead of giving up.
+	bad := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+	}))
+	defer bad.Close()
+	good := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", dohContentType)
+		_, _ = w.Write(packAResponse(t, "example.com.", [4]byte{1, 2, 3, 4}))
+	}))
+	defer good.Close()
+
+	got, err := resolveViaDoHServers(context.Background(), []string{bad.URL, good.URL}, "example.com")
+	if err != nil {
+		t.Fatalf("resolveViaDoHServers() error: %v", err)
+	}
+	if want := netip.AddrFrom4([4]byte{1, 2, 3, 4}); got != want {
+		t.Fatalf("resolveViaDoHServers() = %v, want %v", got, want)
+	}
+}
+
+func TestResolveViaDoHServersNoServers(t *testing.T) {
+	t.Parallel()
+
+	if _, err := resolveViaDoHServers(context.Background(), nil, "example.com"); err == nil {
+		t.Fatal("resolveViaDoHServers() with no servers returned nil error, want error")
+	}
+}
+
+// packAResponse builds a minimal DNS response carrying a single A record.
+func packAResponse(t *testing.T, name string, ip [4]byte) []byte {
+	t.Helper()
+	dnsName, err := dnsmessage.NewName(name)
+	if err != nil {
+		t.Fatalf("NewName: %v", err)
+	}
+	msg := dnsmessage.Message{
+		Header: dnsmessage.Header{Response: true},
+		Answers: []dnsmessage.Resource{
+			{
+				Header: dnsmessage.ResourceHeader{
+					Name:  dnsName,
+					Type:  dnsmessage.TypeA,
+					Class: dnsmessage.ClassINET,
+				},
+				Body: &dnsmessage.AResource{A: ip},
+			},
+		},
+	}
+	b, err := msg.Pack()
+	if err != nil {
+		t.Fatalf("Pack: %v", err)
+	}
+	return b
+}

main.go

@@ -28,6 +28,11 @@ func serviceLogic(configPath string, isTsnetDebug bool, configURL string, logger
 		os.Exit(1)
 	}
 
+	core.SetDoHServers(cfg.DNS.DoHServers)
+	if len(cfg.DNS.DoHServers) > 0 {
+		logger.Info("DNS-over-HTTPS fallback enabled", "servers", cfg.DNS.DoHServers)
+	}
+
 	ctx, cancelAll := context.WithCancel(context.Background())
 	defer cancelAll()
 	logger.Info("initializing tsnet server")