cmd/cloner, tailcfg: fix nil vs len 0 issues, add tests, use for Hostinfo

Also use go:generate and https://golang.org/s/generatedcode header style.

.github/workflows/cross-darwin.yml

@@ -3,7 +3,7 @@ name: Darwin-Cross
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -16,10 +16,10 @@ jobs:
 
     steps:
 
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
       id: go
 
     - name: Check out code into the Go module directory

.github/workflows/cross-freebsd.yml

@@ -3,7 +3,7 @@ name: FreeBSD-Cross
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -16,10 +16,10 @@ jobs:
 
     steps:
 
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
       id: go
 
     - name: Check out code into the Go module directory

.github/workflows/cross-openbsd.yml

@@ -3,7 +3,7 @@ name: OpenBSD-Cross
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -16,10 +16,10 @@ jobs:
 
     steps:
 
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
       id: go
 
     - name: Check out code into the Go module directory

.github/workflows/cross-windows.yml

@@ -3,7 +3,7 @@ name: Windows-Cross
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -16,10 +16,10 @@ jobs:
 
     steps:
 
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
       id: go
 
     - name: Check out code into the Go module directory

.github/workflows/license.yml

@@ -0,0 +1,40 @@
+name: license
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - '*'
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Set up Go
+      uses: actions/setup-go@v1
+      with:
+        go-version: 1.14
+
+    - name: Check out code
+      uses: actions/checkout@v1
+
+    - name: Run license checker
+      run: ./scripts/check_license_headers.sh .
+
+    - uses: k0kubun/action-slack@v2.0.0
+      with:
+        payload: |
+          {
+            "attachments": [{
+              "text": "${{ job.status }}: ${{ github.workflow }} <https://github.com/${{ github.repository }}/commit/${{ github.sha }}/checks|${{ env.COMMIT_DATE }} #${{ env.COMMIT_NUMBER_OF_DAY }}> " +
+                      "(<https://github.com/${{ github.repository }}/commit/${{ github.sha }}|" + "${{ github.sha }}".substring(0, 10) + ">) " +
+                      "of ${{ github.repository }}@" + "${{ github.ref }}".split('/').reverse()[0] + " by ${{ github.event.head_commit.committer.name }}",
+              "color": "danger"
+            }]
+          }
+      env:
+        SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
+      if: failure() && github.event_name == 'push'

.github/workflows/linux.yml

@@ -3,7 +3,7 @@ name: Linux
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -16,10 +16,10 @@ jobs:
 
     steps:
 
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
       id: go
 
     - name: Check out code into the Go module directory

.github/workflows/linux32.yml

@@ -0,0 +1,48 @@
+name: Linux 32-bit
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - '*'
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+
+    - name: Set up Go
+      uses: actions/setup-go@v1
+      with:
+        go-version: 1.14
+      id: go
+
+    - name: Check out code into the Go module directory
+      uses: actions/checkout@v1
+
+    - name: Basic build
+      run: GOARCH=386 go build ./cmd/...
+
+    - name: Run tests on linux
+      run: GOARCH=386 go test ./...
+
+    - uses: k0kubun/action-slack@v2.0.0
+      with:
+        payload: |
+          {
+            "attachments": [{
+              "text": "${{ job.status }}: ${{ github.workflow }} <https://github.com/${{ github.repository }}/commit/${{ github.sha }}/checks|${{ env.COMMIT_DATE }} #${{ env.COMMIT_NUMBER_OF_DAY }}> " +
+                      "(<https://github.com/${{ github.repository }}/commit/${{ github.sha }}|" + "${{ github.sha }}".substring(0, 10) + ">) " +
+                      "of ${{ github.repository }}@" + "${{ github.ref }}".split('/').reverse()[0] + " by ${{ github.event.head_commit.committer.name }}",
+              "color": "danger"
+            }]
+          }
+      env:
+        SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}
+      if: failure() && github.event_name == 'push'
+

.github/workflows/staticcheck.yml

@@ -3,7 +3,7 @@ name: staticcheck
 on:
   push:
     branches:
-      - master
+      - main
   pull_request:
     branches:
       - '*'
@@ -13,16 +13,22 @@ jobs:
     runs-on: ubuntu-latest
 
     steps:
-    - name: Set up Go 1.13
+    - name: Set up Go
       uses: actions/setup-go@v1
       with:
-        go-version: 1.13
+        go-version: 1.14
 
     - name: Check out code
       uses: actions/checkout@v1
 
+    - name: Run go vet
+      run: go vet ./...
+
+    - name: Print staticcheck version
+      run: go run honnef.co/go/tools/cmd/staticcheck -version
+
     - name: Run staticcheck
-      run: go run honnef.co/go/tools/cmd/staticcheck -- ./...
+      run: "go run honnef.co/go/tools/cmd/staticcheck -- $(go list ./... | grep -v tempfork)"
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

Dockerfile

@@ -2,7 +2,26 @@
 # Use of this source code is governed by a BSD-style
 # license that can be found in the LICENSE file.
 
-FROM golang:1.13-alpine AS build-env
+# This Dockerfile includes all the tailscale binaries.
+#
+# To build the Dockerfile:
+#
+#     $ docker build -t tailscale:tailscale .
+#
+# To run the tailscaled agent:
+#
+#     $ docker run -d --name=tailscaled -v /var/lib:/var/lib -v /dev/net/tun:/dev/net/tun --network=host --privileged tailscale:tailscale tailscaled
+#
+# To then log in:
+#
+#     $ docker exec tailscaled tailscale up
+#
+# To see status:
+#
+#     $ docker exec tailscaled tailscale status
+
+
+FROM golang:1.14-alpine AS build-env
 
 WORKDIR /go/src/tailscale
 
@@ -15,5 +34,5 @@ COPY . .
 RUN go install -v ./cmd/...
 
 FROM alpine:3.11
-RUN apk add --no-cache ca-certificates iptables
+RUN apk add --no-cache ca-certificates iptables iproute2
 COPY --from=build-env /go/bin/* /usr/local/bin/

Makefile

@@ -0,0 +1,7 @@
+usage:
+	echo "See Makefile"
+
+check: staticcheck
+
+staticcheck:
+	go run honnef.co/go/tools/cmd/staticcheck -- $$(go list ./... | grep -v tempfork)

README.md

@@ -12,16 +12,21 @@ It currently includes the Linux client.
 The Linux client is currently `cmd/relaynode`, but will
 soon be replaced by `cmd/tailscaled`.
 
+## Using
+
+We serve packages for a variety of distros at
+https://pkgs.tailscale.com .
+
 ## Building
 
 ```
 go install tailscale.com/cmd/tailscale{,d}
 ```
 
-We only support the latest Go release and any Go beta or release
-candidate builds (currently Go 1.13.x or Go 1.14) in module mode. It
-might work in earlier Go versions or in GOPATH mode, but we're making
-no effort to keep those working.
+We only guarantee to support the latest Go release and any Go beta or
+release candidate builds (currently Go 1.14) in module mode. It might
+work in earlier Go versions or in GOPATH mode, but we're making no
+effort to keep those working.
 
 ## Bugs
 

atomicfile/atomicfile.go

@@ -1,4 +1,4 @@
-// Copyright 2019 Tailscale & AUTHORS. All rights reserved.
+// Copyright (c) 2019 Tailscale Inc & AUTHORS All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
@@ -9,20 +9,39 @@
 package atomicfile // import "tailscale.com/atomicfile"
 
 import (
-	"fmt"
 	"io/ioutil"
 	"os"
+	"path/filepath"
+	"runtime"
 )
 
 // WriteFile writes data to filename+some suffix, then renames it
 // into filename.
-func WriteFile(filename string, data []byte, perm os.FileMode) error {
-	tmpname := filename + ".new.tmp"
-	if err := ioutil.WriteFile(tmpname, data, perm); err != nil {
-		return fmt.Errorf("%#v: %v", tmpname, err)
+func WriteFile(filename string, data []byte, perm os.FileMode) (err error) {
+	f, err := ioutil.TempFile(filepath.Dir(filename), filepath.Base(filename)+".tmp")
+	if err != nil {
+		return err
 	}
-	if err := os.Rename(tmpname, filename); err != nil {
-		return fmt.Errorf("%#v->%#v: %v", tmpname, filename, err)
+	tmpName := f.Name()
+	defer func() {
+		if err != nil {
+			f.Close()
+			os.Remove(tmpName)
+		}
+	}()
+	if _, err := f.Write(data); err != nil {
+		return err
 	}
-	return nil
+	if runtime.GOOS != "windows" {
+		if err := f.Chmod(perm); err != nil {
+			return err
+		}
+	}
+	if err := f.Sync(); err != nil {
+		return err
+	}
+	if err := f.Close(); err != nil {
+		return err
+	}
+	return os.Rename(tmpName, filename)
 }

cmd/cloner/cloner.go

@@ -0,0 +1,264 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Cloner is a tool to automate the creation of a Clone method.
+//
+// The result of the Clone method aliases no memory that can be edited
+// with the original.
+//
+// This tool makes lots of implicit assumptions about the types you feed it.
+// In particular, it can only write relatively "shallow" Clone methods.
+// That is, if a type contains another named struct type, cloner assumes that
+// named type will also have a Clone method.
+package main
+
+import (
+	"bytes"
+	"flag"
+	"fmt"
+	"go/ast"
+	"go/format"
+	"go/token"
+	"go/types"
+	"io/ioutil"
+	"log"
+	"os"
+	"strings"
+
+	"golang.org/x/tools/go/packages"
+)
+
+var (
+	flagTypes     = flag.String("type", "", "comma-separated list of types; required")
+	flagOutput    = flag.String("output", "", "output file; required")
+	flagBuildTags = flag.String("tags", "", "compiler build tags to apply")
+)
+
+func main() {
+	log.SetFlags(0)
+	log.SetPrefix("cloner: ")
+	flag.Parse()
+	if len(*flagTypes) == 0 {
+		flag.Usage()
+		os.Exit(2)
+	}
+	typeNames := strings.Split(*flagTypes, ",")
+
+	cfg := &packages.Config{
+		Mode:  packages.NeedTypes | packages.NeedTypesInfo | packages.NeedSyntax | packages.NeedName,
+		Tests: false,
+	}
+	if *flagBuildTags != "" {
+		cfg.BuildFlags = []string{"-tags=" + *flagBuildTags}
+	}
+	pkgs, err := packages.Load(cfg, ".")
+	if err != nil {
+		log.Fatal(err)
+	}
+	if len(pkgs) != 1 {
+		log.Fatalf("wrong number of packages: %d", len(pkgs))
+	}
+	pkg := pkgs[0]
+	buf := new(bytes.Buffer)
+	imports := make(map[string]struct{})
+	for _, typeName := range typeNames {
+		found := false
+		for _, file := range pkg.Syntax {
+			//var fbuf bytes.Buffer
+			//ast.Fprint(&fbuf, pkg.Fset, file, nil)
+			//fmt.Println(fbuf.String())
+
+			for _, d := range file.Decls {
+				decl, ok := d.(*ast.GenDecl)
+				if !ok || decl.Tok != token.TYPE {
+					continue
+				}
+				for _, s := range decl.Specs {
+					spec, ok := s.(*ast.TypeSpec)
+					if !ok || spec.Name.Name != typeName {
+						continue
+					}
+					typeNameObj := pkg.TypesInfo.Defs[spec.Name]
+					typ, ok := typeNameObj.Type().(*types.Named)
+					if !ok {
+						continue
+					}
+					pkg := typeNameObj.Pkg()
+					gen(buf, imports, typeName, typ, pkg)
+				}
+				found = true
+			}
+		}
+		if !found {
+			log.Fatalf("could not find type %s", typeName)
+		}
+	}
+
+	contents := new(bytes.Buffer)
+	fmt.Fprintf(contents, header, *flagTypes, pkg.Name)
+	fmt.Fprintf(contents, "import (\n")
+	for s := range imports {
+		fmt.Fprintf(contents, "\t%q\n", s)
+	}
+	fmt.Fprintf(contents, ")\n\n")
+	contents.Write(buf.Bytes())
+
+	out, err := format.Source(contents.Bytes())
+	if err != nil {
+		log.Fatalf("%s, in source:\n%s", err, contents.Bytes())
+	}
+
+	output := *flagOutput
+	if output == "" {
+		flag.Usage()
+		os.Exit(2)
+	}
+	if err := ioutil.WriteFile(output, out, 0666); err != nil {
+		log.Fatal(err)
+	}
+}
+
+const header = `// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Code generated by tailscale.com/cmd/cloner -type %s; DO NOT EDIT.
+
+package %s
+
+`
+
+func gen(buf *bytes.Buffer, imports map[string]struct{}, name string, typ *types.Named, thisPkg *types.Package) {
+	pkgQual := func(pkg *types.Package) string {
+		if thisPkg == pkg {
+			return ""
+		}
+		imports[pkg.Path()] = struct{}{}
+		return pkg.Name()
+	}
+	importedName := func(t types.Type) string {
+		return types.TypeString(t, pkgQual)
+	}
+
+	switch t := typ.Underlying().(type) {
+	case *types.Struct:
+		_ = t
+		name := typ.Obj().Name()
+		fmt.Fprintf(buf, "// Clone makes a deep copy of %s.\n", name)
+		fmt.Fprintf(buf, "// The result aliases no memory with the original.\n")
+		fmt.Fprintf(buf, "func (src *%s) Clone() *%s {\n", name, name)
+		writef := func(format string, args ...interface{}) {
+			fmt.Fprintf(buf, "\t"+format+"\n", args...)
+		}
+		writef("if src == nil {")
+		writef("\treturn nil")
+		writef("}")
+		writef("dst := new(%s)", name)
+		writef("*dst = *src")
+		for i := 0; i < t.NumFields(); i++ {
+			fname := t.Field(i).Name()
+			ft := t.Field(i).Type()
+			if !containsPointers(ft) {
+				continue
+			}
+			if named, _ := ft.(*types.Named); named != nil && !hasBasicUnderlying(ft) {
+				writef("dst.%s = *src.%s.Clone()", fname, fname)
+				continue
+			}
+			switch ft := ft.Underlying().(type) {
+			case *types.Slice:
+				if containsPointers(ft.Elem()) {
+					n := importedName(ft.Elem())
+					writef("dst.%s = make([]%s, len(src.%s))", fname, n, fname)
+					writef("for i := range dst.%s {", fname)
+					if _, isPtr := ft.Elem().(*types.Pointer); isPtr {
+						writef("\tdst.%s[i] = src.%s[i].Clone()", fname, fname)
+					} else {
+						writef("\tdst.%s[i] = *src.%s[i].Clone()", fname, fname)
+					}
+					writef("}")
+				} else {
+					writef("dst.%s = append(src.%s[:0:0], src.%s...)", fname, fname, fname)
+				}
+			case *types.Pointer:
+				if named, _ := ft.Elem().(*types.Named); named != nil && containsPointers(ft.Elem()) {
+					writef("dst.%s = src.%s.Clone()", fname, fname)
+					continue
+				}
+				n := importedName(ft.Elem())
+				writef("if dst.%s != nil {", fname)
+				writef("\tdst.%s = new(%s)", fname, n)
+				writef("\t*dst.%s = *src.%s", fname, fname)
+				if containsPointers(ft.Elem()) {
+					writef("\t" + `panic("TODO pointers in pointers")`)
+				}
+				writef("}")
+			case *types.Map:
+				writef("if dst.%s != nil {", fname)
+				writef("\tdst.%s = map[%s]%s{}", fname, importedName(ft.Key()), importedName(ft.Elem()))
+				if sliceType, isSlice := ft.Elem().(*types.Slice); isSlice {
+					n := importedName(sliceType.Elem())
+					writef("\tfor k := range src.%s {", fname)
+					// use zero-length slice instead of nil to ensure
+					// the key is always copied.
+					writef("\t\tdst.%s[k] = append([]%s{}, src.%s[k]...)", fname, n, fname)
+					writef("\t}")
+				} else if containsPointers(ft.Elem()) {
+					writef("\t\t" + `panic("TODO map value pointers")`)
+				} else {
+					writef("\tfor k, v := range src.%s {", fname)
+					writef("\t\tdst.%s[k] = v", fname)
+					writef("\t}")
+				}
+				writef("}")
+			case *types.Struct:
+				writef(`panic("TODO struct %s")`, fname)
+			default:
+				writef(`panic(fmt.Sprintf("TODO: %T", ft))`)
+			}
+		}
+		writef("return dst")
+		fmt.Fprintf(buf, "}\n\n")
+	}
+}
+
+func hasBasicUnderlying(typ types.Type) bool {
+	switch typ.Underlying().(type) {
+	case *types.Slice, *types.Map:
+		return true
+	default:
+		return false
+	}
+}
+
+func containsPointers(typ types.Type) bool {
+	switch typ.String() {
+	case "time.Time":
+		// time.Time contains a pointer that does not need copying
+		return false
+	case "inet.af/netaddr.IP":
+		return false
+	}
+	switch ft := typ.Underlying().(type) {
+	case *types.Array:
+		return containsPointers(ft.Elem())
+	case *types.Chan:
+		return true
+	case *types.Interface:
+		return true // a little too broad
+	case *types.Map:
+		return true
+	case *types.Pointer:
+		return true
+	case *types.Slice:
+		return true
+	case *types.Struct:
+		for i := 0; i < ft.NumFields(); i++ {
+			if containsPointers(ft.Field(i).Type()) {
+				return true
+			}
+		}
+	}
+	return false
+}

cmd/derper/derper.go

@@ -6,9 +6,10 @@
 package main // import "tailscale.com/cmd/derper"
 
 import (
+	"context"
 	"encoding/json"
+	"errors"
 	"expvar"
-	_ "expvar"
 	"flag"
 	"fmt"
 	"io"
@@ -16,9 +17,9 @@ import (
 	"log"
 	"net"
 	"net/http"
-	_ "net/http/pprof"
 	"os"
 	"path/filepath"
+	"regexp"
 	"strings"
 	"time"
 
@@ -27,8 +28,10 @@ import (
 	"tailscale.com/atomicfile"
 	"tailscale.com/derp"
 	"tailscale.com/derp/derphttp"
-	"tailscale.com/interfaces"
 	"tailscale.com/logpolicy"
+	"tailscale.com/metrics"
+	"tailscale.com/net/stun"
+	"tailscale.com/tsweb"
 	"tailscale.com/types/key"
 )
 
@@ -36,20 +39,14 @@ var (
 	dev           = flag.Bool("dev", false, "run in localhost development mode")
 	addr          = flag.String("a", ":443", "server address")
 	configPath    = flag.String("c", "", "config file path")
-	certDir       = flag.String("certdir", defaultCertDir(), "directory to store LetsEncrypt certs, if addr's port is :443")
+	certDir       = flag.String("certdir", tsweb.DefaultCertDir("derper-certs"), "directory to store LetsEncrypt certs, if addr's port is :443")
 	hostname      = flag.String("hostname", "derp.tailscale.com", "LetsEncrypt host name, if addr's port is :443")
-	mbps          = flag.Int("mbps", 5, "Mbps (mebibit/s) per-client rate limit; 0 means unlimited")
 	logCollection = flag.String("logcollection", "", "If non-empty, logtail collection to log to")
+	runSTUN       = flag.Bool("stun", false, "also run a STUN server")
+	meshPSKFile   = flag.String("mesh-psk-file", defaultMeshPSKFile(), "if non-empty, path to file containing the mesh pre-shared key file. It should contain some hex string; whitespace is trimmed.")
+	meshWith      = flag.String("mesh-with", "", "optional comma-separated list of hostnames to mesh with; the server's own hostname can be in the list")
 )
 
-func defaultCertDir() string {
-	cacheDir, err := os.UserCacheDir()
-	if err == nil {
-		return filepath.Join(cacheDir, "tailscale", "derper-certs")
-	}
-	return ""
-}
-
 type config struct {
 	PrivateKey wgcfg.PrivateKey
 }
@@ -110,6 +107,7 @@ func main() {
 		*logCollection = ""
 		*addr = ":3340" // above the keys DERP
 		log.Printf("Running in dev mode.")
+		tsweb.DevMode = true
 	}
 
 	var logPol *logpolicy.Policy
@@ -120,24 +118,30 @@ func main() {
 
 	cfg := loadConfig()
 
-	letsEncrypt := false
-	if _, port, _ := net.SplitHostPort(*addr); port == "443" {
-		letsEncrypt = true
-	}
+	letsEncrypt := tsweb.IsProd443(*addr)
 
 	s := derp.NewServer(key.Private(cfg.PrivateKey), log.Printf)
-	if *mbps != 0 {
-		s.BytesPerSecond = (*mbps << 20) / 8
+
+	if *meshPSKFile != "" {
+		b, err := ioutil.ReadFile(*meshPSKFile)
+		if err != nil {
+			log.Fatal(err)
+		}
+		key := strings.TrimSpace(string(b))
+		if matched, _ := regexp.MatchString(`(?i)^[0-9a-f]{64,}$`, key); !matched {
+			log.Fatalf("key in %s must contain 64+ hex digits", *meshPSKFile)
+		}
+		s.SetMeshKey(key)
+		log.Printf("DERP mesh key configured")
+	}
+	if err := startMesh(s); err != nil {
+		log.Fatalf("startMesh: %v", err)
 	}
 	expvar.Publish("derp", s.ExpVar())
-	expvar.Publish("uptime", uptimeVar{})
 
 	// Create our own mux so we don't expose /debug/ stuff to the world.
-	mux := http.NewServeMux()
+	mux := tsweb.NewMux(debugHandler(s))
 	mux.Handle("/derp", derphttp.Handler(s))
-	mux.Handle("/debug/", protected(debugHandler(s)))
-	mux.Handle("/debug/pprof/", protected(http.DefaultServeMux)) // to net/http/pprof
-	mux.Handle("/debug/vars", protected(http.DefaultServeMux))   // to expvar
 	mux.Handle("/", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		w.Header().Set("Content-Type", "text/html; charset=utf-8")
 		w.WriteHeader(200)
@@ -150,11 +154,15 @@ func main() {
   server.
 </p>
 `)
-		if allowDebugAccess(r) {
+		if tsweb.AllowDebugAccess(r) {
 			io.WriteString(w, "<p>Debug info at <a href='/debug/'>/debug/</a>.</p>\n")
 		}
 	}))
 
+	if *runSTUN {
+		go serveSTUN()
+	}
+
 	httpsrv := &http.Server{
 		Addr:    *addr,
 		Handler: mux,
@@ -171,9 +179,13 @@ func main() {
 			HostPolicy: autocert.HostWhitelist(*hostname),
 			Cache:      autocert.DirCache(*certDir),
 		}
+		if *hostname == "derp.tailscale.com" {
+			certManager.HostPolicy = prodAutocertHostPolicy
+			certManager.Email = "security@tailscale.com"
+		}
 		httpsrv.TLSConfig = certManager.TLSConfig()
 		go func() {
-			err := http.ListenAndServe(":80", certManager.HTTPHandler(port80Handler{mux}))
+			err := http.ListenAndServe(":80", certManager.HTTPHandler(tsweb.Port80Handler{Main: mux}))
 			if err != nil {
 				if err != http.ErrServerClosed {
 					log.Fatal(err)
@@ -190,82 +202,120 @@ func main() {
 	}
 }
 
-func protected(h http.Handler) http.Handler {
-	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if !allowDebugAccess(r) {
-			http.Error(w, "debug access denied", http.StatusForbidden)
-			return
-		}
-		h.ServeHTTP(w, r)
-	})
-}
-
-func allowDebugAccess(r *http.Request) bool {
-	if r.Header.Get("X-Forwarded-For") != "" {
-		// TODO if/when needed. For now, conservative:
-		return false
-	}
-	ipStr, _, err := net.SplitHostPort(r.RemoteAddr)
-	if err != nil {
-		return false
-	}
-	ip := net.ParseIP(ipStr)
-	return interfaces.IsTailscaleIP(ip) || ip.IsLoopback() || ipStr == os.Getenv("ALLOW_DEBUG_IP")
-}
-
-type port80Handler struct{ tlsHandler http.Handler }
-
-func (h port80Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
-	path := r.RequestURI
-	if path == "/debug" || strings.HasPrefix(path, "/debug") {
-		h.tlsHandler.ServeHTTP(w, r)
-		return
-	}
-	if r.Method != "GET" && r.Method != "HEAD" {
-		http.Error(w, "Use HTTPS", http.StatusBadRequest)
-		return
-	}
-	if path == "/" && allowDebugAccess(r) {
-		// Redirect authorized user to the debug handler.
-		path = "/debug/"
-	}
-	target := "https://" + stripPort(r.Host) + path
-	http.Redirect(w, r, target, http.StatusFound)
-}
-
-func stripPort(hostport string) string {
-	host, _, err := net.SplitHostPort(hostport)
-	if err != nil {
-		return hostport
-	}
-	return net.JoinHostPort(host, "443")
-}
-
 func debugHandler(s *derp.Server) http.Handler {
 	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.RequestURI == "/debug/check" {
+			err := s.ConsistencyCheck()
+			if err != nil {
+				http.Error(w, err.Error(), 500)
+			} else {
+				io.WriteString(w, "derp.Server ConsistencyCheck okay")
+			}
+			return
+		}
 		f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
 		f(`<html><body>
 <h1>DERP debug</h1>
 <ul>
 `)
 		f("<li><b>Hostname:</b> %v</li>\n", *hostname)
-		f("<li><b>Rate Limit:</b> %v Mbps</li>\n", *mbps)
-		f("<li><b>Uptime:</b> %v</li>\n", uptime().Round(time.Second))
+		f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
+		f("<li><b>Mesh Key:</b> %v</li>\n", s.HasMeshKey())
 
-		f(`<li><a href="/debug/vars">/debug/vars</a></li>
+		f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
+   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
    <li><a href="/debug/pprof/">/debug/pprof/</a></li>
    <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
    <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
+   <li><a href="/debug/check">/debug/check</a> internal consistency check</li>
 <ul>
 </html>
 `)
 	})
 }
 
-var timeStart = time.Now()
+func serveSTUN() {
+	pc, err := net.ListenPacket("udp", ":3478")
+	if err != nil {
+		log.Fatalf("failed to open STUN listener: %v", err)
+	}
+	log.Printf("running STUN server on %v", pc.LocalAddr())
 
-func uptime() time.Duration { return time.Since(timeStart) }
+	var (
+		stats           = new(metrics.Set)
+		stunDisposition = &metrics.LabelMap{Label: "disposition"}
+		stunAddrFamily  = &metrics.LabelMap{Label: "family"}
 
-type uptimeVar struct{}
+		stunReadError  = stunDisposition.Get("read_error")
+		stunNotSTUN    = stunDisposition.Get("not_stun")
+		stunWriteError = stunDisposition.Get("write_error")
+		stunSuccess    = stunDisposition.Get("success")
 
-func (uptimeVar) String() string { return fmt.Sprint(int64(uptime().Seconds())) }
+		stunIPv4 = stunAddrFamily.Get("ipv4")
+		stunIPv6 = stunAddrFamily.Get("ipv6")
+	)
+	stats.Set("counter_requests", stunDisposition)
+	stats.Set("counter_addrfamily", stunAddrFamily)
+	expvar.Publish("stun", stats)
+
+	var buf [64 << 10]byte
+	for {
+		n, addr, err := pc.ReadFrom(buf[:])
+		if err != nil {
+			log.Printf("STUN ReadFrom: %v", err)
+			time.Sleep(time.Second)
+			stunReadError.Add(1)
+			continue
+		}
+		ua, ok := addr.(*net.UDPAddr)
+		if !ok {
+			log.Printf("STUN unexpected address %T %v", addr, addr)
+			stunReadError.Add(1)
+			continue
+		}
+		pkt := buf[:n]
+		if !stun.Is(pkt) {
+			stunNotSTUN.Add(1)
+			continue
+		}
+		txid, err := stun.ParseBindingRequest(pkt)
+		if err != nil {
+			stunNotSTUN.Add(1)
+			continue
+		}
+		if ua.IP.To4() != nil {
+			stunIPv4.Add(1)
+		} else {
+			stunIPv6.Add(1)
+		}
+		res := stun.Response(txid, ua.IP, uint16(ua.Port))
+		_, err = pc.WriteTo(res, addr)
+		if err != nil {
+			stunWriteError.Add(1)
+		} else {
+			stunSuccess.Add(1)
+		}
+	}
+}
+
+var validProdHostname = regexp.MustCompile(`^derp([^.]*)\.tailscale\.com\.?$`)
+
+func prodAutocertHostPolicy(_ context.Context, host string) error {
+	if validProdHostname.MatchString(host) {
+		return nil
+	}
+	return errors.New("invalid hostname")
+}
+
+func defaultMeshPSKFile() string {
+	try := []string{
+		"/home/derp/keys/derp-mesh.key",
+		filepath.Join(os.Getenv("HOME"), "keys", "derp-mesh.key"),
+	}
+	for _, p := range try {
+		if _, err := os.Stat(p); err == nil {
+			return p
+		}
+	}
+	return ""
+}

cmd/derper/derper_test.go

@@ -0,0 +1,35 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"context"
+	"testing"
+)
+
+func TestProdAutocertHostPolicy(t *testing.T) {
+	tests := []struct {
+		in     string
+		wantOK bool
+	}{
+		{"derp.tailscale.com", true},
+		{"derp.tailscale.com.", true},
+		{"derp1.tailscale.com", true},
+		{"derp1b.tailscale.com", true},
+		{"derp2.tailscale.com", true},
+		{"derp02.tailscale.com", true},
+		{"derp-nyc.tailscale.com", true},
+		{"derpfoo.tailscale.com", true},
+		{"derp02.bar.tailscale.com", false},
+		{"example.net", false},
+	}
+	for _, tt := range tests {
+		got := prodAutocertHostPolicy(context.Background(), tt.in) == nil
+		if got != tt.wantOK {
+			t.Errorf("f(%q) = %v; want %v", tt.in, got, tt.wantOK)
+		}
+	}
+
+}

cmd/derper/mesh.go

@@ -0,0 +1,45 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import (
+	"errors"
+	"fmt"
+	"log"
+	"strings"
+
+	"tailscale.com/derp"
+	"tailscale.com/derp/derphttp"
+	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
+)
+
+func startMesh(s *derp.Server) error {
+	if *meshWith == "" {
+		return nil
+	}
+	if !s.HasMeshKey() {
+		return errors.New("--mesh-with requires --mesh-psk-file")
+	}
+	for _, host := range strings.Split(*meshWith, ",") {
+		if err := startMeshWithHost(s, host); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func startMeshWithHost(s *derp.Server, host string) error {
+	logf := logger.WithPrefix(log.Printf, fmt.Sprintf("mesh(%q): ", host))
+	c, err := derphttp.NewClient(s.PrivateKey(), "https://"+host+"/derp", logf)
+	if err != nil {
+		return err
+	}
+	c.MeshKey = s.MeshKey()
+	add := func(k key.Public) { s.AddPacketForwarder(k, c) }
+	remove := func(k key.Public) { s.RemovePacketForwarder(k, c) }
+	go c.RunWatchConnectionLoop(s.PublicKey(), add, remove)
+	return nil
+}

cmd/microproxy/microproxy.go

@@ -0,0 +1,175 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// microproxy proxies incoming HTTPS connections to another
+// destination. Instead of managing its own TLS certificates, it
+// borrows issued certificates and keys from an autocert directory.
+package main
+
+import (
+	"crypto/tls"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"log"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"tailscale.com/logpolicy"
+	"tailscale.com/tsweb"
+)
+
+var (
+	addr          = flag.String("addr", ":4430", "server address")
+	certdir       = flag.String("certdir", "", "directory to borrow LetsEncrypt certificates from")
+	hostname      = flag.String("hostname", "", "hostname to serve")
+	logCollection = flag.String("logcollection", "", "If non-empty, logtail collection to log to")
+	nodeExporter  = flag.String("node-exporter", "http://localhost:9100", "URL of the local prometheus node exporter")
+	goVarsURL     = flag.String("go-vars-url", "http://localhost:8383/debug/vars", "URL of a local Go server's /debug/vars endpoint")
+)
+
+func main() {
+	flag.Parse()
+
+	if *logCollection != "" {
+		logpolicy.New(*logCollection)
+	}
+
+	ne, err := url.Parse(*nodeExporter)
+	if err != nil {
+		log.Fatalf("Couldn't parse URL %q: %v", *nodeExporter, err)
+	}
+	proxy := httputil.NewSingleHostReverseProxy(ne)
+	proxy.FlushInterval = time.Second
+
+	if _, err = url.Parse(*goVarsURL); err != nil {
+		log.Fatalf("Couldn't parse URL %q: %v", *goVarsURL, err)
+	}
+
+	mux := tsweb.NewMux(http.HandlerFunc(debugHandler))
+	mux.Handle("/metrics", tsweb.Protected(proxy))
+	mux.Handle("/varz", tsweb.Protected(tsweb.StdHandler(&goVarsHandler{*goVarsURL}, log.Printf)))
+
+	ch := &certHolder{
+		hostname: *hostname,
+		path:     filepath.Join(*certdir, *hostname),
+	}
+
+	httpsrv := &http.Server{
+		Addr:    *addr,
+		Handler: mux,
+		TLSConfig: &tls.Config{
+			GetCertificate: ch.GetCertificate,
+		},
+	}
+
+	if err := httpsrv.ListenAndServeTLS("", ""); err != nil && err != http.ErrServerClosed {
+		log.Fatal(err)
+	}
+}
+
+type goVarsHandler struct {
+	url string
+}
+
+func promPrint(w io.Writer, prefix string, obj map[string]interface{}) {
+	for k, i := range obj {
+		if prefix != "" {
+			k = prefix + "_" + k
+		}
+		switch v := i.(type) {
+		case map[string]interface{}:
+			promPrint(w, k, v)
+		case float64:
+			fmt.Fprintf(w, "%s %f\n", k, v)
+		default:
+			fmt.Fprintf(w, "# Skipping key %q, unhandled type %T\n", k, v)
+		}
+	}
+}
+
+func (h *goVarsHandler) ServeHTTPReturn(w http.ResponseWriter, r *http.Request) error {
+	resp, err := http.Get(h.url)
+	if err != nil {
+		return tsweb.Error(http.StatusInternalServerError, "fetch failed", err)
+	}
+	defer resp.Body.Close()
+	var mon map[string]interface{}
+	if err := json.NewDecoder(resp.Body).Decode(&mon); err != nil {
+		return tsweb.Error(http.StatusInternalServerError, "fetch failed", err)
+	}
+
+	w.WriteHeader(http.StatusOK)
+	promPrint(w, "", mon)
+	return nil
+}
+
+// certHolder loads and caches a TLS certificate from disk, reloading
+// it every hour.
+type certHolder struct {
+	hostname string // only hostname allowed in SNI
+	path     string // path of certificate+key combined PEM file
+
+	mu     sync.Mutex
+	cert   *tls.Certificate // cached parsed cert+key
+	loaded time.Time
+}
+
+func (c *certHolder) GetCertificate(ch *tls.ClientHelloInfo) (*tls.Certificate, error) {
+	if ch.ServerName != c.hostname {
+		return nil, fmt.Errorf("wrong client SNI %q", ch.ServerName)
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if time.Since(c.loaded) > time.Hour {
+		if err := c.loadLocked(); err != nil {
+			log.Printf("Reloading cert %q: %v", c.path, err)
+			// continue anyway, we might be able to serve off the stale cert.
+		}
+	}
+	return c.cert, nil
+}
+
+// load reloads the TLS certificate and key from disk. Caller must
+// hold mu.
+func (c *certHolder) loadLocked() error {
+	bs, err := ioutil.ReadFile(c.path)
+	if err != nil {
+		return fmt.Errorf("reading %q: %v", c.path, err)
+	}
+	cert, err := tls.X509KeyPair(bs, bs)
+	if err != nil {
+		return fmt.Errorf("parsing %q: %v", c.path, err)
+	}
+
+	c.cert = &cert
+	c.loaded = time.Now()
+	return nil
+}
+
+// debugHandler serves a page with links to tsweb-managed debug URLs
+// at /debug/.
+func debugHandler(w http.ResponseWriter, r *http.Request) {
+	f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
+	f(`<html><body>
+<h1>microproxy debug</h1>
+<ul>
+`)
+	f("<li><b>Hostname:</b> %v</li>\n", *hostname)
+	f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
+	f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
+   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
+   <li><a href="/debug/pprof/">/debug/pprof/</a></li>
+   <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
+   <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
+<ul>
+</html>
+`)
+}

cmd/mkpkg/main.go

@@ -31,16 +31,28 @@ func parseFiles(s string) (map[string]string, error) {
 	return ret, nil
 }
 
+func parseEmptyDirs(s string) []string {
+	// strings.Split("", ",") would return []string{""}, which is not suitable:
+	// this would create an empty dir record with path "", breaking the package
+	if s == "" {
+		return nil
+	}
+	return strings.Split(s, ",")
+}
+
 func main() {
 	out := getopt.StringLong("out", 'o', "", "output file to write")
 	goarch := getopt.StringLong("arch", 'a', "amd64", "GOARCH this package is for")
 	pkgType := getopt.StringLong("type", 't', "deb", "type of package to build (deb or rpm)")
 	files := getopt.StringLong("files", 'F', "", "comma-separated list of files in src:dst form")
 	configFiles := getopt.StringLong("configs", 'C', "", "like --files, but for files marked as user-editable config files")
+	emptyDirs := getopt.StringLong("emptydirs", 'E', "", "comma-separated list of empty directories")
 	version := getopt.StringLong("version", 0, "0.0.0", "version of the package")
 	postinst := getopt.StringLong("postinst", 0, "", "debian postinst script path")
 	prerm := getopt.StringLong("prerm", 0, "", "debian prerm script path")
 	postrm := getopt.StringLong("postrm", 0, "", "debian postrm script path")
+	replaces := getopt.StringLong("replaces", 0, "", "package which this package replaces, if any")
+	depends := getopt.StringLong("depends", 0, "", "comma-separated list of packages this package depends on")
 	getopt.Parse()
 
 	filesMap, err := parseFiles(*files)
@@ -51,6 +63,7 @@ func main() {
 	if err != nil {
 		log.Fatalf("Parsing --configs: %v", err)
 	}
+	emptyDirList := parseEmptyDirs(*emptyDirs)
 	info := nfpm.WithDefaults(&nfpm.Info{
 		Name:        "tailscale",
 		Arch:        *goarch,
@@ -61,20 +74,29 @@ func main() {
 		Homepage:    "https://www.tailscale.com",
 		License:     "MIT",
 		Overridables: nfpm.Overridables{
-			Files:       filesMap,
-			ConfigFiles: configsMap,
+			EmptyFolders: emptyDirList,
+			Files:        filesMap,
+			ConfigFiles:  configsMap,
+			Scripts: nfpm.Scripts{
+				PostInstall: *postinst,
+				PreRemove:   *prerm,
+				PostRemove:  *postrm,
+			},
 		},
 	})
 
+	if len(*depends) != 0 {
+		info.Overridables.Depends = strings.Split(*depends, ",")
+	}
+	if *replaces != "" {
+		info.Overridables.Replaces = []string{*replaces}
+		info.Overridables.Conflicts = []string{*replaces}
+	}
+
 	switch *pkgType {
 	case "deb":
 		info.Section = "net"
 		info.Priority = "extra"
-		info.Overridables.Scripts = nfpm.Scripts{
-			PostInstall: *postinst,
-			PreRemove:   *prerm,
-			PostRemove:  *postrm,
-		}
 	case "rpm":
 		info.Overridables.RPM.Group = "Network"
 	}

cmd/relaynode/.gitignore

@@ -1,14 +0,0 @@
-/*.tar.gz
-/*.deb
-/*.rpm
-/*.spec
-/pkgver
-debian/changelog
-debian/debhelper-build-stamp
-debian/files
-debian/*.log
-debian/*.substvars
-debian/*.debhelper
-debian/tailscale-relay
-/tailscale-relay/
-/tailscale-relay-*

cmd/relaynode/clean.do

@@ -1 +0,0 @@
-rm -f debian/changelog *~ debian/*~

cmd/relaynode/clean.od

@@ -1,13 +0,0 @@
-exec >&2
-read -r package <package
-rm -f *~ .*~ \
-	debian/*~ debian/changelog debian/debhelper-build-stamp \
-	debian/*.log debian/files debian/*.substvars debian/*.debhelper \
-	*.tar.gz *.deb *.rpm *.spec pkgver relaynode *.exe
-[ -n "$package" ] && rm -rf "debian/$package"
-for d in */.stamp; do
-	if [ -e "$d" ]; then
-		dir=$(dirname "$d")
-		rm -rf "$dir"
-	fi
-done

cmd/relaynode/deb.od

@@ -1,10 +0,0 @@
-exec >&2
-dir=${1%/*}
-redo-ifchange "$S/$dir/package" "$S/oss/version/short.txt"
-read -r package <"$S/$dir/package"
-read -r version <"$S/oss/version/short.txt"
-arch=$(dpkg --print-architecture)
-
-redo-ifchange "$dir/${package}_$arch.deb"
-rm -f "$dir/${package}"_*_"$arch.deb"
-ln -sf "${package}_$arch.deb" "$dir/${package}_${version}_$arch.deb"

cmd/relaynode/debian/README.Debian

@@ -1 +0,0 @@
-Tailscale IPN relay daemon.

cmd/relaynode/debian/changelog.do

@@ -1,5 +0,0 @@
-redo-ifchange ../../../version/short.txt gen-changelog
-(
-	cd ..
-	debian/gen-changelog
-) >$3

cmd/relaynode/debian/compat

@@ -1 +0,0 @@
-9

cmd/relaynode/debian/control

@@ -1,14 +0,0 @@
-Source: tailscale-relay
-Section: net
-Priority: extra
-Maintainer: Avery Pennarun <apenwarr@tailscale.com>
-Build-Depends: debhelper (>= 10.2.5), dh-systemd (>= 1.5)
-Standards-Version: 3.9.2
-Homepage: https://tailscale.com/
-Vcs-Git: https://github.com/tailscale/tailscale
-Vcs-Browser: https://github.com/tailscale/tailscale
-
-Package: tailscale-relay
-Architecture: any
-Depends: ${shlibs:Depends}, ${misc:Depends}
-Description: Traffic relay node for Tailscale IPN

cmd/relaynode/debian/copyright

@@ -1,11 +0,0 @@
-Format: http://svn.debian.org/wsvn/dep/web/deps/dep5.mdwn?op=file&rev=173
-Upstream-Name: tailscale-relay
-Upstream-Contact: Avery Pennarun <apenwarr@tailscale.com>
-Source: https://github.com/tailscale/tailscale/
-
-Files: *
-Copyright: © 2019 Tailscale Inc. <info@tailscale.com>
-License: Proprietary
- *
- * Copyright 2019 Tailscale Inc. All rights reserved.
- *

cmd/relaynode/debian/gen-changelog

@@ -1,25 +0,0 @@
-#!/bin/sh
-read junk pkgname <debian/control
-read shortver <../../version/short.txt
-git log --pretty='format:'"$pkgname"' (SHA:%H) unstable; urgency=low
-
-  * %s
-  
- -- %aN <%aE>  %aD
-' . |
-python -Sc '
-import os, re, subprocess, sys
-
-first = True
-def Describe(g):
-  global first
-  if first:
-    s = sys.argv[1]
-    first = False
-  else:
-    sha = g.group(1)
-    s = subprocess.check_output(["git", "describe", "--always", "--", sha]).strip().decode("utf-8")
-  return re.sub(r"^\D*", "", s)
-
-print(re.sub(r"SHA:([0-9a-f]+)", Describe, sys.stdin.read()))
-' "$shortver"

cmd/relaynode/debian/install

@@ -1,3 +0,0 @@
-relaynode			/usr/sbin
-tailscale-login			/usr/sbin
-taillogin			/usr/sbin

cmd/relaynode/debian/postinst

@@ -1,8 +0,0 @@
-#DEBHELPER#
-
-f=/var/lib/tailscale/relay.conf
-if ! [ -e "$f" ]; then
-	echo
-	echo "Note: Run tailscale-login to configure $f." >&2
-	echo
-fi

cmd/relaynode/debian/rules

@@ -1,10 +0,0 @@
-#!/usr/bin/make -f
-DESTDIR=debian/tailscale-relay
-
-override_dh_auto_test:
-override_dh_auto_install:
-	mkdir -p "${DESTDIR}/etc/default"
-	cp tailscale-relay.defaults "${DESTDIR}/etc/default/tailscale-relay"
-
-%:
-	dh $@ --with=systemd

cmd/relaynode/debian/tailscale-relay.service

@@ -1,12 +0,0 @@
-[Unit]
-Description=Traffic relay node for Tailscale IPN
-After=network.target
-ConditionPathExists=/var/lib/tailscale/relay.conf
-
-[Service]
-EnvironmentFile=/etc/default/tailscale-relay
-ExecStart=/usr/sbin/relaynode --config=/var/lib/tailscale/relay.conf --tun=wg0 $PORT $FLAGS
-Restart=on-failure
-
-[Install]
-WantedBy=multi-user.target

cmd/relaynode/default.deb.od

@@ -1,21 +0,0 @@
-exec >&2
-dir=${1%/*}
-redo-ifchange "$S/oss/version/short.txt" "$S/$dir/package" "$dir/debtmp.dir"
-read -r package <"$S/$dir/package"
-read -r version <"$S/oss/version/short.txt"
-arch=$(dpkg --print-architecture)
-
-(
-	cd "$S/$dir"
-	git ls-files debian | xargs redo-ifchange debian/changelog
-)
-cp -a "$S/$dir/debian" "$dir/debtmp/"
-rm -f "$dir/debtmp/debian/$package.debhelper.log"
-rm -f "$dir/${package}_${version}_${arch}.deb"
-(
-	cd "$dir/debtmp" &&
-	debian/rules build &&
-	fakeroot debian/rules binary
-)
-
-mv "$dir/${package}_${version}_${arch}.deb" "$3"

cmd/relaynode/default.dir.od

@@ -1,20 +0,0 @@
-# Generate a directory tree suitable for forming a tarball of
-# this package.
-exec >&2
-dir=${1%/*}
-outdir=$PWD/${1%.dir}
-rm -rf "$outdir"
-mkdir "$outdir"
-touch $outdir/.stamp
-sfiles="
-	tailscale-login
-	debian/*.service
-	*.defaults
-"
-ofiles="
-	relaynode
-	../taillogin/taillogin
-"
-redo-ifchange "$outdir/.stamp"
-(cd "$S/$dir" && redo-ifchange $sfiles && cp $sfiles "$outdir/")
-(cd "$dir" && redo-ifchange $ofiles && cp $ofiles "$outdir/")

cmd/relaynode/default.rpm.od

@@ -1,15 +0,0 @@
-exec >&2
-dir=${1%/*}
-pkg=${1##*/}
-pkg=${pkg%.rpm}
-redo-ifchange "$S/oss/version/short.txt" "$dir/$pkg.tar.gz" "$dir/$pkg.spec"
-read -r pkgver junk <"$S/oss/version/short.txt"
-
-machine=$(uname -m)
-rpmbase=$HOME/rpmbuild
-
-mkdir -p "$rpmbase/SOURCES/"
-cp "$dir/$pkg.tar.gz" "$rpmbase/SOURCES/"
-rm -f "$rpmbase/RPMS/$machine/$pkg-$pkgver.$machine.rpm"
-rpmbuild -bb "$dir/$pkg.spec"
-mv "$rpmbase/RPMS/$machine/$pkg-$pkgver.$machine.rpm" $3

cmd/relaynode/default.spec.od

@@ -1,7 +0,0 @@
-redo-ifchange "$S/$1.in" "$S/oss/version/short.txt"
-read -r pkgver junk <"$S/oss/version/short.txt"
-basever=${pkgver%-*}
-subver=${pkgver#*-}
-sed -e "s/Version: 0.00$/Version: $basever/" \
-    -e "s/Release: 0$/Release: $subver/" \
-	<"$S/$1.in" >"$3"

cmd/relaynode/default.tar.gz.od

@@ -1,8 +0,0 @@
-exec >&2
-xdir=${1%.tar.gz}
-base=${xdir##*/}
-updir=${xdir%/*}
-redo-ifchange "$xdir.dir"
-OUT="$PWD/$3"
-
-cd "$updir" && tar -czvf "$OUT" --exclude "$base/.stamp" "$base"

cmd/relaynode/dist.od

@@ -1,15 +0,0 @@
-# Build packages for customer distribution.
-dir=${1%/*}
-cd "$dir"
-targets="tarball"
-if which dh_clean fakeroot dpkg >/dev/null; then
-	targets="$targets deb"
-else
-	echo "Skipping debian packages: debhelper and/or dpkg build tools missing." >&2
-fi
-if which rpm >/dev/null; then
-	targets="$targets rpm"
-else
-	echo "Skipping rpm packages: rpm build tools missing." >&2
-fi
-redo-ifchange $targets

cmd/relaynode/docker/.gitignore

@@ -1 +0,0 @@
-/relaynode

cmd/relaynode/docker/Dockerfile

@@ -1,17 +0,0 @@
-# Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-# Use of this source code is governed by a BSD-style
-# license that can be found in the LICENSE file.
-
-# Build with: docker build -t tailcontrol-alpine .
-# Run with: docker run --cap-add=NET_ADMIN --device=/dev/net/tun:/dev/net/tun -it tailcontrol-alpine
-
-FROM debian:stretch-slim
-
-RUN apt-get update && apt-get -y install iproute2 iptables
-RUN apt-get -y install ca-certificates
-RUN apt-get -y install nginx-light
-
-COPY relaynode /
-
-# tailcontrol -tun=wg0 -dbdir=$HOME/taildb >> tailcontrol.log 2>&1 &
-CMD ["/relaynode", "-R", "--config", "relay.conf"]

cmd/relaynode/docker/all.do

@@ -1 +0,0 @@
-redo-ifchange build

cmd/relaynode/docker/build.do

@@ -1,3 +0,0 @@
-exec >&2
-redo-ifchange Dockerfile relaynode
-docker build -t tailscale .

cmd/relaynode/docker/relaynode.do

@@ -1,2 +0,0 @@
-redo-ifchange ../relaynode
-cp ../relaynode $3

cmd/relaynode/docker/run.sh

@@ -1,10 +0,0 @@
-#!/bin/sh
-# Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-# Use of this source code is governed by a BSD-style
-# license that can be found in the LICENSE file.
-
-set -e
-redo-ifchange build
-docker run --cap-add=NET_ADMIN \
-	--device=/dev/net/tun:/dev/net/tun \
-	-it tailscale

cmd/relaynode/package

@@ -1 +0,0 @@
-tailscale-relay

cmd/relaynode/relaynode.go

@@ -1,239 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Relaynode is the old Linux Tailscale daemon.
-//
-// Deprecated: this program will be soon deleted. The replacement is
-// cmd/tailscaled.
-package main // import "tailscale.com/cmd/relaynode"
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"io/ioutil"
-	"log"
-	"net/http"
-	"net/http/pprof"
-	"os"
-	"os/signal"
-	"strings"
-	"syscall"
-	"time"
-
-	"github.com/apenwarr/fixconsole"
-	"github.com/google/go-cmp/cmp"
-	"github.com/klauspost/compress/zstd"
-	"github.com/pborman/getopt/v2"
-	"github.com/tailscale/wireguard-go/wgcfg"
-	"tailscale.com/atomicfile"
-	"tailscale.com/control/controlclient"
-	"tailscale.com/logpolicy"
-	"tailscale.com/version"
-	"tailscale.com/wgengine"
-	"tailscale.com/wgengine/filter"
-	"tailscale.com/wgengine/magicsock"
-)
-
-func main() {
-	err := fixconsole.FixConsoleIfNeeded()
-	if err != nil {
-		log.Printf("fixConsoleOutput: %v\n", err)
-	}
-	config := getopt.StringLong("config", 'f', "", "path to config file")
-	server := getopt.StringLong("server", 's', "https://login.tailscale.com", "URL to tailcontrol server")
-	listenport := getopt.Uint16Long("port", 'p', magicsock.DefaultPort, "WireGuard port (0=autoselect)")
-	tunname := getopt.StringLong("tun", 0, "wg0", "tunnel interface name")
-	alwaysrefresh := getopt.BoolLong("always-refresh", 0, "force key refresh at startup")
-	fake := getopt.BoolLong("fake", 0, "fake tunnel+routing instead of tuntap")
-	nuroutes := getopt.BoolLong("no-single-routes", 'N', "disallow (non-subnet) routes to single nodes")
-	rroutes := getopt.BoolLong("remote-routes", 'R', "allow routing subnets to remote nodes")
-	droutes := getopt.BoolLong("default-routes", 'D', "allow default route on remote node")
-	routes := getopt.StringLong("routes", 0, "", "list of IP ranges this node can relay")
-	debug := getopt.StringLong("debug", 0, "", "Address of debug server")
-	getopt.Parse()
-	if len(getopt.Args()) > 0 {
-		log.Fatalf("too many non-flag arguments: %#v", getopt.Args()[0])
-	}
-	uflags := controlclient.UFlagsHelper(!*nuroutes, *rroutes, *droutes)
-	if *config == "" {
-		log.Fatal("no --config file specified")
-	}
-	if *tunname == "" {
-		log.Printf("Warning: no --tun device specified; routing disabled.\n")
-	}
-
-	pol := logpolicy.New("tailnode.log.tailscale.io")
-
-	logf := wgengine.RusagePrefixLog(log.Printf)
-
-	// The wgengine takes a wireguard configuration produced by the
-	// controlclient, and runs the actual tunnels and packets.
-	var e wgengine.Engine
-	if *fake {
-		e, err = wgengine.NewFakeUserspaceEngine(logf, *listenport)
-	} else {
-		e, err = wgengine.NewUserspaceEngine(logf, *tunname, *listenport)
-	}
-	if err != nil {
-		log.Fatalf("Error starting wireguard engine: %v\n", err)
-	}
-
-	e = wgengine.NewWatchdog(e)
-
-	// Default filter blocks everything, until Start() is called.
-	e.SetFilter(filter.NewAllowNone())
-
-	var lastNetMap *controlclient.NetworkMap
-	statusFunc := func(new controlclient.Status) {
-		if new.URL != "" {
-			fmt.Fprintf(os.Stderr, "To authenticate, visit:\n\n\t%s\n\n", new.URL)
-			return
-		}
-		if new.Err != "" {
-			log.Print(new.Err)
-			return
-		}
-		if new.Persist != nil {
-			if err := saveConfig(*config, *new.Persist); err != nil {
-				log.Println(err)
-			}
-		}
-
-		if m := new.NetMap; m != nil {
-			if lastNetMap != nil {
-				s1 := strings.Split(lastNetMap.Concise(), "\n")
-				s2 := strings.Split(new.NetMap.Concise(), "\n")
-				logf("netmap diff:\n%v\n", cmp.Diff(s1, s2))
-			}
-			lastNetMap = m
-
-			if m.Equal(&controlclient.NetworkMap{}) {
-				return
-			}
-
-			log.Printf("packet filter: %v\n", m.PacketFilter)
-			e.SetFilter(filter.New(m.PacketFilter))
-
-			wgcfg, err := m.WGCfg(uflags, m.DNS)
-			if err != nil {
-				log.Fatalf("Error getting wg config: %v\n", err)
-			}
-			err = e.Reconfig(wgcfg, m.DNSDomains)
-			if err != nil {
-				log.Fatalf("Error reconfiguring engine: %v\n", err)
-			}
-		}
-	}
-
-	cfg, err := loadConfig(*config)
-	if err != nil {
-		log.Fatal(err)
-	}
-
-	hi := controlclient.NewHostinfo()
-	hi.FrontendLogID = pol.PublicID.String()
-	hi.BackendLogID = pol.PublicID.String()
-	if *routes != "" {
-		for _, routeStr := range strings.Split(*routes, ",") {
-			cidr, err := wgcfg.ParseCIDR(routeStr)
-			if err != nil {
-				log.Fatalf("--routes: not an IP range: %s", routeStr)
-			}
-			hi.RoutableIPs = append(hi.RoutableIPs, *cidr)
-		}
-	}
-
-	c, err := controlclient.New(controlclient.Options{
-		Persist:   cfg,
-		ServerURL: *server,
-		Hostinfo:  &hi,
-		NewDecompressor: func() (controlclient.Decompressor, error) {
-			return zstd.NewReader(nil)
-		},
-		KeepAlive: true,
-	})
-	c.SetStatusFunc(statusFunc)
-	if err != nil {
-		log.Fatal(err)
-	}
-	lf := controlclient.LoginDefault
-	if *alwaysrefresh {
-		lf |= controlclient.LoginInteractive
-	}
-	c.Login(nil, lf)
-
-	// Print the wireguard status when we get an update.
-	e.SetStatusCallback(func(s *wgengine.Status, err error) {
-		if err != nil {
-			log.Fatalf("Wireguard engine status error: %v\n", err)
-		}
-		var ss []string
-		for _, p := range s.Peers {
-			if p.LastHandshake.IsZero() {
-				ss = append(ss, "x")
-			} else {
-				ss = append(ss, fmt.Sprintf("%d/%d", p.RxBytes, p.TxBytes))
-			}
-		}
-		logf("v%v peers: %v\n", version.LONG, strings.Join(ss, " "))
-		c.UpdateEndpoints(0, s.LocalAddrs)
-	})
-
-	if *debug != "" {
-		go runDebugServer(*debug)
-	}
-
-	sigCh := make(chan os.Signal, 1)
-	signal.Notify(sigCh, os.Interrupt)
-	signal.Notify(sigCh, syscall.SIGTERM)
-
-	<-sigCh
-	logf("signal received, exiting")
-
-	ctx, cancel := context.WithTimeout(context.Background(), 1*time.Second)
-	defer cancel()
-
-	e.Close()
-	pol.Shutdown(ctx)
-}
-
-func loadConfig(path string) (cfg controlclient.Persist, err error) {
-	b, err := ioutil.ReadFile(path)
-	if os.IsNotExist(err) {
-		log.Printf("config %s does not exist", path)
-		return controlclient.Persist{}, nil
-	}
-	if err := json.Unmarshal(b, &cfg); err != nil {
-		return controlclient.Persist{}, fmt.Errorf("load config: %v", err)
-	}
-	return cfg, nil
-}
-
-func saveConfig(path string, cfg controlclient.Persist) error {
-	b, err := json.MarshalIndent(cfg, "", "\t")
-	if err != nil {
-		return fmt.Errorf("save config: %v", err)
-	}
-	if err := atomicfile.WriteFile(path, b, 0666); err != nil {
-		return fmt.Errorf("save config: %v", err)
-	}
-	return nil
-}
-
-func runDebugServer(addr string) {
-	mux := http.NewServeMux()
-	mux.HandleFunc("/debug/pprof/", pprof.Index)
-	mux.HandleFunc("/debug/pprof/cmdline", pprof.Cmdline)
-	mux.HandleFunc("/debug/pprof/profile", pprof.Profile)
-	mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
-	mux.HandleFunc("/debug/pprof/trace", pprof.Trace)
-	srv := http.Server{
-		Addr:    addr,
-		Handler: mux,
-	}
-	if err := srv.ListenAndServe(); err != nil {
-		log.Fatal(err)
-	}
-}

cmd/relaynode/relaynode.od

@@ -0,0 +1 @@
+# placeholder to work around redo bug

cmd/relaynode/rpm.od

@@ -1,9 +0,0 @@
-exec >&2
-dir=${2%/*}
-redo-ifchange "$S/$dir/package" "$S/oss/version/short.txt"
-read -r package <"$S/$dir/package"
-read -r pkgver <"$S/oss/version/short.txt"
-machine=$(uname -m)
-redo-ifchange "$dir/$package.rpm"
-rm -f "$dir/${package}"-*."$machine.rpm"
-ln -sf "$package.rpm" "$dir/$package-$pkgver.$machine.rpm"

cmd/relaynode/tailscale-login

@@ -1,4 +0,0 @@
-#!/bin/sh
-cfg=/var/lib/tailscale/relay.conf
-dir=$(dirname "$0")
-"$dir/taillogin" --config="$cfg"

cmd/relaynode/tailscale-relay.defaults

@@ -1,8 +0,0 @@
-# Set the port to listen on for incoming VPN packets.
-# Remote nodes will automatically be informed about the new port number,
-# but you might want to configure this in order to set external firewall
-# settings.
-PORT="--port=41641"
-
-# Extra flags you might want to pass to relaynode.
-FLAGS=""

cmd/relaynode/tailscale-relay.spec.in

@@ -1,40 +0,0 @@
-Name: tailscale-relay
-Version: 0.00
-Release: 0
-Summary: Traffic relay node for Tailscale
-Group: Network
-License: Proprietary
-URL: https://tailscale.com/
-Vendor: Tailscale Inc.
-#Source: https://github.com/tailscale/tailscale
-Source0: tailscale-relay.tar.gz
-#Prefix: %{_prefix}
-Packager: Avery Pennarun <apenwarr@tailscale.com>
-BuildRoot: %{_tmppath}/%{name}-root
-
-%description
-Traffic relay node for Tailscale.
-
-%prep
-%setup -n tailscale-relay
-
-%build
-
-%install
-D=$RPM_BUILD_ROOT
-[ "$D" = "/" -o -z "$D" ] && exit 99
-rm -rf "$D"
-mkdir -p $D/usr/sbin $D/lib/systemd/system $D/etc/default $D/etc/tailscale
-cp taillogin tailscale-login relaynode $D/usr/sbin
-cp tailscale-relay.service $D/lib/systemd/system/
-cp tailscale-relay.defaults $D/etc/default/tailscale-relay
-
-%clean
-
-%files
-%defattr(-,root,root)
-%config(noreplace) /etc/default/tailscale-relay
-/lib/systemd/system/tailscale-relay.service
-/usr/sbin/taillogin
-/usr/sbin/tailscale-login
-/usr/sbin/relaynode

cmd/relaynode/tarball.od

@@ -1,7 +0,0 @@
-dir=${1%/*}
-redo-ifchange "$S/$dir/package" "$S/oss/version/short.txt"
-read -r package <"$S/$dir/package"
-read -r version <"$S/oss/version/short.txt"
-redo-ifchange "$dir/$package.tar.gz"
-rm -f "$dir/$package"-*.tar.gz
-ln -sf "$package.tar.gz" "$dir/$package-$version.tar.gz"

cmd/taillogin/taillogin.go

@@ -1,99 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// The taillogin command, invoked via the tailscale-login shell script, is shipped
-// with the current (old) Linux client, to log in to Tailscale on a Linux box.
-//
-// Deprecated: this will be deleted, to be replaced by cmd/tailscale.
-package main
-
-import (
-	"encoding/json"
-	"fmt"
-	"io/ioutil"
-	"log"
-	"os"
-
-	"github.com/pborman/getopt/v2"
-	"tailscale.com/atomicfile"
-	"tailscale.com/control/controlclient"
-	"tailscale.com/logpolicy"
-)
-
-func main() {
-	config := getopt.StringLong("config", 'f', "", "path to config file")
-	server := getopt.StringLong("server", 's', "https://login.tailscale.com", "URL to tailgate server")
-	getopt.Parse()
-	if len(getopt.Args()) > 0 {
-		log.Fatal("too many non-flag arguments")
-	}
-	if *config == "" {
-		log.Fatal("no --config file specified")
-	}
-	pol := logpolicy.New("tailnode.log.tailscale.io")
-	defer pol.Close()
-
-	cfg, err := loadConfig(*config)
-	if err != nil {
-		log.Fatal(err)
-	}
-
-	hi := controlclient.NewHostinfo()
-	hi.FrontendLogID = pol.PublicID.String()
-	hi.BackendLogID = pol.PublicID.String()
-
-	done := make(chan struct{}, 1)
-	c, err := controlclient.New(controlclient.Options{
-		Persist:   cfg,
-		ServerURL: *server,
-		Hostinfo:  &hi,
-	})
-	if err != nil {
-		log.Fatal(err)
-	}
-	c.SetStatusFunc(func(new controlclient.Status) {
-		if new.URL != "" {
-			fmt.Fprintf(os.Stderr, "To authenticate, visit:\n\n\t%s\n\n", new.URL)
-			return
-		}
-		if new.Err != "" {
-			log.Print(new.Err)
-			return
-		}
-		if new.Persist != nil {
-			if err := saveConfig(*config, *new.Persist); err != nil {
-				log.Println(err)
-			}
-		}
-		if new.NetMap != nil {
-			done <- struct{}{}
-		}
-	})
-	c.Login(nil, 0)
-	<-done
-	log.Printf("Success.\n")
-}
-
-func loadConfig(path string) (cfg controlclient.Persist, err error) {
-	b, err := ioutil.ReadFile(path)
-	if os.IsNotExist(err) {
-		log.Printf("config %s does not exist", path)
-		return controlclient.Persist{}, nil
-	}
-	if err := json.Unmarshal(b, &cfg); err != nil {
-		return controlclient.Persist{}, fmt.Errorf("load config: %v", err)
-	}
-	return cfg, nil
-}
-
-func saveConfig(path string, cfg controlclient.Persist) error {
-	b, err := json.MarshalIndent(cfg, "", "\t")
-	if err != nil {
-		return fmt.Errorf("save config: %v", err)
-	}
-	if err := atomicfile.WriteFile(path, b, 0666); err != nil {
-		return fmt.Errorf("save config: %v", err)
-	}
-	return nil
-}

cmd/tailscale/cli/cli.go

@@ -0,0 +1,123 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package cli contains the cmd/tailscale CLI code in a package that can be included
+// in other wrapper binaries such as the Mac and Windows clients.
+package cli
+
+import (
+	"context"
+	"flag"
+	"log"
+	"net"
+	"os"
+	"os/signal"
+	"runtime"
+	"strings"
+	"syscall"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/ipn"
+	"tailscale.com/paths"
+	"tailscale.com/safesocket"
+)
+
+// ActLikeCLI reports whether a GUI application should act like the
+// CLI based on os.Args, GOOS, the context the process is running in
+// (pty, parent PID), etc.
+func ActLikeCLI() bool {
+	if len(os.Args) < 2 {
+		return false
+	}
+	switch os.Args[1] {
+	case "up", "status", "netcheck", "version",
+		"-V", "--version", "-h", "--help":
+		return true
+	}
+	return false
+}
+
+// Run runs the CLI. The args do not include the binary name.
+func Run(args []string) error {
+	if len(args) == 1 && (args[0] == "-V" || args[0] == "--version") {
+		args = []string{"version"}
+	}
+
+	rootfs := flag.NewFlagSet("tailscale", flag.ExitOnError)
+	rootfs.StringVar(&rootArgs.socket, "socket", paths.DefaultTailscaledSocket(), "path to tailscaled's unix socket")
+
+	rootCmd := &ffcli.Command{
+		Name:       "tailscale",
+		ShortUsage: "tailscale subcommand [flags]",
+		ShortHelp:  "The easiest, most secure way to use WireGuard.",
+		LongHelp: strings.TrimSpace(`
+This CLI is still under active development. Commands and flags will
+change in the future.
+`),
+		Subcommands: []*ffcli.Command{
+			upCmd,
+			netcheckCmd,
+			statusCmd,
+			versionCmd,
+		},
+		FlagSet: rootfs,
+		Exec:    func(context.Context, []string) error { return flag.ErrHelp },
+	}
+
+	if err := rootCmd.Parse(args); err != nil {
+		return err
+	}
+
+	err := rootCmd.Run(context.Background())
+	if err == flag.ErrHelp {
+		return nil
+	}
+	return err
+}
+
+var rootArgs struct {
+	socket string
+}
+
+func connect(ctx context.Context) (net.Conn, *ipn.BackendClient, context.Context, context.CancelFunc) {
+	c, err := safesocket.Connect(rootArgs.socket, 41112)
+	if err != nil {
+		if runtime.GOOS != "windows" && rootArgs.socket == "" {
+			log.Fatalf("--socket cannot be empty")
+		}
+		log.Fatalf("Failed to connect to connect to tailscaled. (safesocket.Connect: %v)\n", err)
+	}
+	clientToServer := func(b []byte) {
+		ipn.WriteMsg(c, b)
+	}
+
+	ctx, cancel := context.WithCancel(ctx)
+
+	go func() {
+		interrupt := make(chan os.Signal, 1)
+		signal.Notify(interrupt, syscall.SIGINT, syscall.SIGTERM)
+		<-interrupt
+		c.Close()
+		cancel()
+	}()
+
+	bc := ipn.NewBackendClient(log.Printf, clientToServer)
+	return c, bc, ctx, cancel
+}
+
+// pump receives backend messages on conn and pushes them into bc.
+func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) {
+	defer conn.Close()
+	for ctx.Err() == nil {
+		msg, err := ipn.ReadMsg(conn)
+		if err != nil {
+			if ctx.Err() != nil {
+				return
+			}
+			log.Printf("ReadMsg: %v\n", err)
+			break
+		}
+		bc.GotNotifyMsg(msg)
+	}
+}

cmd/tailscale/cli/netcheck.go

@@ -0,0 +1,162 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"log"
+	"os"
+	"sort"
+	"strings"
+	"time"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/derp/derpmap"
+	"tailscale.com/net/dnscache"
+	"tailscale.com/net/netcheck"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
+)
+
+var netcheckCmd = &ffcli.Command{
+	Name:       "netcheck",
+	ShortUsage: "netcheck",
+	ShortHelp:  "Print an analysis of local network conditions",
+	Exec:       runNetcheck,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("netcheck", flag.ExitOnError)
+		fs.StringVar(&netcheckArgs.format, "format", "", `output format; empty (for human-readable), "json" or "json-line"`)
+		fs.DurationVar(&netcheckArgs.every, "every", 0, "if non-zero, do an incremental report with the given frequency")
+		fs.BoolVar(&netcheckArgs.verbose, "verbose", false, "verbose logs")
+		return fs
+	})(),
+}
+
+var netcheckArgs struct {
+	format  string
+	every   time.Duration
+	verbose bool
+}
+
+func runNetcheck(ctx context.Context, args []string) error {
+	c := &netcheck.Client{
+		DNSCache: dnscache.Get(),
+	}
+	if netcheckArgs.verbose {
+		c.Logf = logger.WithPrefix(log.Printf, "netcheck: ")
+		c.Verbose = true
+	} else {
+		c.Logf = logger.Discard
+	}
+
+	if strings.HasPrefix(netcheckArgs.format, "json") {
+		fmt.Fprintln(os.Stderr, "# Warning: this JSON format is not yet considered a stable interface")
+	}
+
+	dm := derpmap.Prod()
+	for {
+		t0 := time.Now()
+		report, err := c.GetReport(ctx, dm)
+		d := time.Since(t0)
+		if netcheckArgs.verbose {
+			c.Logf("GetReport took %v; err=%v", d.Round(time.Millisecond), err)
+		}
+		if err != nil {
+			log.Fatalf("netcheck: %v", err)
+		}
+		if err := printReport(dm, report); err != nil {
+			return err
+		}
+		if netcheckArgs.every == 0 {
+			return nil
+		}
+		time.Sleep(netcheckArgs.every)
+	}
+}
+
+func printReport(dm *tailcfg.DERPMap, report *netcheck.Report) error {
+	var j []byte
+	var err error
+	switch netcheckArgs.format {
+	case "":
+		break
+	case "json":
+		j, err = json.MarshalIndent(report, "", "\t")
+	case "json-line":
+		j, err = json.Marshal(report)
+	default:
+		return fmt.Errorf("unknown output format %q", netcheckArgs.format)
+	}
+	if err != nil {
+		return err
+	}
+	if j != nil {
+		j = append(j, '\n')
+		os.Stdout.Write(j)
+		return nil
+	}
+
+	fmt.Printf("\nReport:\n")
+	fmt.Printf("\t* UDP: %v\n", report.UDP)
+	if report.GlobalV4 != "" {
+		fmt.Printf("\t* IPv4: yes, %v\n", report.GlobalV4)
+	} else {
+		fmt.Printf("\t* IPv4: (no addr found)\n")
+	}
+	if report.GlobalV6 != "" {
+		fmt.Printf("\t* IPv6: yes, %v\n", report.GlobalV6)
+	} else if report.IPv6 {
+		fmt.Printf("\t* IPv6: (no addr found)\n")
+	} else {
+		fmt.Printf("\t* IPv6: no\n")
+	}
+	fmt.Printf("\t* MappingVariesByDestIP: %v\n", report.MappingVariesByDestIP)
+	fmt.Printf("\t* HairPinning: %v\n", report.HairPinning)
+	fmt.Printf("\t* PortMapping: %v\n", portMapping(report))
+
+	// When DERP latency checking failed,
+	// magicsock will try to pick the DERP server that
+	// most of your other nodes are also using
+	if len(report.RegionLatency) == 0 {
+		fmt.Printf("\t* Nearest DERP: unknown (no response to latency probes)\n")
+	} else {
+		fmt.Printf("\t* Nearest DERP: %v (%v)\n", report.PreferredDERP, dm.Regions[report.PreferredDERP].RegionCode)
+		fmt.Printf("\t* DERP latency:\n")
+		var rids []int
+		for rid := range dm.Regions {
+			rids = append(rids, rid)
+		}
+		sort.Ints(rids)
+		for _, rid := range rids {
+			d, ok := report.RegionLatency[rid]
+			var latency string
+			if ok {
+				latency = d.Round(time.Millisecond / 10).String()
+			}
+			fmt.Printf("\t\t- %v, %3s = %s\n", rid, dm.Regions[rid].RegionCode, latency)
+		}
+	}
+	return nil
+}
+
+func portMapping(r *netcheck.Report) string {
+	if !r.AnyPortMappingChecked() {
+		return "not checked"
+	}
+	var got []string
+	if r.UPnP.EqualBool(true) {
+		got = append(got, "UPnP")
+	}
+	if r.PMP.EqualBool(true) {
+		got = append(got, "NAT-PMP")
+	}
+	if r.PCP.EqualBool(true) {
+		got = append(got, "PCP")
+	}
+	return strings.Join(got, ", ")
+}

cmd/tailscale/cli/status.go

@@ -0,0 +1,172 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"log"
+	"net"
+	"net/http"
+	"os"
+	"time"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"github.com/toqueteos/webbrowser"
+	"tailscale.com/ipn"
+	"tailscale.com/ipn/ipnstate"
+	"tailscale.com/net/interfaces"
+)
+
+var statusCmd = &ffcli.Command{
+	Name:       "status",
+	ShortUsage: "status [-active] [-web] [-json]",
+	ShortHelp:  "Show state of tailscaled and its connections",
+	Exec:       runStatus,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("status", flag.ExitOnError)
+		fs.BoolVar(&statusArgs.json, "json", false, "output in JSON format (WARNING: format subject to change)")
+		fs.BoolVar(&statusArgs.web, "web", false, "run webserver with HTML showing status")
+		fs.BoolVar(&statusArgs.active, "active", false, "filter output to only peers with active sessions (not applicable to web mode)")
+		fs.StringVar(&statusArgs.listen, "listen", "127.0.0.1:8384", "listen address; use port 0 for automatic")
+		fs.BoolVar(&statusArgs.browser, "browser", true, "Open a browser in web mode")
+		return fs
+	})(),
+}
+
+var statusArgs struct {
+	json    bool   // JSON output mode
+	web     bool   // run webserver
+	listen  string // in web mode, webserver address to listen on, empty means auto
+	browser bool   // in web mode, whether to open browser
+	active  bool   // in CLI mode, filter output to only peers with active sessions
+}
+
+func runStatus(ctx context.Context, args []string) error {
+	c, bc, ctx, cancel := connect(ctx)
+	defer cancel()
+
+	bc.AllowVersionSkew = true
+
+	ch := make(chan *ipnstate.Status, 1)
+	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.ErrMessage != nil {
+			log.Fatal(*n.ErrMessage)
+		}
+		if n.Status != nil {
+			ch <- n.Status
+		}
+	})
+	go pump(ctx, bc, c)
+
+	getStatus := func() (*ipnstate.Status, error) {
+		bc.RequestStatus()
+		select {
+		case st := <-ch:
+			return st, nil
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		}
+	}
+	st, err := getStatus()
+	if err != nil {
+		return err
+	}
+	if statusArgs.json {
+		if statusArgs.active {
+			for peer, ps := range st.Peer {
+				if !peerActive(ps) {
+					delete(st.Peer, peer)
+				}
+			}
+		}
+		j, err := json.MarshalIndent(st, "", "  ")
+		if err != nil {
+			return err
+		}
+		fmt.Printf("%s", j)
+		return nil
+	}
+	if statusArgs.web {
+		ln, err := net.Listen("tcp", statusArgs.listen)
+		if err != nil {
+			return err
+		}
+		statusURL := interfaces.HTTPOfListener(ln)
+		fmt.Printf("Serving Tailscale status at %v ...\n", statusURL)
+		go func() {
+			<-ctx.Done()
+			ln.Close()
+		}()
+		if statusArgs.browser {
+			go webbrowser.Open(statusURL)
+		}
+		err = http.Serve(ln, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			if r.RequestURI != "/" {
+				http.NotFound(w, r)
+				return
+			}
+			st, err := getStatus()
+			if err != nil {
+				http.Error(w, err.Error(), 500)
+				return
+			}
+			w.Header().Set("Content-Type", "text/html; charset=utf-8")
+			st.WriteHTML(w)
+		}))
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+		return err
+	}
+
+	var buf bytes.Buffer
+	f := func(format string, a ...interface{}) { fmt.Fprintf(&buf, format, a...) }
+	for _, peer := range st.Peers() {
+		ps := st.Peer[peer]
+		active := peerActive(ps)
+		if statusArgs.active && !active {
+			continue
+		}
+		f("%s %-7s %-15s %-18s tx=%8d rx=%8d ",
+			peer.ShortString(),
+			ps.OS,
+			ps.TailAddr,
+			ps.SimpleHostName(),
+			ps.TxBytes,
+			ps.RxBytes,
+		)
+		relay := ps.Relay
+		if active && relay != "" && ps.CurAddr == "" {
+			relay = "*" + relay + "*"
+		} else {
+			relay = " " + relay
+		}
+		f("%-6s", relay)
+		for i, addr := range ps.Addrs {
+			if i != 0 {
+				f(", ")
+			}
+			if addr == ps.CurAddr {
+				f("*%s*", addr)
+			} else {
+				f("%s", addr)
+			}
+		}
+		f("\n")
+	}
+	os.Stdout.Write(buf.Bytes())
+	return nil
+}
+
+// peerActive reports whether ps has recent activity.
+//
+// TODO: have the server report this bool instead.
+func peerActive(ps *ipnstate.PeerStatus) bool {
+	return !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
+}

cmd/tailscale/cli/up.go

@@ -0,0 +1,252 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"bytes"
+	"context"
+	"flag"
+	"fmt"
+	"log"
+	"os"
+	"os/exec"
+	"runtime"
+	"strconv"
+	"strings"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"github.com/tailscale/wireguard-go/wgcfg"
+	"tailscale.com/ipn"
+	"tailscale.com/tailcfg"
+	"tailscale.com/wgengine/router"
+)
+
+// globalStateKey is the ipn.StateKey that tailscaled loads on
+// startup.
+//
+// We have to support multiple state keys for other OSes (Windows in
+// particular), but right now Unix daemons run with a single
+// node-global state. To keep open the option of having per-user state
+// later, the global state key doesn't look like a username.
+const globalStateKey = "_daemon"
+
+var upCmd = &ffcli.Command{
+	Name:       "up",
+	ShortUsage: "up [flags]",
+	ShortHelp:  "Connect to your Tailscale network",
+
+	LongHelp: strings.TrimSpace(`
+"tailscale up" connects this machine to your Tailscale network,
+triggering authentication if necessary.
+
+The flags passed to this command are specific to this machine. If you don't
+specify any flags, options are reset to their default.
+`),
+	FlagSet: (func() *flag.FlagSet {
+		upf := flag.NewFlagSet("up", flag.ExitOnError)
+		upf.StringVar(&upArgs.server, "login-server", "https://login.tailscale.com", "base URL of control server")
+		upf.BoolVar(&upArgs.acceptRoutes, "accept-routes", false, "accept routes advertised by other Tailscale nodes")
+		upf.BoolVar(&upArgs.acceptDNS, "accept-dns", true, "accept DNS configuration from the admin panel")
+		upf.BoolVar(&upArgs.singleRoutes, "host-routes", true, "install host routes to other Tailscale nodes")
+		upf.BoolVar(&upArgs.shieldsUp, "shields-up", false, "don't allow incoming connections")
+		upf.StringVar(&upArgs.advertiseTags, "advertise-tags", "", "ACL tags to request (comma-separated, e.g. eng,montreal,ssh)")
+		upf.StringVar(&upArgs.authKey, "authkey", "", "node authorization key")
+		upf.StringVar(&upArgs.hostname, "hostname", "", "hostname to use instead of the one provided by the OS")
+		upf.BoolVar(&upArgs.enableDERP, "enable-derp", true, "enable the use of DERP servers")
+		if runtime.GOOS == "linux" || isBSD(runtime.GOOS) {
+			upf.StringVar(&upArgs.advertiseRoutes, "advertise-routes", "", "routes to advertise to other nodes (comma-separated, e.g. 10.0.0.0/8,192.168.0.0/24)")
+		}
+		if runtime.GOOS == "linux" {
+			upf.BoolVar(&upArgs.snat, "snat-subnet-routes", true, "source NAT traffic to local routes advertised with -advertise-routes")
+			upf.StringVar(&upArgs.netfilterMode, "netfilter-mode", "on", "netfilter mode (one of on, nodivert, off)")
+		}
+		return upf
+	})(),
+	Exec: runUp,
+}
+
+var upArgs struct {
+	server          string
+	acceptRoutes    bool
+	acceptDNS       bool
+	singleRoutes    bool
+	shieldsUp       bool
+	advertiseRoutes string
+	advertiseTags   string
+	enableDERP      bool
+	snat            bool
+	netfilterMode   string
+	authKey         string
+	hostname        string
+}
+
+// parseIPOrCIDR parses an IP address or a CIDR prefix. If the input
+// is an IP address, it is returned in CIDR form with a /32 mask for
+// IPv4 or a /128 mask for IPv6.
+func parseIPOrCIDR(s string) (wgcfg.CIDR, bool) {
+	if strings.Contains(s, "/") {
+		ret, err := wgcfg.ParseCIDR(s)
+		if err != nil {
+			return wgcfg.CIDR{}, false
+		}
+		return ret, true
+	}
+
+	ip, ok := wgcfg.ParseIP(s)
+	if !ok {
+		return wgcfg.CIDR{}, false
+	}
+	if ip.Is4() {
+		return wgcfg.CIDR{IP: ip, Mask: 32}, true
+	} else {
+		return wgcfg.CIDR{IP: ip, Mask: 128}, true
+	}
+}
+
+func isBSD(s string) bool {
+	return s == "dragonfly" || s == "freebsd" || s == "netbsd" || s == "openbsd"
+}
+
+func warning(format string, args ...interface{}) {
+	fmt.Printf("Warning: "+format+"\n", args...)
+}
+
+// checkIPForwarding prints warnings on linux if IP forwarding is not
+// enabled, or if we were unable to verify the state of IP forwarding.
+func checkIPForwarding() {
+	var key string
+
+	if runtime.GOOS == "linux" {
+		key = "net.ipv4.ip_forward"
+	} else if isBSD(runtime.GOOS) {
+		key = "net.inet.ip.forwarding"
+	} else {
+		return
+	}
+
+	bs, err := exec.Command("sysctl", "-n", key).Output()
+	if err != nil {
+		warning("couldn't check %s (%v).\nSubnet routes won't work without IP forwarding.", key, err)
+		return
+	}
+	on, err := strconv.ParseBool(string(bytes.TrimSpace(bs)))
+	if err != nil {
+		warning("couldn't parse %s (%v).\nSubnet routes won't work without IP forwarding.", key, err)
+		return
+	}
+	if !on {
+		warning("%s is disabled. Subnet routes won't work.", key)
+	}
+}
+
+func runUp(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		log.Fatalf("too many non-flag arguments: %q", args)
+	}
+
+	var routes []wgcfg.CIDR
+	if upArgs.advertiseRoutes != "" {
+		checkIPForwarding()
+		advroutes := strings.Split(upArgs.advertiseRoutes, ",")
+		for _, s := range advroutes {
+			cidr, ok := parseIPOrCIDR(s)
+			if !ok {
+				log.Fatalf("%q is not a valid IP address or CIDR prefix", s)
+			}
+			routes = append(routes, cidr)
+		}
+	}
+
+	var tags []string
+	if upArgs.advertiseTags != "" {
+		tags = strings.Split(upArgs.advertiseTags, ",")
+		for _, tag := range tags {
+			err := tailcfg.CheckTag(tag)
+			if err != nil {
+				log.Fatalf("tag: %q: %s", tag, err)
+			}
+		}
+	}
+
+	if len(upArgs.hostname) > 256 {
+		log.Fatalf("hostname too long: %d bytes (max 256)", len(upArgs.hostname))
+	}
+
+	// TODO(apenwarr): fix different semantics between prefs and uflags
+	// TODO(apenwarr): allow setting/using CorpDNS
+	prefs := ipn.NewPrefs()
+	prefs.ControlURL = upArgs.server
+	prefs.WantRunning = true
+	prefs.RouteAll = upArgs.acceptRoutes
+	prefs.CorpDNS = upArgs.acceptDNS
+	prefs.AllowSingleHosts = upArgs.singleRoutes
+	prefs.ShieldsUp = upArgs.shieldsUp
+	prefs.AdvertiseRoutes = routes
+	prefs.AdvertiseTags = tags
+	prefs.NoSNAT = !upArgs.snat
+	prefs.DisableDERP = !upArgs.enableDERP
+	prefs.Hostname = upArgs.hostname
+	if runtime.GOOS == "linux" {
+		switch upArgs.netfilterMode {
+		case "on":
+			prefs.NetfilterMode = router.NetfilterOn
+		case "nodivert":
+			prefs.NetfilterMode = router.NetfilterNoDivert
+			warning("netfilter=nodivert; add iptables calls to ts-* chains manually.")
+		case "off":
+			prefs.NetfilterMode = router.NetfilterOff
+			warning("netfilter=off; configure iptables yourself.")
+		default:
+			log.Fatalf("invalid value --netfilter-mode: %q", upArgs.netfilterMode)
+		}
+	}
+
+	c, bc, ctx, cancel := connect(ctx)
+	defer cancel()
+
+	var printed bool
+
+	bc.SetPrefs(prefs)
+	opts := ipn.Options{
+		StateKey: globalStateKey,
+		AuthKey:  upArgs.authKey,
+		Notify: func(n ipn.Notify) {
+			if n.ErrMessage != nil {
+				log.Fatalf("backend error: %v\n", *n.ErrMessage)
+			}
+			if s := n.State; s != nil {
+				switch *s {
+				case ipn.NeedsLogin:
+					printed = true
+					bc.StartLoginInteractive()
+				case ipn.NeedsMachineAuth:
+					printed = true
+					fmt.Fprintf(os.Stderr, "\nTo authorize your machine, visit (as admin):\n\n\t%s/admin/machines\n\n", upArgs.server)
+				case ipn.Starting, ipn.Running:
+					// Done full authentication process
+					if printed {
+						// Only need to print an update if we printed the "please click" message earlier.
+						fmt.Fprintf(os.Stderr, "Success.\n")
+					}
+					cancel()
+				}
+			}
+			if url := n.BrowseToURL; url != nil {
+				fmt.Fprintf(os.Stderr, "\nTo authenticate, visit:\n\n\t%s\n\n", *url)
+			}
+		},
+	}
+	// We still have to Start right now because it's the only way to
+	// set up notifications and whatnot. This causes a bunch of churn
+	// every time the CLI touches anything.
+	//
+	// TODO(danderson): redo the frontend/backend API to assume
+	// ephemeral frontends that read/modify/write state, once
+	// Windows/Mac state is moved into backend.
+	bc.Start(opts)
+	pump(ctx, bc, c)
+
+	return nil
+}

cmd/tailscale/cli/version.go

@@ -0,0 +1,69 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cli
+
+import (
+	"context"
+	"flag"
+	"fmt"
+	"log"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/ipn"
+	"tailscale.com/version"
+)
+
+var versionCmd = &ffcli.Command{
+	Name:       "version",
+	ShortUsage: "version [flags]",
+	ShortHelp:  "Print Tailscale version",
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("version", flag.ExitOnError)
+		fs.BoolVar(&versionArgs.daemon, "daemon", false, "also print local node's daemon version")
+		return fs
+	})(),
+	Exec: runVersion,
+}
+
+var versionArgs struct {
+	daemon bool // also check local node's daemon version
+}
+
+func runVersion(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		log.Fatalf("too many non-flag arguments: %q", args)
+	}
+	if !versionArgs.daemon {
+		fmt.Println(version.LONG)
+		return nil
+	}
+	fmt.Printf("Client: %s\n", version.LONG)
+
+	c, bc, ctx, cancel := connect(ctx)
+	defer cancel()
+
+	bc.AllowVersionSkew = true
+
+	done := make(chan struct{})
+
+	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.ErrMessage != nil {
+			log.Fatal(*n.ErrMessage)
+		}
+		if n.Status != nil {
+			fmt.Printf("Daemon: %s\n", n.Version)
+			close(done)
+		}
+	})
+	go pump(ctx, bc, c)
+
+	bc.RequestStatus()
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

cmd/tailscale/tailscale.go

@@ -7,134 +7,22 @@
 package main // import "tailscale.com/cmd/tailscale"
 
 import (
-	"context"
 	"fmt"
 	"log"
-	"net"
 	"os"
-	"os/signal"
-	"syscall"
 
 	"github.com/apenwarr/fixconsole"
-	"github.com/pborman/getopt/v2"
-	"github.com/tailscale/wireguard-go/wgcfg"
-	"tailscale.com/ipn"
-	"tailscale.com/logpolicy"
-	"tailscale.com/safesocket"
+	"tailscale.com/cmd/tailscale/cli"
 )
 
-// globalStateKey is the ipn.StateKey that tailscaled loads on
-// startup.
-//
-// We have to support multiple state keys for other OSes (Windows in
-// particular), but right now Unix daemons run with a single
-// node-global state. To keep open the option of having per-user state
-// later, the global state key doesn't look like a username.
-const globalStateKey = "_daemon"
-
-// pump receives backend messages on conn and pushes them into bc.
-func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) {
-	defer log.Printf("Control connection done.\n")
-	defer conn.Close()
-	for ctx.Err() == nil {
-		msg, err := ipn.ReadMsg(conn)
-		if err != nil {
-			log.Printf("ReadMsg: %v\n", err)
-			break
-		}
-		bc.GotNotifyMsg(msg)
-	}
-}
-
 func main() {
 	err := fixconsole.FixConsoleIfNeeded()
 	if err != nil {
 		log.Printf("fixConsoleOutput: %v\n", err)
 	}
 
-	socket := getopt.StringLong("socket", 0, "/run/tailscale/tailscaled.sock", "path of tailscaled's unix socket")
-	server := getopt.StringLong("server", 's', "https://login.tailscale.com", "URL to tailcontrol server")
-	nuroutes := getopt.BoolLong("no-single-routes", 'N', "disallow (non-subnet) routes to single nodes")
-	routeall := getopt.BoolLong("remote-routes", 'R', "accept routes advertised by remote nodes")
-	nopf := getopt.BoolLong("no-packet-filter", 'F', "disable packet filter")
-	advroutes := getopt.ListLong("routes", 'r', "routes to advertise to other nodes (comma-separated, e.g. 10.0.0.0/8,192.168.1.0/24)")
-	getopt.Parse()
-	pol := logpolicy.New("tailnode.log.tailscale.io")
-	if len(getopt.Args()) > 0 {
-		log.Fatalf("too many non-flag arguments: %#v", getopt.Args()[0])
+	if err := cli.Run(os.Args[1:]); err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
 	}
-
-	defer pol.Close()
-
-	var adv []wgcfg.CIDR
-	for _, s := range *advroutes {
-		cidr, err := wgcfg.ParseCIDR(s)
-		if err != nil {
-			log.Fatalf("%q is not a valid CIDR prefix: %v", s, err)
-		}
-		adv = append(adv, *cidr)
-	}
-
-	// TODO(apenwarr): fix different semantics between prefs and uflags
-	// TODO(apenwarr): allow setting/using CorpDNS
-	prefs := ipn.NewPrefs()
-	prefs.ControlURL = *server
-	prefs.WantRunning = true
-	prefs.RouteAll = *routeall
-	prefs.AllowSingleHosts = !*nuroutes
-	prefs.UsePacketFilter = !*nopf
-	prefs.AdvertiseRoutes = adv
-
-	c, err := safesocket.Connect(*socket, 0)
-	if err != nil {
-		log.Fatalf("safesocket.Connect: %v\n", err)
-	}
-	clientToServer := func(b []byte) {
-		ipn.WriteMsg(c, b)
-	}
-
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-
-	go func() {
-		interrupt := make(chan os.Signal, 1)
-		signal.Notify(interrupt, syscall.SIGINT, syscall.SIGTERM)
-		<-interrupt
-		c.Close()
-	}()
-
-	bc := ipn.NewBackendClient(log.Printf, clientToServer)
-	bc.SetPrefs(prefs)
-	opts := ipn.Options{
-		StateKey: globalStateKey,
-		Notify: func(n ipn.Notify) {
-			if n.ErrMessage != nil {
-				log.Fatalf("backend error: %v\n", *n.ErrMessage)
-			}
-			if s := n.State; s != nil {
-				switch *s {
-				case ipn.NeedsLogin:
-					bc.StartLoginInteractive()
-				case ipn.NeedsMachineAuth:
-					fmt.Fprintf(os.Stderr, "\nTo authorize your machine, visit (as admin):\n\n\t%s/admin/machines\n\n", *server)
-				case ipn.Starting, ipn.Running:
-					// Done full authentication process
-					fmt.Fprintf(os.Stderr, "\ntailscaled is authenticated, nothing more to do.\n\n")
-					cancel()
-				}
-			}
-			if url := n.BrowseToURL; url != nil {
-				fmt.Fprintf(os.Stderr, "\nTo authenticate, visit:\n\n\t%s\n\n", *url)
-			}
-		},
-	}
-	// We still have to Start right now because it's the only way to
-	// set up notifications and whatnot. This causes a bunch of churn
-	// every time the CLI touches anything.
-	//
-	// TODO(danderson): redo the frontend/backend API to assume
-	// ephemeral frontends that read/modify/write state, once
-	// Windows/Mac state is moved into backend.
-	bc.Start(opts)
-	pump(ctx, bc, c)
 }

cmd/tailscaled/tailscaled.defaults

@@ -4,5 +4,5 @@
 # settings.
 PORT="41641"
 
-# Extra flags you might want to pass to relaynode.
+# Extra flags you might want to pass to tailscaled.
 FLAGS=""

cmd/tailscaled/tailscaled.go

@@ -14,13 +14,22 @@ import (
 	"log"
 	"net/http"
 	"net/http/pprof"
+	"os"
+	"os/signal"
+	"runtime"
+	"runtime/debug"
+	"syscall"
+	"time"
 
 	"github.com/apenwarr/fixconsole"
 	"github.com/pborman/getopt/v2"
 	"tailscale.com/ipn/ipnserver"
 	"tailscale.com/logpolicy"
+	"tailscale.com/paths"
+	"tailscale.com/types/logger"
 	"tailscale.com/wgengine"
 	"tailscale.com/wgengine/magicsock"
+	"tailscale.com/wgengine/router"
 )
 
 // globalStateKey is the ipn.StateKey that tailscaled loads on
@@ -32,78 +41,156 @@ import (
 // later, the global state key doesn't look like a username.
 const globalStateKey = "_daemon"
 
-func main() {
-	fake := getopt.BoolLong("fake", 0, "fake tunnel+routing instead of tuntap")
-	debug := getopt.StringLong("debug", 0, "", "Address of debug server")
-	tunname := getopt.StringLong("tun", 0, "tailscale0", "tunnel interface name")
-	listenport := getopt.Uint16Long("port", 'p', magicsock.DefaultPort, "WireGuard port (0=autoselect)")
-	statepath := getopt.StringLong("state", 0, "", "Path of state file")
-	socketpath := getopt.StringLong("socket", 's', "tailscaled.sock", "Path of the service unix socket")
+// defaultTunName returns the default tun device name for the platform.
+func defaultTunName() string {
+	switch runtime.GOOS {
+	case "openbsd":
+		return "tun"
+	case "windows":
+		return "Tailscale"
+	}
+	return "tailscale0"
+}
 
-	logf := wgengine.RusagePrefixLog(log.Printf)
+var args struct {
+	cleanup    bool
+	fake       bool
+	debug      string
+	tunname    string
+	port       uint16
+	statepath  string
+	socketpath string
+}
+
+func main() {
+	// We aren't very performance sensitive, and the parts that are
+	// performance sensitive (wireguard) try hard not to do any memory
+	// allocations. So let's be aggressive about garbage collection,
+	// unless the user specifically overrides it in the usual way.
+	if _, ok := os.LookupEnv("GOGC"); !ok {
+		debug.SetGCPercent(10)
+	}
+
+	// Set default values for getopt.
+	args.tunname = defaultTunName()
+	args.port = magicsock.DefaultPort
+	args.statepath = paths.DefaultTailscaledStateFile()
+	args.socketpath = paths.DefaultTailscaledSocket()
+
+	getopt.FlagLong(&args.cleanup, "cleanup", 0, "clean up system state and exit")
+	getopt.FlagLong(&args.fake, "fake", 0, "fake tunnel+routing instead of tuntap")
+	getopt.FlagLong(&args.debug, "debug", 0, "address of debug server")
+	getopt.FlagLong(&args.tunname, "tun", 0, "tunnel interface name")
+	getopt.FlagLong(&args.port, "port", 'p', "WireGuard port (0=autoselect)")
+	getopt.FlagLong(&args.statepath, "state", 0, "path of state file")
+	getopt.FlagLong(&args.socketpath, "socket", 's', "path of the service unix socket")
 
 	err := fixconsole.FixConsoleIfNeeded()
 	if err != nil {
-		logf("fixConsoleOutput: %v\n", err)
+		log.Fatalf("fixConsoleOutput: %v", err)
 	}
-	pol := logpolicy.New("tailnode.log.tailscale.io")
 
 	getopt.Parse()
 	if len(getopt.Args()) > 0 {
 		log.Fatalf("too many non-flag arguments: %#v", getopt.Args()[0])
 	}
 
-	if *statepath == "" {
+	if args.statepath == "" {
 		log.Fatalf("--state is required")
 	}
 
-	if *socketpath == "" {
+	if args.socketpath == "" && runtime.GOOS != "windows" {
 		log.Fatalf("--socket is required")
 	}
 
-	if *debug != "" {
-		go runDebugServer(*debug)
+	if err := run(); err != nil {
+		// No need to log; the func already did
+		os.Exit(1)
+	}
+}
+
+func run() error {
+	var err error
+
+	pol := logpolicy.New("tailnode.log.tailscale.io")
+	defer func() {
+		// Finish uploading logs after closing everything else.
+		ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+		defer cancel()
+		pol.Shutdown(ctx)
+	}()
+
+	logf := wgengine.RusagePrefixLog(log.Printf)
+	logf = logger.RateLimitedFn(logf, 5*time.Second, 5, 100)
+
+	if args.cleanup {
+		router.Cleanup(logf, args.tunname)
+		return nil
+	}
+
+	var debugMux *http.ServeMux
+	if args.debug != "" {
+		debugMux = newDebugMux()
+		go runDebugServer(debugMux, args.debug)
 	}
 
 	var e wgengine.Engine
-	if *fake {
+	if args.fake {
 		e, err = wgengine.NewFakeUserspaceEngine(logf, 0)
 	} else {
-		e, err = wgengine.NewUserspaceEngine(logf, *tunname, *listenport)
+		e, err = wgengine.NewUserspaceEngine(logf, args.tunname, args.port)
 	}
 	if err != nil {
-		log.Fatalf("wgengine.New: %v\n", err)
+		logf("wgengine.New: %v", err)
+		return err
 	}
 	e = wgengine.NewWatchdog(e)
 
+	ctx, cancel := context.WithCancel(context.Background())
+	// Exit gracefully by cancelling the ipnserver context in most common cases:
+	// interrupted from the TTY or killed by a service manager.
+	go func() {
+		interrupt := make(chan os.Signal, 1)
+		signal.Notify(interrupt, syscall.SIGINT, syscall.SIGTERM)
+		select {
+		case <-interrupt:
+			cancel()
+		case <-ctx.Done():
+			// continue
+		}
+	}()
+
 	opts := ipnserver.Options{
-		SocketPath:         *socketpath,
-		StatePath:          *statepath,
+		SocketPath:         args.socketpath,
+		Port:               41112,
+		StatePath:          args.statepath,
 		AutostartStateKey:  globalStateKey,
-		LegacyConfigPath:   "/var/lib/tailscale/relay.conf",
+		LegacyConfigPath:   paths.LegacyConfigPath(),
 		SurviveDisconnects: true,
+		DebugMux:           debugMux,
 	}
-	err = ipnserver.Run(context.Background(), logf, pol.PublicID.String(), opts, e)
-	if err != nil {
-		log.Fatalf("tailscaled: %v\n", err)
+	err = ipnserver.Run(ctx, logf, pol.PublicID.String(), opts, e)
+	// Cancelation is not an error: it is the only way to stop ipnserver.
+	if err != nil && err != context.Canceled {
+		logf("ipnserver.Run: %v", err)
+		return err
 	}
 
-	// TODO(crawshaw): It would be nice to start a timeout context the moment a signal
-	// is received and use that timeout to give us a moment to finish uploading logs
-	// here. But the signal is handled inside ipnserver.Run, so some plumbing is needed.
-	ctx, cancel := context.WithCancel(context.Background())
-	cancel()
-	pol.Shutdown(ctx)
+	return nil
 }
 
-func runDebugServer(addr string) {
+func newDebugMux() *http.ServeMux {
 	mux := http.NewServeMux()
 	mux.HandleFunc("/debug/pprof/", pprof.Index)
 	mux.HandleFunc("/debug/pprof/cmdline", pprof.Cmdline)
 	mux.HandleFunc("/debug/pprof/profile", pprof.Profile)
 	mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
 	mux.HandleFunc("/debug/pprof/trace", pprof.Trace)
-	srv := http.Server{
+	return mux
+}
+
+func runDebugServer(mux *http.ServeMux, addr string) {
+	srv := &http.Server{
 		Addr:    addr,
 		Handler: mux,
 	}

cmd/tailscaled/tailscaled.service

@@ -9,6 +9,7 @@ StartLimitBurst=0
 [Service]
 EnvironmentFile=/etc/default/tailscaled
 ExecStart=/usr/sbin/tailscaled --state=/var/lib/tailscale/tailscaled.state --socket=/run/tailscale/tailscaled.sock --port $PORT $FLAGS
+ExecStopPost=/usr/sbin/tailscaled --cleanup
 
 Restart=on-failure
 
@@ -16,8 +17,8 @@ RuntimeDirectory=tailscale
 RuntimeDirectoryMode=0755
 StateDirectory=tailscale
 StateDirectoryMode=0750
-User=root
-Group=root
+CacheDirectory=tailscale
+CacheDirectoryMode=0750
 
 [Install]
 WantedBy=multi-user.target

cmd/tsshd/tsshd.go

@@ -32,7 +32,7 @@ import (
 	"github.com/gliderlabs/ssh"
 	"github.com/kr/pty"
 	gossh "golang.org/x/crypto/ssh"
-	"tailscale.com/interfaces"
+	"tailscale.com/net/interfaces"
 )
 
 var (

control/controlclient/auto.go

@@ -17,43 +17,46 @@ import (
 	"sync"
 	"time"
 
+	"github.com/tailscale/wireguard-go/wgcfg"
 	"golang.org/x/oauth2"
 	"tailscale.com/logtail/backoff"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/empty"
 	"tailscale.com/types/logger"
+	"tailscale.com/types/structs"
 )
 
-// TODO(apenwarr): eliminate the 'state' variable, as it's now obsolete.
-//  It's used only by the unit tests.
-type state int
+// State is the high-level state of the client. It is used only in
+// unit tests for proper sequencing, don't depend on it anywhere else.
+// TODO(apenwarr): eliminate 'state', as it's now obsolete.
+type State int
 
 const (
-	stateNew = state(iota)
-	stateNotAuthenticated
-	stateAuthenticating
-	stateURLVisitRequired
-	stateAuthenticated
-	stateSynchronized // connected and received map update
+	StateNew = State(iota)
+	StateNotAuthenticated
+	StateAuthenticating
+	StateURLVisitRequired
+	StateAuthenticated
+	StateSynchronized // connected and received map update
 )
 
-func (s state) MarshalText() ([]byte, error) {
+func (s State) MarshalText() ([]byte, error) {
 	return []byte(s.String()), nil
 }
 
-func (s state) String() string {
+func (s State) String() string {
 	switch s {
-	case stateNew:
+	case StateNew:
 		return "state:new"
-	case stateNotAuthenticated:
+	case StateNotAuthenticated:
 		return "state:not-authenticated"
-	case stateAuthenticating:
+	case StateAuthenticating:
 		return "state:authenticating"
-	case stateURLVisitRequired:
+	case StateURLVisitRequired:
 		return "state:url-visit-required"
-	case stateAuthenticated:
+	case StateAuthenticated:
 		return "state:authenticated"
-	case stateSynchronized:
+	case StateSynchronized:
 		return "state:synchronized"
 	default:
 		return fmt.Sprintf("state:unknown:%d", int(s))
@@ -61,13 +64,14 @@ func (s state) String() string {
 }
 
 type Status struct {
+	_             structs.Incomparable
 	LoginFinished *empty.Message
 	Err           string
 	URL           string
-	Persist       *Persist         // locally persisted configuration
-	NetMap        *NetworkMap      // server-pushed configuration
-	Hostinfo      tailcfg.Hostinfo // current Hostinfo data
-	state         state
+	Persist       *Persist          // locally persisted configuration
+	NetMap        *NetworkMap       // server-pushed configuration
+	Hostinfo      *tailcfg.Hostinfo // current Hostinfo data
+	State         State
 }
 
 // Equal reports whether s and s2 are equal.
@@ -82,7 +86,7 @@ func (s *Status) Equal(s2 *Status) bool {
 		reflect.DeepEqual(s.Persist, s2.Persist) &&
 		reflect.DeepEqual(s.NetMap, s2.NetMap) &&
 		reflect.DeepEqual(s.Hostinfo, s2.Hostinfo) &&
-		s.state == s2.state
+		s.State == s2.State
 }
 
 func (s Status) String() string {
@@ -90,10 +94,11 @@ func (s Status) String() string {
 	if err != nil {
 		panic(err)
 	}
-	return s.state.String() + " " + string(b)
+	return s.State.String() + " " + string(b)
 }
 
 type LoginGoal struct {
+	_            structs.Incomparable
 	wantLoggedIn bool          // true if we *want* to be logged in
 	token        *oauth2.Token // oauth token to use when logging in
 	flags        LoginFlags    // flags to use when logging in
@@ -115,10 +120,10 @@ type Client struct {
 	loggedIn     bool       // true if currently logged in
 	loginGoal    *LoginGoal // non-nil if some login activity is desired
 	synced       bool       // true if our netmap is up-to-date
-	hostinfo     tailcfg.Hostinfo
+	hostinfo     *tailcfg.Hostinfo
 	inPollNetMap bool // true if currently running a PollNetMap
 	inSendStatus int  // number of sendStatus calls currently in progress
-	state        state
+	state        State
 
 	authCtx    context.Context // context used for auth requests
 	mapCtx     context.Context // context used for netmap requests
@@ -147,6 +152,9 @@ func NewNoStart(opts Options) (*Client, error) {
 	if opts.Logf == nil {
 		opts.Logf = func(fmt string, args ...interface{}) {}
 	}
+	if opts.TimeNow == nil {
+		opts.TimeNow = time.Now
+	}
 	c := &Client{
 		direct:   direct,
 		timeNow:  opts.TimeNow,
@@ -199,7 +207,7 @@ func (c *Client) cancelMapSafely() {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
-	c.logf("cancelMapSafely: synced=%v\n", c.synced)
+	c.logf("cancelMapSafely: synced=%v", c.synced)
 
 	if c.inPollNetMap {
 		// received at least one netmap since the last
@@ -221,23 +229,23 @@ func (c *Client) cancelMapSafely() {
 		// request.
 		select {
 		case c.newMapCh <- struct{}{}:
-			c.logf("cancelMapSafely: wrote to channel\n")
+			c.logf("cancelMapSafely: wrote to channel")
 		default:
 			// if channel write failed, then there was already
 			// an outstanding newMapCh request. One is enough,
 			// since it'll always use the latest endpoints.
-			c.logf("cancelMapSafely: channel was full\n")
+			c.logf("cancelMapSafely: channel was full")
 		}
 	}
 }
 
 func (c *Client) authRoutine() {
 	defer close(c.authDone)
-	bo := backoff.Backoff{Name: "authRoutine"}
+	bo := backoff.NewBackoff("authRoutine", c.logf)
 
 	for {
 		c.mu.Lock()
-		c.logf("authRoutine: %s\n", c.state)
+		c.logf("authRoutine: %s", c.state)
 		expiry := c.expiry
 		goal := c.loginGoal
 		ctx := c.authCtx
@@ -246,13 +254,13 @@ func (c *Client) authRoutine() {
 
 		select {
 		case <-c.quit:
-			c.logf("authRoutine: quit\n")
+			c.logf("authRoutine: quit")
 			return
 		default:
 		}
 
 		report := func(err error, msg string) {
-			c.logf("%s: %v\n", msg, err)
+			c.logf("%s: %v", msg, err)
 			err = fmt.Errorf("%s: %v", msg, err)
 			// don't send status updates for context errors,
 			// since context cancelation is always on purpose.
@@ -281,14 +289,14 @@ func (c *Client) authRoutine() {
 			}
 			select {
 			case <-ctx.Done():
-				c.logf("authRoutine: context done.\n")
+				c.logf("authRoutine: context done.")
 			case <-exp:
 				// Unfortunately the key expiry isn't provided
 				// by the control server until mapRequest.
 				// So we have to do some hackery with c.expiry
 				// in here.
 				// TODO(apenwarr): add a key expiry field in RegisterResponse.
-				c.logf("authRoutine: key expiration check.\n")
+				c.logf("authRoutine: key expiration check.")
 				if synced && expiry != nil && !expiry.IsZero() && expiry.Before(c.timeNow()) {
 					c.logf("Key expired; setting loggedIn=false.")
 
@@ -313,7 +321,7 @@ func (c *Client) authRoutine() {
 			c.mu.Lock()
 			c.loggedIn = false
 			c.loginGoal = nil
-			c.state = stateNotAuthenticated
+			c.state = StateNotAuthenticated
 			c.synced = false
 			c.mu.Unlock()
 
@@ -322,9 +330,9 @@ func (c *Client) authRoutine() {
 		} else { // ie. goal.wantLoggedIn
 			c.mu.Lock()
 			if goal.url != "" {
-				c.state = stateURLVisitRequired
+				c.state = StateURLVisitRequired
 			} else {
-				c.state = stateAuthenticating
+				c.state = StateAuthenticating
 			}
 			c.mu.Unlock()
 
@@ -347,13 +355,14 @@ func (c *Client) authRoutine() {
 					err = fmt.Errorf("weird: server required a new url?")
 					report(err, "WaitLoginURL")
 				}
-				goal.url = url
-				goal.token = nil
-				goal.flags = LoginDefault
 
 				c.mu.Lock()
-				c.loginGoal = goal
-				c.state = stateURLVisitRequired
+				c.loginGoal = &LoginGoal{
+					wantLoggedIn: true,
+					flags:        LoginDefault,
+					url:          url,
+				}
+				c.state = StateURLVisitRequired
 				c.synced = false
 				c.mu.Unlock()
 
@@ -366,7 +375,7 @@ func (c *Client) authRoutine() {
 			c.mu.Lock()
 			c.loggedIn = true
 			c.loginGoal = nil
-			c.state = stateAuthenticated
+			c.state = StateAuthenticated
 			c.mu.Unlock()
 
 			c.sendStatus("authRoutine4", nil, "", nil)
@@ -376,26 +385,40 @@ func (c *Client) authRoutine() {
 	}
 }
 
+// Expiry returns the credential expiration time, or the zero time if
+// the expiration time isn't known. Used in tests only.
+func (c *Client) Expiry() *time.Time {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.expiry
+}
+
+// Direct returns the underlying direct client object. Used in tests
+// only.
+func (c *Client) Direct() *Direct {
+	return c.direct
+}
+
 func (c *Client) mapRoutine() {
 	defer close(c.mapDone)
-	bo := backoff.Backoff{Name: "mapRoutine"}
+	bo := backoff.NewBackoff("mapRoutine", c.logf)
 
 	for {
 		c.mu.Lock()
-		c.logf("mapRoutine: %s\n", c.state)
+		c.logf("mapRoutine: %s", c.state)
 		loggedIn := c.loggedIn
 		ctx := c.mapCtx
 		c.mu.Unlock()
 
 		select {
 		case <-c.quit:
-			c.logf("mapRoutine: quit\n")
+			c.logf("mapRoutine: quit")
 			return
 		default:
 		}
 
 		report := func(err error, msg string) {
-			c.logf("%s: %v\n", msg, err)
+			c.logf("%s: %v", msg, err)
 			err = fmt.Errorf("%s: %v", msg, err)
 			// don't send status updates for context errors,
 			// since context cancelation is always on purpose.
@@ -413,9 +436,9 @@ func (c *Client) mapRoutine() {
 
 			select {
 			case <-ctx.Done():
-				c.logf("mapRoutine: context done.\n")
+				c.logf("mapRoutine: context done.")
 			case <-c.newMapCh:
-				c.logf("mapRoutine: new map needed while idle.\n")
+				c.logf("mapRoutine: new map needed while idle.")
 			}
 		} else {
 			// Be sure this is false when we're not inside
@@ -430,7 +453,7 @@ func (c *Client) mapRoutine() {
 
 				select {
 				case <-c.newMapCh:
-					c.logf("mapRoutine: new map request during PollNetMap. canceling.\n")
+					c.logf("mapRoutine: new map request during PollNetMap. canceling.")
 					c.cancelMapLocked()
 
 					// Don't emit this netmap; we're
@@ -443,7 +466,7 @@ func (c *Client) mapRoutine() {
 				c.synced = true
 				c.inPollNetMap = true
 				if c.loggedIn {
-					c.state = stateSynchronized
+					c.state = StateSynchronized
 				}
 				exp := nm.Expiry
 				c.expiry = &exp
@@ -452,7 +475,7 @@ func (c *Client) mapRoutine() {
 
 				c.mu.Unlock()
 
-				c.logf("mapRoutine: netmap received: %s\n", state)
+				c.logf("mapRoutine: netmap received: %s", state)
 				if stillAuthed {
 					c.sendStatus("mapRoutine2", nil, "", nm)
 				}
@@ -461,8 +484,8 @@ func (c *Client) mapRoutine() {
 			c.mu.Lock()
 			c.synced = false
 			c.inPollNetMap = false
-			if c.state == stateSynchronized {
-				c.state = stateAuthenticated
+			if c.state == StateSynchronized {
+				c.state = StateAuthenticated
 			}
 			c.mu.Unlock()
 
@@ -489,12 +512,34 @@ func (c *Client) SetStatusFunc(fn func(Status)) {
 	c.mu.Unlock()
 }
 
-func (c *Client) SetHostinfo(hi tailcfg.Hostinfo) {
-	c.direct.SetHostinfo(hi)
+func (c *Client) SetHostinfo(hi *tailcfg.Hostinfo) {
+	if hi == nil {
+		panic("nil Hostinfo")
+	}
+	if !c.direct.SetHostinfo(hi) {
+		c.logf("[unexpected] duplicate Hostinfo: %v", hi)
+		return
+	}
+	c.logf("Hostinfo: %v", hi)
+
 	// Send new Hostinfo to server
 	c.cancelMapSafely()
 }
 
+func (c *Client) SetNetInfo(ni *tailcfg.NetInfo) {
+	if ni == nil {
+		panic("nil NetInfo")
+	}
+	if !c.direct.SetNetInfo(ni) {
+		c.logf("[unexpected] duplicate NetInfo: %v", ni)
+		return
+	}
+	c.logf("NetInfo: %v", ni)
+
+	// Send new Hostinfo (which includes NetInfo) to server
+	c.cancelMapSafely()
+}
+
 func (c *Client) sendStatus(who string, err error, url string, nm *NetworkMap) {
 	c.mu.Lock()
 	state := c.state
@@ -505,11 +550,11 @@ func (c *Client) sendStatus(who string, err error, url string, nm *NetworkMap) {
 	c.inSendStatus++
 	c.mu.Unlock()
 
-	c.logf("sendStatus: %s: %v\n", who, state)
+	c.logf("sendStatus: %s: %v", who, state)
 
 	var p *Persist
 	var fin *empty.Message
-	if state == stateAuthenticated {
+	if state == StateAuthenticated {
 		fin = new(empty.Message)
 	}
 	if nm != nil && loggedIn && synced {
@@ -526,7 +571,7 @@ func (c *Client) sendStatus(who string, err error, url string, nm *NetworkMap) {
 		Persist:       p,
 		NetMap:        nm,
 		Hostinfo:      hi,
-		state:         state,
+		State:         state,
 	}
 	if err != nil {
 		new.Err = err.Error()
@@ -541,7 +586,7 @@ func (c *Client) sendStatus(who string, err error, url string, nm *NetworkMap) {
 }
 
 func (c *Client) Login(t *oauth2.Token, flags LoginFlags) {
-	c.logf("client.Login(%v, %v)\n", t != nil, flags)
+	c.logf("client.Login(%v, %v)", t != nil, flags)
 
 	c.mu.Lock()
 	c.loginGoal = &LoginGoal{
@@ -555,7 +600,7 @@ func (c *Client) Login(t *oauth2.Token, flags LoginFlags) {
 }
 
 func (c *Client) Logout() {
-	c.logf("client.Logout()\n")
+	c.logf("client.Logout()")
 
 	c.mu.Lock()
 	c.loginGoal = &LoginGoal{
@@ -574,7 +619,7 @@ func (c *Client) UpdateEndpoints(localPort uint16, endpoints []string) {
 }
 
 func (c *Client) Shutdown() {
-	c.logf("client.Shutdown()\n")
+	c.logf("client.Shutdown()")
 
 	c.mu.Lock()
 	inSendStatus := c.inSendStatus
@@ -585,13 +630,30 @@ func (c *Client) Shutdown() {
 	}
 	c.mu.Unlock()
 
-	c.logf("client.Shutdown: inSendStatus=%v\n", inSendStatus)
+	c.logf("client.Shutdown: inSendStatus=%v", inSendStatus)
 	if !closed {
 		close(c.quit)
 		c.cancelAuth()
 		<-c.authDone
 		c.cancelMapUnsafely()
 		<-c.mapDone
-		c.logf("Client.Shutdown done.\n")
+		c.logf("Client.Shutdown done.")
 	}
 }
+
+// NodePublicKey returns the node public key currently in use. This is
+// used exclusively in tests.
+func (c *Client) TestOnlyNodePublicKey() wgcfg.Key {
+	priv := c.direct.GetPersist()
+	return priv.PrivateNodeKey.Public()
+}
+
+func (c *Client) TestOnlySetAuthKey(authkey string) {
+	c.direct.mu.Lock()
+	defer c.direct.mu.Unlock()
+	c.direct.authKey = authkey
+}
+
+func (c *Client) TestOnlyTimeNow() time.Time {
+	return c.timeNow()
+}

control/controlclient/controlclient_test.go

@@ -13,14 +13,16 @@ import (
 
 func fieldsOf(t reflect.Type) (fields []string) {
 	for i := 0; i < t.NumField(); i++ {
-		fields = append(fields, t.Field(i).Name)
+		if name := t.Field(i).Name; name != "_" {
+			fields = append(fields, name)
+		}
 	}
 	return
 }
 
 func TestStatusEqual(t *testing.T) {
 	// Verify that the Equal method stays in sync with reality
-	equalHandles := []string{"LoginFinished", "Err", "URL", "Persist", "NetMap", "Hostinfo", "state"}
+	equalHandles := []string{"LoginFinished", "Err", "URL", "Persist", "NetMap", "Hostinfo", "State"}
 	if have := fieldsOf(reflect.TypeOf(Status{})); !reflect.DeepEqual(have, equalHandles) {
 		t.Errorf("Status.Equal check might be out of sync\nfields: %q\nhandled: %q\n",
 			have, equalHandles)
@@ -46,13 +48,13 @@ func TestStatusEqual(t *testing.T) {
 			true,
 		},
 		{
-			&Status{state: stateNew},
-			&Status{state: stateNew},
+			&Status{State: StateNew},
+			&Status{State: StateNew},
 			true,
 		},
 		{
-			&Status{state: stateNew},
-			&Status{state: stateAuthenticated},
+			&Status{State: StateNew},
+			&Status{State: StateAuthenticated},
 			false,
 		},
 		{

control/controlclient/direct.go

@@ -16,8 +16,9 @@ import (
 	"io/ioutil"
 	"log"
 	"net/http"
+	"net/url"
 	"os"
-	"runtime"
+	"reflect"
 	"strconv"
 	"strings"
 	"sync"
@@ -26,13 +27,17 @@ import (
 	"github.com/tailscale/wireguard-go/wgcfg"
 	"golang.org/x/crypto/nacl/box"
 	"golang.org/x/oauth2"
+	"tailscale.com/log/logheap"
+	"tailscale.com/net/netns"
+	"tailscale.com/net/tlsdial"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/logger"
+	"tailscale.com/types/structs"
 	"tailscale.com/version"
-	"tailscale.com/wgengine/filter"
 )
 
 type Persist struct {
+	_                 structs.Incomparable
 	PrivateMachineKey wgcfg.PrivateKey
 	PrivateNodeKey    wgcfg.PrivateKey
 	OldPrivateNodeKey wgcfg.PrivateKey // needed to request key rotation
@@ -76,29 +81,35 @@ type Direct struct {
 	httpc           *http.Client // HTTP client used to talk to tailcontrol
 	serverURL       string       // URL of the tailcontrol server
 	timeNow         func() time.Time
+	lastPrintMap    time.Time
 	newDecompressor func() (Decompressor, error)
 	keepAlive       bool
 	logf            logger.Logf
+	discoPubKey     tailcfg.DiscoKey
 
 	mu           sync.Mutex // mutex guards the following fields
 	serverKey    wgcfg.Key
 	persist      Persist
+	authKey      string
 	tryingNewKey wgcfg.PrivateKey
 	expiry       *time.Time
-	hostinfo     tailcfg.Hostinfo
-	endpoints    []string
-	localPort    uint16 // or zero to mean auto
+	// hostinfo is mutated in-place while mu is held.
+	hostinfo  *tailcfg.Hostinfo // always non-nil
+	endpoints []string
+	localPort uint16 // or zero to mean auto
 }
 
 type Options struct {
-	Persist         Persist          // initial persistent data
-	HTTPC           *http.Client     // HTTP client used to talk to tailcontrol
-	ServerURL       string           // URL of the tailcontrol server
-	TimeNow         func() time.Time // time.Now implementation used by Client
-	Hostinfo        *tailcfg.Hostinfo
+	Persist         Persist           // initial persistent data
+	ServerURL       string            // URL of the tailcontrol server
+	AuthKey         string            // optional node auth key for auto registration
+	TimeNow         func() time.Time  // time.Now implementation used by Client
+	Hostinfo        *tailcfg.Hostinfo // non-nil passes ownership, nil means to use default using os.Hostname, etc
+	DiscoPublicKey  tailcfg.DiscoKey
 	NewDecompressor func() (Decompressor, error)
 	KeepAlive       bool
 	Logf            logger.Logf
+	HTTPTestClient  *http.Client // optional HTTP client to use (for tests only)
 }
 
 type Decompressor interface {
@@ -112,61 +123,91 @@ func NewDirect(opts Options) (*Direct, error) {
 		return nil, errors.New("controlclient.New: no server URL specified")
 	}
 	opts.ServerURL = strings.TrimRight(opts.ServerURL, "/")
-	if opts.HTTPC == nil {
-		opts.HTTPC = http.DefaultClient
+	serverURL, err := url.Parse(opts.ServerURL)
+	if err != nil {
+		return nil, err
 	}
 	if opts.TimeNow == nil {
 		opts.TimeNow = time.Now
 	}
 	if opts.Logf == nil {
 		// TODO(apenwarr): remove this default and fail instead.
+		// TODO(bradfitz): ... but then it shouldn't be in Options.
 		opts.Logf = log.Printf
 	}
 
+	httpc := opts.HTTPTestClient
+	if httpc == nil {
+		dialer := netns.NewDialer()
+		tr := http.DefaultTransport.(*http.Transport).Clone()
+		tr.DialContext = dialer.DialContext
+		tr.ForceAttemptHTTP2 = true
+		tr.TLSClientConfig = tlsdial.Config(serverURL.Host, tr.TLSClientConfig)
+		httpc = &http.Client{Transport: tr}
+	}
+
 	c := &Direct{
-		httpc:           opts.HTTPC,
+		httpc:           httpc,
 		serverURL:       opts.ServerURL,
 		timeNow:         opts.TimeNow,
 		logf:            opts.Logf,
 		newDecompressor: opts.NewDecompressor,
 		keepAlive:       opts.KeepAlive,
 		persist:         opts.Persist,
+		authKey:         opts.AuthKey,
+		discoPubKey:     opts.DiscoPublicKey,
 	}
 	if opts.Hostinfo == nil {
 		c.SetHostinfo(NewHostinfo())
 	} else {
-		c.SetHostinfo(*opts.Hostinfo)
+		c.SetHostinfo(opts.Hostinfo)
 	}
-
 	return c, nil
 }
 
-func NewHostinfo() tailcfg.Hostinfo {
+func NewHostinfo() *tailcfg.Hostinfo {
 	hostname, _ := os.Hostname()
-	os := runtime.GOOS
-	switch os {
-	case "darwin":
-		switch runtime.GOARCH {
-		case "arm", "arm64":
-			os = "iOS"
-		default:
-			os = "macOS"
-		}
-	}
-
-	return tailcfg.Hostinfo{
+	return &tailcfg.Hostinfo{
 		IPNVersion: version.LONG,
 		Hostname:   hostname,
-		OS:         os,
+		OS:         version.OS(),
 	}
 }
 
-func (c *Direct) SetHostinfo(hi tailcfg.Hostinfo) {
+// SetHostinfo clones the provided Hostinfo and remembers it for the
+// next update. It reports whether the Hostinfo has changed.
+func (c *Direct) SetHostinfo(hi *tailcfg.Hostinfo) bool {
+	if hi == nil {
+		panic("nil Hostinfo")
+	}
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
-	c.logf("Hostinfo: %v\n", hi)
-	c.hostinfo = hi
+	if hi.Equal(c.hostinfo) {
+		return false
+	}
+	c.hostinfo = hi.Clone()
+	return true
+}
+
+// SetNetInfo clones the provided NetInfo and remembers it for the
+// next update. It reports whether the NetInfo has changed.
+func (c *Direct) SetNetInfo(ni *tailcfg.NetInfo) bool {
+	if ni == nil {
+		panic("nil NetInfo")
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	if c.hostinfo == nil {
+		c.logf("[unexpected] SetNetInfo called with no HostInfo; ignoring NetInfo update: %+v", ni)
+		return false
+	}
+	if reflect.DeepEqual(ni, c.hostinfo.NetInfo) {
+		return false
+	}
+	c.hostinfo.NetInfo = ni.Clone()
+	return true
 }
 
 func (c *Direct) GetPersist() Persist {
@@ -183,7 +224,7 @@ const (
 )
 
 func (c *Direct) TryLogout(ctx context.Context) error {
-	c.logf("direct.TryLogout()\n")
+	c.logf("direct.TryLogout()")
 
 	c.mu.Lock()
 	defer c.mu.Unlock()
@@ -199,12 +240,12 @@ func (c *Direct) TryLogout(ctx context.Context) error {
 }
 
 func (c *Direct) TryLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags) (url string, err error) {
-	c.logf("direct.TryLogin(%v, %v)\n", t != nil, flags)
+	c.logf("direct.TryLogin(%v, %v)", t != nil, flags)
 	return c.doLoginOrRegen(ctx, t, flags, false, "")
 }
 
 func (c *Direct) WaitLoginURL(ctx context.Context, url string) (newUrl string, err error) {
-	c.logf("direct.WaitLoginURL\n")
+	c.logf("direct.WaitLoginURL")
 	return c.doLoginOrRegen(ctx, nil, LoginDefault, false, url)
 }
 
@@ -224,11 +265,14 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	persist := c.persist
 	tryingNewKey := c.tryingNewKey
 	serverKey := c.serverKey
+	authKey := c.authKey
+	hostinfo := c.hostinfo.Clone()
+	backendLogID := hostinfo.BackendLogID
 	expired := c.expiry != nil && !c.expiry.IsZero() && c.expiry.Before(c.timeNow())
 	c.mu.Unlock()
 
 	if persist.PrivateMachineKey == (wgcfg.PrivateKey{}) {
-		c.logf("Generating a new machinekey.\n")
+		c.logf("Generating a new machinekey.")
 		mkey, err := wgcfg.NewPrivateKey()
 		if err != nil {
 			log.Fatal(err)
@@ -237,15 +281,15 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	}
 
 	if expired {
-		c.logf("Old key expired -> regen=true\n")
+		c.logf("Old key expired -> regen=true")
 		regen = true
 	}
 	if (flags & LoginInteractive) != 0 {
-		c.logf("LoginInteractive -> regen=true\n")
+		c.logf("LoginInteractive -> regen=true")
 		regen = true
 	}
 
-	c.logf("doLogin(regen=%v, hasUrl=%v)\n", regen, url != "")
+	c.logf("doLogin(regen=%v, hasUrl=%v)", regen, url != "")
 	if serverKey == (wgcfg.Key{}) {
 		var err error
 		serverKey, err = loadServerKey(ctx, c.httpc, c.serverURL)
@@ -261,7 +305,7 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	var oldNodeKey wgcfg.Key
 	if url != "" {
 	} else if regen || persist.PrivateNodeKey == (wgcfg.PrivateKey{}) {
-		c.logf("Generating a new nodekey.\n")
+		c.logf("Generating a new nodekey.")
 		persist.OldPrivateNodeKey = persist.PrivateNodeKey
 		key, err := wgcfg.NewPrivateKey()
 		if err != nil {
@@ -278,9 +322,9 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	}
 
 	if tryingNewKey == (wgcfg.PrivateKey{}) {
-		log.Fatalf("tryingNewKey is empty, give up\n")
+		log.Fatalf("tryingNewKey is empty, give up")
 	}
-	if c.hostinfo.BackendLogID == "" {
+	if backendLogID == "" {
 		err = errors.New("hostinfo: BackendLogID missing")
 		return regen, url, err
 	}
@@ -288,15 +332,16 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 		Version:    1,
 		OldNodeKey: tailcfg.NodeKey(oldNodeKey),
 		NodeKey:    tailcfg.NodeKey(tryingNewKey.Public()),
-		Hostinfo:   c.hostinfo,
+		Hostinfo:   hostinfo,
 		Followup:   url,
 	}
-	c.logf("RegisterReq: onode=%v node=%v fup=%v\n",
-		request.OldNodeKey.AbbrevString(),
-		request.NodeKey.AbbrevString(), url != "")
+	c.logf("RegisterReq: onode=%v node=%v fup=%v",
+		request.OldNodeKey.ShortString(),
+		request.NodeKey.ShortString(), url != "")
 	request.Auth.Oauth2Token = t
 	request.Auth.Provider = persist.Provider
 	request.Auth.LoginName = persist.LoginName
+	request.Auth.AuthKey = authKey
 	bodyData, err := encode(request, &serverKey, &persist.PrivateMachineKey)
 	if err != nil {
 		return regen, url, err
@@ -314,7 +359,7 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	if err != nil {
 		return regen, url, fmt.Errorf("register request: %v", err)
 	}
-	c.logf("RegisterReq: returned.\n")
+	c.logf("RegisterReq: returned.")
 	resp := tailcfg.RegisterResponse{}
 	if err := decode(res, &resp, &serverKey, &persist.PrivateMachineKey); err != nil {
 		return regen, url, fmt.Errorf("register request: %v", err)
@@ -325,7 +370,7 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 			return true, "", fmt.Errorf("weird: regen=true but server says NodeKeyExpired: %v", request.NodeKey)
 		}
 		c.logf("server reports new node key %v has expired",
-			request.NodeKey.AbbrevString())
+			request.NodeKey.ShortString())
 		return true, "", nil
 	}
 	if persist.Provider == "" {
@@ -342,9 +387,9 @@ func (c *Direct) doLogin(ctx context.Context, t *oauth2.Token, flags LoginFlags,
 	//	- user is disabled
 
 	if resp.AuthURL != "" {
-		c.logf("AuthURL is %.20v...\n", resp.AuthURL)
+		c.logf("AuthURL is %v", resp.AuthURL)
 	} else {
-		c.logf("No AuthURL\n")
+		c.logf("No AuthURL")
 	}
 
 	c.mu.Lock()
@@ -391,7 +436,7 @@ func (c *Direct) newEndpoints(localPort uint16, endpoints []string) (changed boo
 	if c.localPort == localPort && sameStrings(c.endpoints, endpoints) {
 		return false // unchanged
 	}
-	c.logf("client.newEndpoints(%v, %v)\n", localPort, endpoints)
+	c.logf("client.newEndpoints(%v, %v)", localPort, endpoints)
 	c.localPort = localPort
 	c.endpoints = append(c.endpoints[:0], endpoints...)
 	return true // changed
@@ -411,25 +456,34 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 	persist := c.persist
 	serverURL := c.serverURL
 	serverKey := c.serverKey
-	hostinfo := c.hostinfo
+	hostinfo := c.hostinfo.Clone()
+	backendLogID := hostinfo.BackendLogID
 	localPort := c.localPort
 	ep := append([]string(nil), c.endpoints...)
 	c.mu.Unlock()
 
-	if hostinfo.BackendLogID == "" {
+	if backendLogID == "" {
 		return errors.New("hostinfo: BackendLogID missing")
 	}
 
 	allowStream := maxPolls != 1
-	c.logf("PollNetMap: stream=%v :%v %v\n", maxPolls, localPort, ep)
+	c.logf("PollNetMap: stream=%v :%v %v", maxPolls, localPort, ep)
+
+	vlogf := logger.Discard
+	if Debug.NetMap {
+		vlogf = c.logf
+	}
 
 	request := tailcfg.MapRequest{
-		Version:   4,
-		KeepAlive: c.keepAlive,
-		NodeKey:   tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
-		Endpoints: ep,
-		Stream:    allowStream,
-		Hostinfo:  hostinfo,
+		Version:         4,
+		IncludeIPv6:     true,
+		KeepAlive:       c.keepAlive,
+		NodeKey:         tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
+		DiscoKey:        c.discoPubKey,
+		Endpoints:       ep,
+		Stream:          allowStream,
+		Hostinfo:        hostinfo,
+		DebugForceDisco: Debug.ForceDisco,
 	}
 	if c.newDecompressor != nil {
 		request.Compress = "zstd"
@@ -437,9 +491,11 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 
 	bodyData, err := encode(request, &serverKey, &persist.PrivateMachineKey)
 	if err != nil {
+		vlogf("netmap: encode: %v", err)
 		return err
 	}
 
+	t0 := time.Now()
 	u := fmt.Sprintf("%s/machine/%s/map", serverURL, persist.PrivateMachineKey.Public().HexString())
 	req, err := http.NewRequest("POST", u, bytes.NewReader(bodyData))
 	if err != nil {
@@ -451,8 +507,10 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 
 	res, err := c.httpc.Do(req)
 	if err != nil {
+		vlogf("netmap: Do: %v", err)
 		return err
 	}
+	vlogf("netmap: Do = %v after %v", res.StatusCode, time.Since(t0).Round(time.Millisecond))
 	if res.StatusCode != 200 {
 		msg, _ := ioutil.ReadAll(res.Body)
 		res.Body.Close()
@@ -467,26 +525,35 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 	const pollTimeout = 120 * time.Second
 	timeout := time.NewTimer(pollTimeout)
 	timeoutReset := make(chan struct{})
-	defer close(timeoutReset)
+	pollDone := make(chan struct{})
+	defer close(pollDone)
 	go func() {
 		for {
 			select {
+			case <-pollDone:
+				vlogf("netmap: ending timeout goroutine")
+				return
 			case <-timeout.C:
 				c.logf("map response long-poll timed out!")
 				cancel()
 				return
-			case _, ok := <-timeoutReset:
-				if !ok {
-					return // channel closed, shut down goroutine
-				}
+			case <-timeoutReset:
 				if !timeout.Stop() {
-					<-timeout.C
+					select {
+					case <-timeout.C:
+					case <-pollDone:
+						vlogf("netmap: ending timeout goroutine")
+						return
+					}
 				}
+				vlogf("netmap: reset timeout timer")
 				timeout.Reset(pollTimeout)
 			}
 		}
 	}()
 
+	var lastDERPMap *tailcfg.DERPMap
+
 	// If allowStream, then the server will use an HTTP long poll to
 	// return incremental results. There is always one response right
 	// away, followed by a delay, and eventually others.
@@ -495,33 +562,58 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 	// We can use this same read loop either way.
 	var msg []byte
 	for i := 0; i < maxPolls || maxPolls < 0; i++ {
+		vlogf("netmap: starting size read after %v (poll %v)", time.Since(t0).Round(time.Millisecond), i)
 		var siz [4]byte
 		if _, err := io.ReadFull(res.Body, siz[:]); err != nil {
+			vlogf("netmap: size read error after %v: %v", time.Since(t0).Round(time.Millisecond), err)
 			return err
 		}
 		size := binary.LittleEndian.Uint32(siz[:])
+		vlogf("netmap: read size %v after %v", size, time.Since(t0).Round(time.Millisecond))
 		msg = append(msg[:0], make([]byte, size)...)
 		if _, err := io.ReadFull(res.Body, msg); err != nil {
+			vlogf("netmap: body read error: %v", err)
 			return err
 		}
+		vlogf("netmap: read body after %v", time.Since(t0).Round(time.Millisecond))
 
 		var resp tailcfg.MapResponse
-
-		// Default filter if the key is missing from the incoming
-		// json (ie. old tailcontrol server without PacketFilter
-		// support). If even an empty PacketFilter is provided, this
-		// will be overwritten.
-		// TODO(apenwarr 2020-02-01): remove after tailcontrol is fully deployed.
-		resp.PacketFilter = filter.MatchAllowAll
-
 		if err := c.decodeMsg(msg, &resp); err != nil {
+			vlogf("netmap: decode error: %v")
 			return err
 		}
 		if resp.KeepAlive {
-			c.logf("map response keep alive received")
-			timeoutReset <- struct{}{}
+			vlogf("netmap: got keep-alive")
+			select {
+			case timeoutReset <- struct{}{}:
+				vlogf("netmap: sent keep-alive timer reset")
+			case <-ctx.Done():
+				c.logf("netmap: not resetting timer for keep-alive due to: %v", ctx.Err())
+				return ctx.Err()
+			}
 			continue
 		}
+		vlogf("netmap: got new map")
+
+		if resp.DERPMap != nil {
+			vlogf("netmap: new map contains DERP map")
+			lastDERPMap = resp.DERPMap
+		}
+		if resp.Debug != nil && resp.Debug.LogHeapPprof {
+			go logheap.LogHeap(resp.Debug.LogHeapURL)
+		}
+		// Temporarily (2020-06-29) support removing all but
+		// discovery-supporting nodes during development, for
+		// less noise.
+		if Debug.OnlyDisco {
+			filtered := resp.Peers[:0]
+			for _, p := range resp.Peers {
+				if !p.DiscoKey.IsZero() {
+					filtered = append(filtered, p)
+				}
+			}
+			resp.Peers = filtered
+		}
 
 		nm := &NetworkMap{
 			NodeKey:      tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
@@ -537,15 +629,9 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 			DNS:          resp.DNS,
 			DNSDomains:   resp.SearchPaths,
 			Hostinfo:     resp.Node.Hostinfo,
-			PacketFilter: resp.PacketFilter,
-		}
-		// Temporary (2020-02-21) knob to force debug, during DERP testing:
-		if ok, _ := strconv.ParseBool(os.Getenv("DEBUG_FORCE_DERP")); ok {
-			c.logf("debug: adding DERP endpoints to all peers")
-			for i := range nm.Peers {
-				peer := &nm.Peers[i]
-				peer.Endpoints = append([]string{"127.3.3.40:1"}, peer.Endpoints...)
-			}
+			PacketFilter: c.parsePacketFilter(resp.PacketFilter),
+			DERPMap:      lastDERPMap,
+			Debug:        resp.Debug,
 		}
 		for _, profile := range resp.UserProfiles {
 			nm.UserProfiles[profile.ID] = profile
@@ -555,7 +641,17 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 		} else {
 			nm.MachineStatus = tailcfg.MachineUnauthorized
 		}
-		//c.logf("new network map[%d]:\n%s", i, nm.Concise())
+
+		// Printing the netmap can be extremely verbose, but is very
+		// handy for debugging. Let's limit how often we do it.
+		// Code elsewhere prints netmap diffs every time, so this
+		// occasional full dump, plus incremental diffs, should do
+		// the job.
+		now := c.timeNow()
+		if now.Sub(c.lastPrintMap) >= 5*time.Minute {
+			c.lastPrintMap = now
+			c.logf("new network map[%d]:\n%s", i, nm.Concise())
+		}
 
 		c.mu.Lock()
 		c.expiry = &nm.Expiry
@@ -582,8 +678,10 @@ func decode(res *http.Response, v interface{}, serverKey *wgcfg.Key, mkey *wgcfg
 }
 
 func (c *Direct) decodeMsg(msg []byte, v interface{}) error {
+	c.mu.Lock()
 	mkey := c.persist.PrivateMachineKey
 	serverKey := c.serverKey
+	c.mu.Unlock()
 
 	decrypted, err := decryptMsg(msg, &serverKey, &mkey)
 	if err != nil {
@@ -593,7 +691,6 @@ func (c *Direct) decodeMsg(msg []byte, v interface{}) error {
 	if c.newDecompressor == nil {
 		b = decrypted
 	} else {
-		//decoder, err := zstd.NewReader(nil)
 		decoder, err := c.newDecompressor()
 		if err != nil {
 			return err
@@ -643,6 +740,12 @@ func encode(v interface{}, serverKey *wgcfg.Key, mkey *wgcfg.PrivateKey) ([]byte
 	if err != nil {
 		return nil, err
 	}
+	const debugMapRequests = false
+	if debugMapRequests {
+		if _, ok := v.(tailcfg.MapRequest); ok {
+			log.Printf("MapRequest: %s", b)
+		}
+	}
 	var nonce [24]byte
 	if _, err := io.ReadFull(rand.Reader, nonce[:]); err != nil {
 		panic(err)
@@ -676,3 +779,39 @@ func loadServerKey(ctx context.Context, httpc *http.Client, serverURL string) (w
 	}
 	return key, nil
 }
+
+// Debug contains temporary internal-only debug knobs.
+// They're unexported to not draw attention to them.
+var Debug = initDebug()
+
+type debug struct {
+	NetMap     bool
+	OnlyDisco  bool
+	Disco      bool
+	ForceDisco bool // ask control server to not filter out our disco key
+}
+
+func initDebug() debug {
+	d := debug{
+		NetMap:     envBool("TS_DEBUG_NETMAP"),
+		OnlyDisco:  os.Getenv("TS_DEBUG_USE_DISCO") == "only",
+		ForceDisco: os.Getenv("TS_DEBUG_USE_DISCO") == "only" || envBool("TS_DEBUG_USE_DISCO"),
+	}
+	if d.ForceDisco || os.Getenv("TS_DEBUG_USE_DISCO") == "" {
+		// This is now defaults to on.
+		d.Disco = true
+	}
+	return d
+}
+
+func envBool(k string) bool {
+	e := os.Getenv(k)
+	if e == "" {
+		return false
+	}
+	v, err := strconv.ParseBool(e)
+	if err != nil {
+		panic(fmt.Sprintf("invalid non-bool %q for env var %q", e, k))
+	}
+	return v
+}

control/controlclient/direct_test.go

@@ -1,305 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build depends_on_currently_unreleased
-
-package controlclient
-
-import (
-	"context"
-	"io/ioutil"
-	"net/http"
-	"net/http/httptest"
-	"os"
-	"testing"
-	"time"
-
-	"github.com/klauspost/compress/zstd"
-	"github.com/tailscale/wireguard-go/wgcfg"
-	"tailscale.com/tailcfg"
-	"tailscale.io/control" // not yet released
-)
-
-func TestClientsReusingKeys(t *testing.T) {
-	tmpdir, err := ioutil.TempDir("", "control-test-")
-	if err != nil {
-		t.Fatal(err)
-	}
-	var server *control.Server
-	httpsrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		server.ServeHTTP(w, r)
-	}))
-	server, err = control.New(tmpdir, httpsrv.URL, true)
-	if err != nil {
-		t.Fatal(err)
-	}
-	server.QuietLogging = true
-	defer func() {
-		httpsrv.CloseClientConnections()
-		httpsrv.Close()
-		os.RemoveAll(tmpdir)
-	}()
-
-	hi := NewHostinfo()
-	hi.FrontendLogID = "go-test-only"
-	hi.BackendLogID = "go-test-only"
-	c1, err := NewDirect(Options{
-		ServerURL: httpsrv.URL,
-		HTTPC:     httpsrv.Client(),
-		//TimeNow:   s.control.TimeNow,
-		Logf: func(fmt string, args ...interface{}) {
-			t.Helper()
-			t.Logf("c1: "+fmt, args...)
-		},
-		Hostinfo: &hi,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-	authURL, err := c1.TryLogin(ctx, nil, 0)
-	if err != nil {
-		t.Fatal(err)
-	}
-	const user = "testuser1@tailscale.onmicrosoft.com"
-	postAuthURL(t, ctx, httpsrv.Client(), user, authURL)
-	newURL, err := c1.WaitLoginURL(ctx, authURL)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if newURL != "" {
-		t.Fatalf("unexpected newURL: %s", newURL)
-	}
-
-	pollErrCh := make(chan error)
-	go func() {
-		err := c1.PollNetMap(ctx, -1, func(netMap *NetworkMap) {})
-		pollErrCh <- err
-	}()
-
-	select {
-	case err := <-pollErrCh:
-		t.Fatal(err)
-	default:
-	}
-
-	c2, err := NewDirect(Options{
-		ServerURL: httpsrv.URL,
-		HTTPC:     httpsrv.Client(),
-		Logf: func(fmt string, args ...interface{}) {
-			t.Helper()
-			t.Logf("c2: "+fmt, args...)
-		},
-		Persist:  c1.GetPersist(),
-		Hostinfo: &hi,
-		NewDecompressor: func() (Decompressor, error) {
-			return zstd.NewReader(nil)
-		},
-		KeepAlive: true,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	authURL, err = c2.TryLogin(ctx, nil, 0)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if authURL != "" {
-		t.Errorf("unexpected authURL %s", authURL)
-	}
-
-	err = c2.PollNetMap(ctx, 1, func(netMap *NetworkMap) {})
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	select {
-	case err := <-pollErrCh:
-		t.Logf("expected poll error: %v", err)
-	case <-time.After(5 * time.Second):
-		t.Fatal("first client poll failed to close")
-	}
-}
-
-func TestClientsReusingOldKey(t *testing.T) {
-	tmpdir, err := ioutil.TempDir("", "control-test-")
-	if err != nil {
-		t.Fatal(err)
-	}
-	var server *control.Server
-	httpsrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		server.ServeHTTP(w, r)
-	}))
-	server, err = control.New(tmpdir, httpsrv.URL, true)
-	if err != nil {
-		t.Fatal(err)
-	}
-	server.QuietLogging = true
-	defer func() {
-		httpsrv.CloseClientConnections()
-		httpsrv.Close()
-		os.RemoveAll(tmpdir)
-	}()
-
-	hi := NewHostinfo()
-	hi.FrontendLogID = "go-test-only"
-	hi.BackendLogID = "go-test-only"
-	genOpts := func() Options {
-		return Options{
-			ServerURL: httpsrv.URL,
-			HTTPC:     httpsrv.Client(),
-			//TimeNow:   s.control.TimeNow,
-			Logf: func(fmt string, args ...interface{}) {
-				t.Helper()
-				t.Logf("c1: "+fmt, args...)
-			},
-			Hostinfo: &hi,
-		}
-	}
-
-	// Login with a new node key. This requires authorization.
-	c1, err := NewDirect(genOpts())
-	if err != nil {
-		t.Fatal(err)
-	}
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
-	authURL, err := c1.TryLogin(ctx, nil, 0)
-	if err != nil {
-		t.Fatal(err)
-	}
-	const user = "testuser1@tailscale.onmicrosoft.com"
-	postAuthURL(t, ctx, httpsrv.Client(), user, authURL)
-	newURL, err := c1.WaitLoginURL(ctx, authURL)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if newURL != "" {
-		t.Fatalf("unexpected newURL: %s", newURL)
-	}
-
-	if err := c1.PollNetMap(ctx, 1, func(netMap *NetworkMap) {}); err != nil {
-		t.Fatal(err)
-	}
-
-	newPrivKey := func(t *testing.T) wgcfg.PrivateKey {
-		t.Helper()
-		k, err := wgcfg.NewPrivateKey()
-		if err != nil {
-			t.Fatal(err)
-		}
-		return k
-	}
-
-	// Replace the previous key with a new key.
-	persist1 := c1.GetPersist()
-	persist2 := Persist{
-		PrivateMachineKey: persist1.PrivateMachineKey,
-		OldPrivateNodeKey: persist1.PrivateNodeKey,
-		PrivateNodeKey:    newPrivKey(t),
-	}
-	opts := genOpts()
-	opts.Persist = persist2
-
-	c1, err = NewDirect(opts)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if authURL, err := c1.TryLogin(ctx, nil, 0); err != nil {
-		t.Fatal(err)
-	} else if authURL == "" {
-		t.Fatal("expected authURL for reused oldNodeKey, got none")
-	} else {
-		postAuthURL(t, ctx, httpsrv.Client(), user, authURL)
-		if newURL, err := c1.WaitLoginURL(ctx, authURL); err != nil {
-			t.Fatal(err)
-		} else if newURL != "" {
-			t.Fatalf("unexpected newURL: %s", newURL)
-		}
-	}
-	if p := c1.GetPersist(); p.PrivateNodeKey != opts.Persist.PrivateNodeKey {
-		t.Error("unexpected node key change")
-	} else {
-		persist2 = p
-	}
-
-	// Here we simulate a client using using old persistent data.
-	// We use the key we have already replaced as the old node key.
-	// This requires the user to authenticate.
-	persist3 := Persist{
-		PrivateMachineKey: persist1.PrivateMachineKey,
-		OldPrivateNodeKey: persist1.PrivateNodeKey,
-		PrivateNodeKey:    newPrivKey(t),
-	}
-	opts = genOpts()
-	opts.Persist = persist3
-
-	c1, err = NewDirect(opts)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if authURL, err := c1.TryLogin(ctx, nil, 0); err != nil {
-		t.Fatal(err)
-	} else if authURL == "" {
-		t.Fatal("expected authURL for reused oldNodeKey, got none")
-	} else {
-		postAuthURL(t, ctx, httpsrv.Client(), user, authURL)
-		if newURL, err := c1.WaitLoginURL(ctx, authURL); err != nil {
-			t.Fatal(err)
-		} else if newURL != "" {
-			t.Fatalf("unexpected newURL: %s", newURL)
-		}
-	}
-	if err := c1.PollNetMap(ctx, 1, func(netMap *NetworkMap) {}); err != nil {
-		t.Fatal(err)
-	}
-
-	// At this point, there should only be one node for the machine key
-	// registered as active in the server.
-	mkey := tailcfg.MachineKey(persist1.PrivateMachineKey.Public())
-	nodeIDs, err := server.DB().MachineNodes(mkey)
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(nodeIDs) != 1 {
-		t.Logf("active nodes for machine key %v:", mkey)
-		for i, nodeID := range nodeIDs {
-			nodeKey := server.DB().NodeKey(nodeID)
-			t.Logf("\tnode %d: id=%v, key=%v", i, nodeID, nodeKey)
-		}
-		t.Fatalf("want 1 active node for the client machine, got %d", len(nodeIDs))
-	}
-
-	// Now try the previous node key. It should fail.
-	opts = genOpts()
-	opts.Persist = persist2
-	c1, err = NewDirect(opts)
-	if err != nil {
-		t.Fatal(err)
-	}
-	// TODO(crawshaw): make this return an actual error.
-	// Have cfgdb track expired keys, and when an expired key is reused
-	// produce an error.
-	if authURL, err := c1.TryLogin(ctx, nil, 0); err != nil {
-		t.Fatal(err)
-	} else if authURL == "" {
-		t.Fatal("expected authURL for reused nodeKey, got none")
-	} else {
-		postAuthURL(t, ctx, httpsrv.Client(), user, authURL)
-		if newURL, err := c1.WaitLoginURL(ctx, authURL); err != nil {
-			t.Fatal(err)
-		} else if newURL != "" {
-			t.Fatalf("unexpected newURL: %s", newURL)
-		}
-	}
-	if err := c1.PollNetMap(ctx, 1, func(netMap *NetworkMap) {}); err != nil {
-		t.Fatal(err)
-	}
-	if nodeIDs, err := server.DB().MachineNodes(mkey); err != nil {
-		t.Fatal(err)
-	} else if len(nodeIDs) != 1 {
-		t.Fatalf("want 1 active node for the client machine, got %d", len(nodeIDs))
-	}
-}

control/controlclient/filter.go

@@ -0,0 +1,84 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"fmt"
+	"net"
+	"tailscale.com/tailcfg"
+	"tailscale.com/wgengine/filter"
+)
+
+func parseIP(host string, defaultBits int) (filter.Net, error) {
+	ip := net.ParseIP(host)
+	if ip != nil && ip.IsUnspecified() {
+		// For clarity, reject 0.0.0.0 as an input
+		return filter.NetNone, fmt.Errorf("ports=%#v: to allow all IP addresses, use *:port, not 0.0.0.0:port", host)
+	} else if ip == nil && host == "*" {
+		// User explicitly requested wildcard dst ip
+		return filter.NetAny, nil
+	} else {
+		if ip != nil {
+			ip = ip.To4()
+		}
+		if ip == nil || len(ip) != 4 {
+			return filter.NetNone, fmt.Errorf("ports=%#v: invalid IPv4 address", host)
+		}
+		return filter.Net{
+			IP:   filter.NewIP(ip),
+			Mask: filter.Netmask(defaultBits),
+		}, nil
+	}
+}
+
+// Parse a backward-compatible FilterRule used by control's wire format,
+// producing the most current filter.Matches format.
+func (c *Direct) parsePacketFilter(pf []tailcfg.FilterRule) filter.Matches {
+	mm := make([]filter.Match, 0, len(pf))
+	var erracc error
+
+	for _, r := range pf {
+		m := filter.Match{}
+
+		for i, s := range r.SrcIPs {
+			bits := 32
+			if len(r.SrcBits) > i {
+				bits = r.SrcBits[i]
+			}
+			net, err := parseIP(s, bits)
+			if err != nil && erracc == nil {
+				erracc = err
+				continue
+			}
+			m.Srcs = append(m.Srcs, net)
+		}
+
+		for _, d := range r.DstPorts {
+			bits := 32
+			if d.Bits != nil {
+				bits = *d.Bits
+			}
+			net, err := parseIP(d.IP, bits)
+			if err != nil && erracc == nil {
+				erracc = err
+				continue
+			}
+			m.Dsts = append(m.Dsts, filter.NetPortRange{
+				Net: net,
+				Ports: filter.PortRange{
+					First: d.Ports.First,
+					Last:  d.Ports.Last,
+				},
+			})
+		}
+
+		mm = append(mm, m)
+	}
+
+	if erracc != nil {
+		c.logf("parsePacketFilter: %s\n", erracc)
+	}
+	return mm
+}

control/controlclient/netmap.go

@@ -5,18 +5,18 @@
 package controlclient
 
 import (
-	"bytes"
-	"encoding/base64"
 	"encoding/json"
 	"fmt"
 	"log"
 	"net"
-	"runtime"
+	"reflect"
+	"strconv"
 	"strings"
 	"time"
 
 	"github.com/tailscale/wireguard-go/wgcfg"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
 	"tailscale.com/wgengine/filter"
 )
 
@@ -29,12 +29,19 @@ type NetworkMap struct {
 	Addresses     []wgcfg.CIDR
 	LocalPort     uint16 // used for debugging
 	MachineStatus tailcfg.MachineStatus
-	Peers         []tailcfg.Node
+	Peers         []*tailcfg.Node // sorted by Node.ID
 	DNS           []wgcfg.IP
 	DNSDomains    []string
 	Hostinfo      tailcfg.Hostinfo
 	PacketFilter  filter.Matches
 
+	// DERPMap is the last DERP server map received. It's reused
+	// between updates and should not be modified.
+	DERPMap *tailcfg.DERPMap
+
+	// Debug knobs from control server for debug or feature gating.
+	Debug *tailcfg.Debug
+
 	// ACLs
 
 	User   tailcfg.UserID
@@ -47,55 +54,148 @@ type NetworkMap struct {
 	// TODO(crawshaw): Capabilities []tailcfg.Capability
 }
 
-func (n *NetworkMap) Equal(n2 *NetworkMap) bool {
-	// TODO(crawshaw): this is crude, but is an easy way to avoid bugs.
-	b, err := json.Marshal(n)
-	if err != nil {
-		panic(err)
-	}
-	b2, err := json.Marshal(n2)
-	if err != nil {
-		panic(err)
-	}
-	return bytes.Equal(b, b2)
-}
-
 func (nm NetworkMap) String() string {
 	return nm.Concise()
 }
 
-func keyString(key [32]byte) string {
-	b64 := base64.StdEncoding.EncodeToString(key[:])
-	abbrev := "invalid"
-	if len(b64) == 44 {
-		abbrev = b64[0:4] + "…" + b64[39:43]
-	}
-	return fmt.Sprintf("[%s]", abbrev)
-}
-
 func (nm *NetworkMap) Concise() string {
 	buf := new(strings.Builder)
-	fmt.Fprintf(buf, "NetworkMap: self: %v auth=%v :%v %v\n",
-		keyString(nm.NodeKey), nm.MachineStatus,
-		nm.LocalPort, nm.Addresses)
+
+	nm.printConciseHeader(buf)
 	for _, p := range nm.Peers {
-		aip := make([]string, len(p.AllowedIPs))
-		for i, a := range p.AllowedIPs {
-			aip[i] = fmt.Sprint(a)
-		}
-		u := fmt.Sprint(p.User)
-		if strings.HasPrefix(u, "userid:") {
-			u = "u:" + u[7:]
-		}
-		f1 := fmt.Sprintf(" %v %-6v %v",
-			keyString(p.Key), u, p.Endpoints)
-		f2 := fmt.Sprintf(" %*v\n", 70-len(f1),
-			strings.Join(aip, " "))
-		fmt.Fprintf(buf, "%s%s", f1, f2)
+		printPeerConcise(buf, p)
 	}
 	return buf.String()
 }
 
+// printConciseHeader prints a concise header line representing nm to buf.
+//
+// If this function is changed to access different fields of nm, keep
+// in equalConciseHeader in sync.
+func (nm *NetworkMap) printConciseHeader(buf *strings.Builder) {
+	fmt.Fprintf(buf, "netmap: self: %v auth=%v",
+		nm.NodeKey.ShortString(), nm.MachineStatus)
+	if nm.LocalPort != 0 {
+		fmt.Fprintf(buf, " port=%v", nm.LocalPort)
+	}
+	if nm.Debug != nil {
+		j, _ := json.Marshal(nm.Debug)
+		fmt.Fprintf(buf, " debug=%s", j)
+	}
+	fmt.Fprintf(buf, " %v", nm.Addresses)
+	buf.WriteByte('\n')
+}
+
+// equalConciseHeader reports whether a and b are equal for the fields
+// used by printConciseHeader.
+func (a *NetworkMap) equalConciseHeader(b *NetworkMap) bool {
+	if a.NodeKey != b.NodeKey ||
+		a.MachineStatus != b.MachineStatus ||
+		a.LocalPort != b.LocalPort ||
+		len(a.Addresses) != len(b.Addresses) {
+		return false
+	}
+	for i, a := range a.Addresses {
+		if b.Addresses[i] != a {
+			return false
+		}
+	}
+	return (a.Debug == nil && b.Debug == nil) || reflect.DeepEqual(a.Debug, b.Debug)
+}
+
+// printPeerConcise appends to buf a line repsenting the peer p.
+//
+// If this function is changed to access different fields of p, keep
+// in nodeConciseEqual in sync.
+func printPeerConcise(buf *strings.Builder, p *tailcfg.Node) {
+	aip := make([]string, len(p.AllowedIPs))
+	for i, a := range p.AllowedIPs {
+		s := strings.TrimSuffix(fmt.Sprint(a), "/32")
+		aip[i] = s
+	}
+
+	ep := make([]string, len(p.Endpoints))
+	for i, e := range p.Endpoints {
+		// Align vertically on the ':' between IP and port
+		colon := strings.IndexByte(e, ':')
+		spaces := 0
+		for colon > 0 && len(e)+spaces-colon < 6 {
+			spaces++
+			colon--
+		}
+		ep[i] = fmt.Sprintf("%21v", e+strings.Repeat(" ", spaces))
+	}
+
+	derp := p.DERP
+	const derpPrefix = "127.3.3.40:"
+	if strings.HasPrefix(derp, derpPrefix) {
+		derp = "D" + derp[len(derpPrefix):]
+	}
+
+	// Most of the time, aip is just one element, so format the
+	// table to look good in that case. This will also make multi-
+	// subnet nodes stand out visually.
+	fmt.Fprintf(buf, " %v %-2v %-15v : %v\n",
+		p.Key.ShortString(), derp,
+		strings.Join(aip, " "),
+		strings.Join(ep, " "))
+}
+
+// nodeConciseEqual reports whether a and b are equal for the fields accessed by printPeerConcise.
+func nodeConciseEqual(a, b *tailcfg.Node) bool {
+	return a.Key == b.Key &&
+		a.DERP == b.DERP &&
+		eqCIDRsIgnoreNil(a.AllowedIPs, b.AllowedIPs) &&
+		eqStringsIgnoreNil(a.Endpoints, b.Endpoints)
+}
+
+func (b *NetworkMap) ConciseDiffFrom(a *NetworkMap) string {
+	var diff strings.Builder
+
+	// See if header (non-peers, "bare") part of the network map changed.
+	// If so, print its diff lines first.
+	if !a.equalConciseHeader(b) {
+		diff.WriteByte('-')
+		a.printConciseHeader(&diff)
+		diff.WriteByte('+')
+		b.printConciseHeader(&diff)
+	}
+
+	aps, bps := a.Peers, b.Peers
+	for len(aps) > 0 && len(bps) > 0 {
+		pa, pb := aps[0], bps[0]
+		switch {
+		case pa.ID == pb.ID:
+			if !nodeConciseEqual(pa, pb) {
+				diff.WriteByte('-')
+				printPeerConcise(&diff, pa)
+				diff.WriteByte('+')
+				printPeerConcise(&diff, pb)
+			}
+			aps, bps = aps[1:], bps[1:]
+		case pa.ID > pb.ID:
+			// New peer in b.
+			diff.WriteByte('+')
+			printPeerConcise(&diff, pb)
+			bps = bps[1:]
+		case pb.ID > pa.ID:
+			// Deleted peer in b.
+			diff.WriteByte('-')
+			printPeerConcise(&diff, pa)
+			aps = aps[1:]
+		}
+	}
+	for _, pa := range aps {
+		diff.WriteByte('-')
+		printPeerConcise(&diff, pa)
+	}
+	for _, pb := range bps {
+		diff.WriteByte('+')
+		printPeerConcise(&diff, pb)
+	}
+	return diff.String()
+}
+
 func (nm *NetworkMap) JSON() string {
 	b, err := json.MarshalIndent(*nm, "", "  ")
 	if err != nil {
@@ -104,184 +204,141 @@ func (nm *NetworkMap) JSON() string {
 	return string(b)
 }
 
-// TODO(apenwarr): delete me once relaynode doesn't need this anymore.
-// control.go:userMap() supercedes it. This does not belong in the client.
-func (nm *NetworkMap) UserMap() map[string][]filter.IP {
-	// Make a lookup table of roles
-	log.Printf("roles list is: %v\n", nm.Roles)
-	roles := make(map[tailcfg.RoleID]tailcfg.Role)
-	for _, role := range nm.Roles {
-		roles[role.ID] = role
-	}
-
-	// First, go through each node's addresses and make a lookup table
-	// of IP->User.
-	fwd := make(map[wgcfg.IP]string)
-	for _, node := range nm.Peers {
-		for _, addr := range node.Addresses {
-			if addr.Mask == 32 && addr.IP.Is4() {
-				user, ok := nm.UserProfiles[node.User]
-				if ok {
-					fwd[addr.IP] = user.LoginName
-				}
-			}
-		}
-	}
-
-	// Next, reverse the mapping into User->IP.
-	rev := make(map[string][]filter.IP)
-	for ip, username := range fwd {
-		ip4 := ip.To4()
-		if ip4 != nil {
-			fip := filter.NewIP(net.IP(ip4))
-			rev[username] = append(rev[username], fip)
-		}
-	}
-
-	// Now add roles, which are lists of users, and therefore lists
-	// of those users' IP addresses.
-	for _, user := range nm.UserProfiles {
-		for _, roleid := range user.Roles {
-			role, ok := roles[roleid]
-			if ok {
-				rolename := "role:" + role.Name
-				rev[rolename] = append(rev[rolename], rev[user.LoginName]...)
-			}
-		}
-	}
-
-	//log.Printf("Usermap is: %v\n", rev)
-	return rev
-}
-
-var iOS = runtime.GOOS == "darwin" && (runtime.GOARCH == "arm" || runtime.GOARCH == "arm64")
-var keepalive = !iOS
+// WGConfigFlags is a bitmask of flags to control the behavior of the
+// wireguard configuration generation done by NetMap.WGCfg.
+type WGConfigFlags int
 
 const (
-	UAllowSingleHosts = 1 << iota
-	UAllowSubnetRoutes
-	UAllowDefaultRoute
-	UHackDefaultRoute
-
-	UDefault = 0
+	AllowSingleHosts WGConfigFlags = 1 << iota
+	AllowSubnetRoutes
+	AllowDefaultRoute
+	HackDefaultRoute
 )
 
-// Several programs need to parse these arguments into uflags, so let's
-// centralize it here.
-func UFlagsHelper(uroutes, rroutes, droutes bool) int {
-	uflags := 0
-	if uroutes {
-		uflags |= UAllowSingleHosts
-	}
-	if rroutes {
-		uflags |= UAllowSubnetRoutes
-	}
-	if droutes {
-		uflags |= UAllowDefaultRoute
-	}
-	return uflags
-}
-
-func (nm *NetworkMap) UAPI(uflags int, dnsOverride []wgcfg.IP) string {
-	wgcfg, err := nm.WGCfg(uflags, dnsOverride)
+// TODO(bradfitz): UAPI seems to only be used by the old confnode and
+// pingnode; delete this when those are deleted/rewritten?
+func (nm *NetworkMap) UAPI(flags WGConfigFlags, dnsOverride []wgcfg.IP) string {
+	wgcfg, err := nm.WGCfg(log.Printf, flags, dnsOverride)
 	if err != nil {
-		log.Fatalf("WGCfg() failed unexpectedly: %v\n", err)
+		log.Fatalf("WGCfg() failed unexpectedly: %v", err)
 	}
 	s, err := wgcfg.ToUAPI()
 	if err != nil {
-		log.Fatalf("ToUAPI() failed unexpectedly: %v\n", err)
+		log.Fatalf("ToUAPI() failed unexpectedly: %v", err)
 	}
 	return s
 }
 
-func (nm *NetworkMap) WGCfg(uflags int, dnsOverride []wgcfg.IP) (*wgcfg.Config, error) {
-	s := nm._WireGuardConfig(uflags, dnsOverride, true)
-	return wgcfg.FromWgQuick(s, "tailscale")
-}
+// EndpointDiscoSuffix is appended to the hex representation of a peer's discovery key
+// and is then the sole wireguard endpoint for peers with a non-zero discovery key.
+// This form is then recognize by magicsock's CreateEndpoint.
+const EndpointDiscoSuffix = ".disco.tailscale:12345"
 
-// TODO(apenwarr): This mode is dangerous.
-// Discarding the extra endpoints is almost universally the wrong choice.
-// Except that plain wireguard can't handle a peer with multiple endpoints.
-// (Yet?)
-func (nm *NetworkMap) WireGuardConfigOneEndpoint(uflags int, dnsOverride []wgcfg.IP) string {
-	return nm._WireGuardConfig(uflags, dnsOverride, false)
-}
-
-func (nm *NetworkMap) _WireGuardConfig(uflags int, dnsOverride []wgcfg.IP, allEndpoints bool) string {
-	buf := new(strings.Builder)
-	fmt.Fprintf(buf, "[Interface]\n")
-	fmt.Fprintf(buf, "PrivateKey = %s\n", base64.StdEncoding.EncodeToString(nm.PrivateKey[:]))
-	if len(nm.Addresses) > 0 {
-		fmt.Fprintf(buf, "Address = ")
-		for i, cidr := range nm.Addresses {
-			if i > 0 {
-				fmt.Fprintf(buf, ", ")
-			}
-			fmt.Fprintf(buf, "%s", cidr)
-		}
-		fmt.Fprintf(buf, "\n")
+// WGCfg returns the NetworkMaps's Wireguard configuration.
+func (nm *NetworkMap) WGCfg(logf logger.Logf, flags WGConfigFlags, dnsOverride []wgcfg.IP) (*wgcfg.Config, error) {
+	cfg := &wgcfg.Config{
+		Name:       "tailscale",
+		PrivateKey: nm.PrivateKey,
+		Addresses:  nm.Addresses,
+		ListenPort: nm.LocalPort,
+		DNS:        append([]wgcfg.IP(nil), dnsOverride...),
+		Peers:      make([]wgcfg.Peer, 0, len(nm.Peers)),
 	}
-	fmt.Fprintf(buf, "ListenPort = %d\n", nm.LocalPort)
-	if len(dnsOverride) > 0 {
-		dnss := []string{}
-		for _, ip := range dnsOverride {
-			dnss = append(dnss, ip.String())
-		}
-		fmt.Fprintf(buf, "DNS = %s\n", strings.Join(dnss, ","))
-	}
-	fmt.Fprintf(buf, "\n")
 
-	for i, peer := range nm.Peers {
-		if (uflags&UAllowSingleHosts) == 0 && len(peer.AllowedIPs) < 2 {
-			log.Printf("wgcfg: %v skipping a single-host peer.\n", peer.Key.AbbrevString())
+	for _, peer := range nm.Peers {
+		if Debug.OnlyDisco && peer.DiscoKey.IsZero() {
 			continue
 		}
-		if i > 0 {
-			fmt.Fprintf(buf, "\n")
+		if (flags&AllowSingleHosts) == 0 && len(peer.AllowedIPs) < 2 {
+			logf("wgcfg: %v skipping a single-host peer.", peer.Key.ShortString())
+			continue
 		}
-		fmt.Fprintf(buf, "[Peer]\n")
-		fmt.Fprintf(buf, "PublicKey = %s\n", base64.StdEncoding.EncodeToString(peer.Key[:]))
-		if len(peer.Endpoints) > 0 {
-			if len(peer.Endpoints) == 1 {
-				fmt.Fprintf(buf, "Endpoint = %s", peer.Endpoints[0])
-			} else if allEndpoints {
-				// TODO(apenwarr): This mode is incompatible.
-				// Normal wireguard clients don't know how to
-				// parse it (yet?)
-				fmt.Fprintf(buf, "Endpoint = %s",
-					strings.Join(peer.Endpoints, ","))
-			} else {
-				fmt.Fprintf(buf, "Endpoint = %s # other endpoints: %s",
-					peer.Endpoints[0],
-					strings.Join(peer.Endpoints[1:], ", "))
+		cfg.Peers = append(cfg.Peers, wgcfg.Peer{
+			PublicKey: wgcfg.Key(peer.Key),
+		})
+		cpeer := &cfg.Peers[len(cfg.Peers)-1]
+		if peer.KeepAlive {
+			cpeer.PersistentKeepalive = 25 // seconds
+		}
+
+		if !peer.DiscoKey.IsZero() {
+			if err := appendEndpoint(cpeer, fmt.Sprintf("%x%s", peer.DiscoKey[:], EndpointDiscoSuffix)); err != nil {
+				return nil, err
+			}
+			cpeer.Endpoints = []wgcfg.Endpoint{{Host: fmt.Sprintf("%x.disco.tailscale", peer.DiscoKey[:]), Port: 12345}}
+		} else {
+			if err := appendEndpoint(cpeer, peer.DERP); err != nil {
+				return nil, err
+			}
+			for _, ep := range peer.Endpoints {
+				if err := appendEndpoint(cpeer, ep); err != nil {
+					return nil, err
+				}
 			}
-			buf.WriteByte('\n')
 		}
-		var aips []string
 		for _, allowedIP := range peer.AllowedIPs {
-			aip := allowedIP.String()
 			if allowedIP.Mask == 0 {
-				if (uflags & UAllowDefaultRoute) == 0 {
-					log.Printf("wgcfg: %v skipping default route\n", peer.Key.AbbrevString())
+				if (flags & AllowDefaultRoute) == 0 {
+					logf("wgcfg: %v skipping default route", peer.Key.ShortString())
 					continue
 				}
-				if (uflags & UHackDefaultRoute) != 0 {
-					aip = "10.0.0.0/8"
-					log.Printf("wgcfg: %v converting default route => %v\n", peer.Key.AbbrevString(), aip)
+				if (flags & HackDefaultRoute) != 0 {
+					allowedIP = wgcfg.CIDR{IP: wgcfg.IPv4(10, 0, 0, 0), Mask: 8}
+					logf("wgcfg: %v converting default route => %v", peer.Key.ShortString(), allowedIP.String())
 				}
 			} else if allowedIP.Mask < 32 {
-				if (uflags & UAllowSubnetRoutes) == 0 {
-					log.Printf("wgcfg: %v skipping subnet route\n", peer.Key.AbbrevString())
+				if (flags & AllowSubnetRoutes) == 0 {
+					logf("wgcfg: %v skipping subnet route", peer.Key.ShortString())
 					continue
 				}
 			}
-			aips = append(aips, aip)
-		}
-		fmt.Fprintf(buf, "AllowedIPs = %s\n", strings.Join(aips, ", "))
-		if keepalive {
-			fmt.Fprintf(buf, "PersistentKeepalive = 25\n")
+			cpeer.AllowedIPs = append(cpeer.AllowedIPs, allowedIP)
 		}
 	}
 
-	return buf.String()
+	return cfg, nil
+}
+
+func appendEndpoint(peer *wgcfg.Peer, epStr string) error {
+	if epStr == "" {
+		return nil
+	}
+	host, port, err := net.SplitHostPort(epStr)
+	if err != nil {
+		return fmt.Errorf("malformed endpoint %q for peer %v", epStr, peer.PublicKey.ShortString())
+	}
+	port16, err := strconv.ParseUint(port, 10, 16)
+	if err != nil {
+		return fmt.Errorf("invalid port in endpoint %q for peer %v", epStr, peer.PublicKey.ShortString())
+	}
+	peer.Endpoints = append(peer.Endpoints, wgcfg.Endpoint{Host: host, Port: uint16(port16)})
+	return nil
+}
+
+// eqStringsIgnoreNil reports whether a and b have the same length and
+// contents, but ignore whether a or b are nil.
+func eqStringsIgnoreNil(a, b []string) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	for i, v := range a {
+		if v != b[i] {
+			return false
+		}
+	}
+	return true
+}
+
+// eqCIDRsIgnoreNil reports whether a and b have the same length and
+// contents, but ignore whether a or b are nil.
+func eqCIDRsIgnoreNil(a, b []wgcfg.CIDR) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	for i, v := range a {
+		if v != b[i] {
+			return false
+		}
+	}
+	return true
 }

control/controlclient/netmap_test.go

@@ -0,0 +1,217 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"testing"
+
+	"tailscale.com/tailcfg"
+)
+
+func testNodeKey(b byte) (ret tailcfg.NodeKey) {
+	for i := range ret {
+		ret[i] = b
+	}
+	return
+}
+
+func TestNetworkMapConcise(t *testing.T) {
+	for _, tt := range []struct {
+		name string
+		nm   *NetworkMap
+		want string
+	}{
+		{
+			name: "basic",
+			nm: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+					{
+						Key:       testNodeKey(3),
+						DERP:      "127.3.3.40:4",
+						Endpoints: []string{"10.2.0.100:12", "10.1.0.100:12345"},
+					},
+				},
+			},
+			want: "netmap: self: [AQEBA] auth=machine-unknown []\n [AgICA] D2                 :    192.168.0.100:12     192.168.0.100:12354\n [AwMDA] D4                 :       10.2.0.100:12        10.1.0.100:12345\n",
+		},
+		{
+			name: "debug_non_nil",
+			nm: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Debug:   &tailcfg.Debug{},
+			},
+			want: "netmap: self: [AQEBA] auth=machine-unknown debug={} []\n",
+		},
+		{
+			name: "debug_values",
+			nm: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Debug:   &tailcfg.Debug{LogHeapPprof: true},
+			},
+			want: "netmap: self: [AQEBA] auth=machine-unknown debug={\"LogHeapPprof\":true} []\n",
+		},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			var got string
+			n := int(testing.AllocsPerRun(1000, func() {
+				got = tt.nm.Concise()
+			}))
+			t.Logf("Allocs = %d", n)
+			if got != tt.want {
+				t.Errorf("Wrong output\n Got: %q\nWant: %q\n## Got (unescaped):\n%s\n## Want (unescaped):\n%s\n", got, tt.want, got, tt.want)
+			}
+		})
+	}
+}
+
+func TestConciseDiffFrom(t *testing.T) {
+	for _, tt := range []struct {
+		name string
+		a, b *NetworkMap
+		want string
+	}{
+		{
+			name: "no_change",
+			a: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			b: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			want: "",
+		},
+		{
+			name: "header_change",
+			a: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			b: &NetworkMap{
+				NodeKey: testNodeKey(2),
+				Peers: []*tailcfg.Node{
+					{
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			want: "-netmap: self: [AQEBA] auth=machine-unknown []\n+netmap: self: [AgICA] auth=machine-unknown []\n",
+		},
+		{
+			name: "peer_add",
+			a: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						ID:        2,
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			b: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						ID:        1,
+						Key:       testNodeKey(1),
+						DERP:      "127.3.3.40:1",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+					{
+						ID:        2,
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+					{
+						ID:        3,
+						Key:       testNodeKey(3),
+						DERP:      "127.3.3.40:3",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			want: "+ [AQEBA] D1                 :    192.168.0.100:12     192.168.0.100:12354\n+ [AwMDA] D3                 :    192.168.0.100:12     192.168.0.100:12354\n",
+		},
+		{
+			name: "peer_remove",
+			a: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						ID:        1,
+						Key:       testNodeKey(1),
+						DERP:      "127.3.3.40:1",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+					{
+						ID:        2,
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+					{
+						ID:        3,
+						Key:       testNodeKey(3),
+						DERP:      "127.3.3.40:3",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			b: &NetworkMap{
+				NodeKey: testNodeKey(1),
+				Peers: []*tailcfg.Node{
+					{
+						ID:        2,
+						Key:       testNodeKey(2),
+						DERP:      "127.3.3.40:2",
+						Endpoints: []string{"192.168.0.100:12", "192.168.0.100:12354"},
+					},
+				},
+			},
+			want: "- [AQEBA] D1                 :    192.168.0.100:12     192.168.0.100:12354\n- [AwMDA] D3                 :    192.168.0.100:12     192.168.0.100:12354\n",
+		},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			var got string
+			n := int(testing.AllocsPerRun(50, func() {
+				got = tt.b.ConciseDiffFrom(tt.a)
+			}))
+			t.Logf("Allocs = %d", n)
+			if got != tt.want {
+				t.Errorf("Wrong output\n Got: %q\nWant: %q\n## Got (unescaped):\n%s\n## Want (unescaped):\n%s\n", got, tt.want, got, tt.want)
+			}
+		})
+	}
+}

derp/derp.go

@@ -32,10 +32,20 @@ const MaxPacketSize = 64 << 10
 const magic = "DERP🔑" // 8 bytes: 0x44 45 52 50 f0 9f 94 91
 
 const (
-	nonceLen   = 24
-	keyLen     = 32
-	maxInfoLen = 1 << 20
-	keepAlive  = 60 * time.Second
+	nonceLen       = 24
+	frameHeaderLen = 1 + 4 // frameType byte + 4 byte length
+	keyLen         = 32
+	maxInfoLen     = 1 << 20
+	keepAlive      = 60 * time.Second
+)
+
+// protocolVersion is bumped whenever there's a wire-incompatible change.
+//   * version 1 (zero on wire): consistent box headers, in use by employee dev nodes a bit
+//   * version 2: received packets have src addrs in frameRecvPacket at beginning
+const protocolVersion = 2
+
+const (
+	protocolSrcAddrs = 2 // protocol version at which client expects src addresses
 )
 
 // frameType is the one byte frame type at the beginning of the frame
@@ -58,12 +68,39 @@ Steady state:
 * server then sends frameRecvPacket to recipient
 */
 const (
-	frameServerKey  = frameType(0x01) // 8B magic + 32B public key + (0+ bytes future use)
-	frameClientInfo = frameType(0x02) // 32B pub key + 24B nonce + naclbox(json)
-	frameServerInfo = frameType(0x03) // 24B nonce + naclbox(json)
-	frameSendPacket = frameType(0x04) // 32B dest pub key + packet bytes
-	frameRecvPacket = frameType(0x05) // packet bytes
-	frameKeepAlive  = frameType(0x06) // no payload, no-op (to be replaced with ping/pong)
+	frameServerKey     = frameType(0x01) // 8B magic + 32B public key + (0+ bytes future use)
+	frameClientInfo    = frameType(0x02) // 32B pub key + 24B nonce + naclbox(json)
+	frameServerInfo    = frameType(0x03) // 24B nonce + naclbox(json)
+	frameSendPacket    = frameType(0x04) // 32B dest pub key + packet bytes
+	frameForwardPacket = frameType(0x0a) // 32B src pub key + 32B dst pub key + packet bytes
+	frameRecvPacket    = frameType(0x05) // v0/1: packet bytes, v2: 32B src pub key + packet bytes
+	frameKeepAlive     = frameType(0x06) // no payload, no-op (to be replaced with ping/pong)
+	frameNotePreferred = frameType(0x07) // 1 byte payload: 0x01 or 0x00 for whether this is client's home node
+
+	// framePeerGone is sent from server to client to signal that
+	// a previous sender is no longer connected. That is, if A
+	// sent to B, and then if A disconnects, the server sends
+	// framePeerGone to B so B can forget that a reverse path
+	// exists on that connection to get back to A.
+	framePeerGone = frameType(0x08) // 32B pub key of peer that's gone
+
+	// framePeerPresent is like framePeerGone, but for other
+	// members of the DERP region when they're meshed up together.
+	framePeerPresent = frameType(0x09) // 32B pub key of peer that's connected
+
+	// frameWatchConns is how one DERP node in a regional mesh
+	// subscribes to the others in the region.
+	// There's no payload. If the sender doesn't have permission, the connection
+	// is closed. Otherwise, the client is initially flooded with
+	// framePeerPresent for all connected nodes, and then a stream of
+	// framePeerPresent & framePeerGone has peers connect and disconnect.
+	frameWatchConns = frameType(0x10)
+
+	// frameClosePeer is a privileged frame type (requires the
+	// mesh key for now) that closes the provided peer's
+	// connection. (To be used for cluster load balancing
+	// purposes, when clients end up on a non-ideal node)
+	frameClosePeer = frameType(0x11) // 32B pub key of peer to close.
 )
 
 var bin = binary.BigEndian
@@ -121,7 +158,8 @@ func readFrame(br *bufio.Reader, maxSize uint32, b []byte) (t frameType, frameLe
 	if frameLen > maxSize {
 		return 0, 0, fmt.Errorf("frame header size %d exceeds reader limit of %d", frameLen, maxSize)
 	}
-	n, err := io.ReadFull(br, b[:frameLen])
+
+	n, err := io.ReadFull(br, b[:minUint32(frameLen, uint32(len(b)))])
 	if err != nil {
 		return 0, 0, err
 	}
@@ -155,3 +193,17 @@ func writeFrame(bw *bufio.Writer, t frameType, b []byte) error {
 	}
 	return bw.Flush()
 }
+
+func minInt(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+func minUint32(a, b uint32) uint32 {
+	if a < b {
+		return a
+	}
+	return b
+}

derp/derp_client.go

@@ -11,7 +11,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"net"
+	"sync"
 	"time"
 
 	"golang.org/x/crypto/nacl/box"
@@ -19,18 +19,57 @@ import (
 	"tailscale.com/types/logger"
 )
 
+// Client is a DERP client.
 type Client struct {
-	serverKey  key.Public // of the DERP server; not a machine or node key
-	privateKey key.Private
-	publicKey  key.Public // of privateKey
-	logf       logger.Logf
-	nc         net.Conn
-	br         *bufio.Reader
-	bw         *bufio.Writer
-	readErr    error // sticky read error
+	serverKey    key.Public // of the DERP server; not a machine or node key
+	privateKey   key.Private
+	publicKey    key.Public // of privateKey
+	protoVersion int        // min of server+client
+	logf         logger.Logf
+	nc           Conn
+	br           *bufio.Reader
+	meshKey      string
+
+	wmu sync.Mutex // hold while writing to bw
+	bw  *bufio.Writer
+
+	// Owned by Recv:
+	peeked  int   // bytes to discard on next Recv
+	readErr error // sticky read error
 }
 
-func NewClient(privateKey key.Private, nc net.Conn, brw *bufio.ReadWriter, logf logger.Logf) (*Client, error) {
+// ClientOpt is an option passed to NewClient.
+type ClientOpt interface {
+	update(*clientOpt)
+}
+
+type clientOptFunc func(*clientOpt)
+
+func (f clientOptFunc) update(o *clientOpt) { f(o) }
+
+// clientOpt are the options passed to newClient.
+type clientOpt struct {
+	MeshKey string
+}
+
+// MeshKey returns a ClientOpt to pass to the DERP server during connect to get
+// access to join the mesh.
+//
+// An empty key means to not use a mesh key.
+func MeshKey(key string) ClientOpt { return clientOptFunc(func(o *clientOpt) { o.MeshKey = key }) }
+
+func NewClient(privateKey key.Private, nc Conn, brw *bufio.ReadWriter, logf logger.Logf, opts ...ClientOpt) (*Client, error) {
+	var opt clientOpt
+	for _, o := range opts {
+		if o == nil {
+			return nil, errors.New("nil ClientOpt")
+		}
+		o.update(&opt)
+	}
+	return newClient(privateKey, nc, brw, logf, opt)
+}
+
+func newClient(privateKey key.Private, nc Conn, brw *bufio.ReadWriter, logf logger.Logf, opt clientOpt) (*Client, error) {
 	c := &Client{
 		privateKey: privateKey,
 		publicKey:  privateKey.Public(),
@@ -38,19 +77,19 @@ func NewClient(privateKey key.Private, nc net.Conn, brw *bufio.ReadWriter, logf
 		nc:         nc,
 		br:         brw.Reader,
 		bw:         brw.Writer,
+		meshKey:    opt.MeshKey,
 	}
-
 	if err := c.recvServerKey(); err != nil {
 		return nil, fmt.Errorf("derp.Client: failed to receive server key: %v", err)
 	}
 	if err := c.sendClientKey(); err != nil {
 		return nil, fmt.Errorf("derp.Client: failed to send client key: %v", err)
 	}
-	_, err := c.recvServerInfo()
+	info, err := c.recvServerInfo()
 	if err != nil {
 		return nil, fmt.Errorf("derp.Client: failed to receive server info: %v", err)
 	}
-
+	c.protoVersion = minInt(protocolVersion, info.Version)
 	return c, nil
 }
 
@@ -104,12 +143,28 @@ func (c *Client) recvServerInfo() (*serverInfo, error) {
 	return info, nil
 }
 
+type clientInfo struct {
+	Version int // `json:"version,omitempty"`
+
+	// MeshKey optionally specifies a pre-shared key used by
+	// trusted clients.  It's required to subscribe to the
+	// connection list & forward packets. It's empty for regular
+	// users.
+	MeshKey string // `json:"meshKey,omitempty"`
+}
+
 func (c *Client) sendClientKey() error {
 	var nonce [nonceLen]byte
 	if _, err := crand.Read(nonce[:]); err != nil {
 		return err
 	}
-	msg := []byte("{}") // no clientInfo for now
+	msg, err := json.Marshal(clientInfo{
+		Version: protocolVersion,
+		MeshKey: c.meshKey,
+	})
+	if err != nil {
+		return err
+	}
 	msgbox := box.Seal(nil, msg, &nonce, c.serverKey.B32(), c.privateKey.B32())
 
 	buf := make([]byte, 0, nonceLen+keyLen+len(msgbox))
@@ -119,6 +174,9 @@ func (c *Client) sendClientKey() error {
 	return writeFrame(c.bw, frameClientInfo, buf)
 }
 
+// ServerPublicKey returns the server's public key.
+func (c *Client) ServerPublicKey() key.Public { return c.serverKey }
+
 // Send sends a packet to the Tailscale node identified by dstKey.
 //
 // It is an error if the packet is larger than 64KB.
@@ -127,7 +185,7 @@ func (c *Client) Send(dstKey key.Public, pkt []byte) error { return c.send(dstKe
 func (c *Client) send(dstKey key.Public, pkt []byte) (ret error) {
 	defer func() {
 		if ret != nil {
-			ret = fmt.Errorf("derp.Send: %v", ret)
+			ret = fmt.Errorf("derp.Send: %w", ret)
 		}
 	}()
 
@@ -135,6 +193,9 @@ func (c *Client) send(dstKey key.Public, pkt []byte) (ret error) {
 		return fmt.Errorf("packet too big: %d", len(pkt))
 	}
 
+	c.wmu.Lock()
+	defer c.wmu.Unlock()
+
 	if err := writeFrameHeader(c.bw, frameSendPacket, uint32(len(dstKey)+len(pkt))); err != nil {
 		return err
 	}
@@ -147,6 +208,85 @@ func (c *Client) send(dstKey key.Public, pkt []byte) (ret error) {
 	return c.bw.Flush()
 }
 
+func (c *Client) ForwardPacket(srcKey, dstKey key.Public, pkt []byte) (err error) {
+	defer func() {
+		if err != nil {
+			err = fmt.Errorf("derp.ForwardPacket: %w", err)
+		}
+	}()
+
+	if len(pkt) > MaxPacketSize {
+		return fmt.Errorf("packet too big: %d", len(pkt))
+	}
+
+	c.wmu.Lock()
+	defer c.wmu.Unlock()
+
+	timer := time.AfterFunc(5*time.Second, c.writeTimeoutFired)
+	defer timer.Stop()
+
+	if err := writeFrameHeader(c.bw, frameForwardPacket, uint32(keyLen*2+len(pkt))); err != nil {
+		return err
+	}
+	if _, err := c.bw.Write(srcKey[:]); err != nil {
+		return err
+	}
+	if _, err := c.bw.Write(dstKey[:]); err != nil {
+		return err
+	}
+	if _, err := c.bw.Write(pkt); err != nil {
+		return err
+	}
+	return c.bw.Flush()
+}
+
+func (c *Client) writeTimeoutFired() { c.nc.Close() }
+
+// NotePreferred sends a packet that tells the server whether this
+// client is the user's preferred server. This is only used in the
+// server for stats.
+func (c *Client) NotePreferred(preferred bool) (err error) {
+	defer func() {
+		if err != nil {
+			err = fmt.Errorf("derp.NotePreferred: %v", err)
+		}
+	}()
+
+	c.wmu.Lock()
+	defer c.wmu.Unlock()
+
+	if err := writeFrameHeader(c.bw, frameNotePreferred, 1); err != nil {
+		return err
+	}
+	var b byte = 0x00
+	if preferred {
+		b = 0x01
+	}
+	if err := c.bw.WriteByte(b); err != nil {
+		return err
+	}
+	return c.bw.Flush()
+}
+
+// WatchConnectionChanges sends a request to subscribe to the peer's connection list.
+// It's a fatal error if the client wasn't created using MeshKey.
+func (c *Client) WatchConnectionChanges() error {
+	c.wmu.Lock()
+	defer c.wmu.Unlock()
+	if err := writeFrameHeader(c.bw, frameWatchConns, 0); err != nil {
+		return err
+	}
+	return c.bw.Flush()
+}
+
+// ClosePeer asks the server to close target's TCP connection.
+// It's a fatal error if the client wasn't created using MeshKey.
+func (c *Client) ClosePeer(target key.Public) error {
+	c.wmu.Lock()
+	defer c.wmu.Unlock()
+	return writeFrame(c.bw, frameClosePeer, target[:])
+}
+
 // ReceivedMessage represents a type returned by Client.Recv. Unless
 // otherwise documented, the returned message aliases the byte slice
 // provided to Recv and thus the message is only as good as that
@@ -156,31 +296,89 @@ type ReceivedMessage interface {
 }
 
 // ReceivedPacket is a ReceivedMessage representing an incoming packet.
-type ReceivedPacket []byte
+type ReceivedPacket struct {
+	Source key.Public
+	// Data is the received packet bytes. It aliases the memory
+	// passed to Client.Recv.
+	Data []byte
+}
 
 func (ReceivedPacket) msg() {}
 
+// PeerGoneMessage is a ReceivedMessage that indicates that the client
+// identified by the underlying public key had previously sent you a
+// packet but has now disconnected from the server.
+type PeerGoneMessage key.Public
+
+func (PeerGoneMessage) msg() {}
+
+// PeerPresentMessage is a ReceivedMessage that indicates that the client
+// is connected to the server. (Only used by trusted mesh clients)
+type PeerPresentMessage key.Public
+
+func (PeerPresentMessage) msg() {}
+
 // Recv reads a message from the DERP server.
-// The provided buffer must be large enough to receive a complete packet,
-// which in practice are are 1.5-4 KB, but can be up to 64 KB.
+//
+// The returned message may alias memory owned by the Client; it
+// should only be accessed until the next call to Client.
+//
 // Once Recv returns an error, the Client is dead forever.
-func (c *Client) Recv(b []byte) (m ReceivedMessage, err error) {
+func (c *Client) Recv() (m ReceivedMessage, err error) {
+	return c.recvTimeout(120 * time.Second)
+}
+
+func (c *Client) recvTimeout(timeout time.Duration) (m ReceivedMessage, err error) {
 	if c.readErr != nil {
 		return nil, c.readErr
 	}
 	defer func() {
 		if err != nil {
-			err = fmt.Errorf("derp.Recv: %v", err)
+			err = fmt.Errorf("derp.Recv: %w", err)
 			c.readErr = err
 		}
 	}()
 
 	for {
-		c.nc.SetReadDeadline(time.Now().Add(120 * time.Second))
-		t, n, err := readFrame(c.br, 1<<20, b)
+		c.nc.SetReadDeadline(time.Now().Add(timeout))
+
+		// Discard any peeked bytes from a previous Recv call.
+		if c.peeked != 0 {
+			if n, err := c.br.Discard(c.peeked); err != nil || n != c.peeked {
+				// Documented to never fail, but might as well check.
+				return nil, fmt.Errorf("bufio.Reader.Discard(%d bytes): got %v, %v", c.peeked, n, err)
+			}
+			c.peeked = 0
+		}
+
+		t, n, err := readFrameHeader(c.br)
 		if err != nil {
 			return nil, err
 		}
+		if n > 1<<20 {
+			return nil, fmt.Errorf("unexpectedly large frame of %d bytes returned", n)
+		}
+
+		var b []byte // frame payload (past the 5 byte header)
+
+		// If the frame fits in our bufio.Reader buffer, just use it.
+		// In practice it's 4KB (from derphttp.Client's bufio.NewReader(httpConn)) and
+		// in practive, WireGuard packets (and thus DERP frames) are under 1.5KB.
+		// So This is the common path.
+		if int(n) <= c.br.Size() {
+			b, err = c.br.Peek(int(n))
+			c.peeked = int(n)
+		} else {
+			// But if for some reason we read a large DERP message (which isn't necessarily
+			// a Wireguard packet), then just allocate memory for it.
+			// TODO(bradfitz): use a pool if large frames ever happen in practice.
+			b = make([]byte, n)
+			_, err = io.ReadFull(c.br, b)
+		}
+		if err != nil {
+			return nil, err
+		}
+
 		switch t {
 		default:
 			continue
@@ -188,8 +386,37 @@ func (c *Client) Recv(b []byte) (m ReceivedMessage, err error) {
 			// TODO: eventually we'll have server->client pings that
 			// require ack pongs.
 			continue
+		case framePeerGone:
+			if n < keyLen {
+				c.logf("[unexpected] dropping short peerGone frame from DERP server")
+				continue
+			}
+			var pg PeerGoneMessage
+			copy(pg[:], b[:keyLen])
+			return pg, nil
+
+		case framePeerPresent:
+			if n < keyLen {
+				c.logf("[unexpected] dropping short peerPresent frame from DERP server")
+				continue
+			}
+			var pg PeerPresentMessage
+			copy(pg[:], b[:keyLen])
+			return pg, nil
+
 		case frameRecvPacket:
-			return ReceivedPacket(b[:n]), nil
+			var rp ReceivedPacket
+			if c.protoVersion < protocolSrcAddrs {
+				rp.Data = b[:n]
+			} else {
+				if n < keyLen {
+					c.logf("[unexpected] dropping short packet from DERP server")
+					continue
+				}
+				copy(rp.Source[:], b[:keyLen])
+				rp.Data = b[keyLen:n]
+			}
+			return rp, nil
 		}
 	}
 }

derp/derp_test.go

@@ -6,15 +6,25 @@ package derp
 
 import (
 	"bufio"
+	"context"
 	crand "crypto/rand"
+	"errors"
+	"expvar"
+	"fmt"
+	"io"
 	"net"
+	"reflect"
+	"sync"
 	"testing"
 	"time"
 
+	"tailscale.com/net/nettest"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 )
 
 func newPrivateKey(t *testing.T) (k key.Private) {
+	t.Helper()
 	if _, err := crand.Read(k[:]); err != nil {
 		t.Fatal(err)
 	}
@@ -35,13 +45,14 @@ func TestSendRecv(t *testing.T) {
 		clientKeys = append(clientKeys, priv.Public())
 	}
 
-	ln, err := net.Listen("tcp", ":0")
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
 	if err != nil {
 		t.Fatal(err)
 	}
 	defer ln.Close()
 
 	var clients []*Client
+	var connsOut []Conn
 	var recvChs []chan []byte
 	errCh := make(chan error, 3)
 
@@ -52,13 +63,15 @@ func TestSendRecv(t *testing.T) {
 			t.Fatal(err)
 		}
 		defer cout.Close()
+		connsOut = append(connsOut, cout)
 
 		cin, err := ln.Accept()
 		if err != nil {
 			t.Fatal(err)
 		}
 		defer cin.Close()
-		go s.Accept(cin, bufio.NewReadWriter(bufio.NewReader(cin), bufio.NewWriter(cin)))
+		brwServer := bufio.NewReadWriter(bufio.NewReader(cin), bufio.NewWriter(cin))
+		go s.Accept(cin, brwServer, fmt.Sprintf("test-client-%d", i))
 
 		key := clientPrivateKeys[i]
 		brw := bufio.NewReadWriter(bufio.NewReader(cout), bufio.NewWriter(cout))
@@ -71,12 +84,13 @@ func TestSendRecv(t *testing.T) {
 		t.Logf("Connected client %d.", i)
 	}
 
+	var peerGoneCount expvar.Int
+
 	t.Logf("Starting read loops")
 	for i := 0; i < numClients; i++ {
 		go func(i int) {
 			for {
-				b := make([]byte, 1<<16)
-				m, err := clients[i].Recv(b)
+				m, err := clients[i].Recv()
 				if err != nil {
 					errCh <- err
 					return
@@ -85,8 +99,13 @@ func TestSendRecv(t *testing.T) {
 				default:
 					t.Errorf("unexpected message type %T", m)
 					continue
+				case PeerGoneMessage:
+					peerGoneCount.Add(1)
 				case ReceivedPacket:
-					recvChs[i] <- []byte(m)
+					if m.Source.IsZero() {
+						t.Errorf("zero Source address in ReceivedPacket")
+					}
+					recvChs[i] <- append([]byte(nil), m.Data...)
 				}
 			}
 		}(i)
@@ -112,6 +131,32 @@ func TestSendRecv(t *testing.T) {
 		}
 	}
 
+	wantActive := func(total, home int64) {
+		t.Helper()
+		dl := time.Now().Add(5 * time.Second)
+		var gotTotal, gotHome int64
+		for time.Now().Before(dl) {
+			gotTotal, gotHome = s.curClients.Value(), s.curHomeClients.Value()
+			if gotTotal == total && gotHome == home {
+				return
+			}
+			time.Sleep(10 * time.Millisecond)
+		}
+		t.Errorf("total/home=%v/%v; want %v/%v", gotTotal, gotHome, total, home)
+	}
+
+	wantClosedPeers := func(want int64) {
+		t.Helper()
+		var got int64
+		dl := time.Now().Add(5 * time.Second)
+		for time.Now().Before(dl) {
+			if got = peerGoneCount.Value(); got == want {
+				return
+			}
+		}
+		t.Errorf("peer gone count = %v; want %v", got, want)
+	}
+
 	msg1 := []byte("hello 0->1\n")
 	if err := clients[0].Send(clientKeys[1], msg1); err != nil {
 		t.Fatal(err)
@@ -128,6 +173,572 @@ func TestSendRecv(t *testing.T) {
 	recvNothing(0)
 	recvNothing(1)
 
+	wantActive(3, 0)
+	clients[0].NotePreferred(true)
+	wantActive(3, 1)
+	clients[0].NotePreferred(true)
+	wantActive(3, 1)
+	clients[0].NotePreferred(false)
+	wantActive(3, 0)
+	clients[0].NotePreferred(false)
+	wantActive(3, 0)
+	clients[1].NotePreferred(true)
+	wantActive(3, 1)
+	connsOut[1].Close()
+	wantActive(2, 0)
+	wantClosedPeers(1)
+	clients[2].NotePreferred(true)
+	wantActive(2, 1)
+	clients[2].NotePreferred(false)
+	wantActive(2, 0)
+	connsOut[2].Close()
+	wantActive(1, 0)
+	wantClosedPeers(1)
+
 	t.Logf("passed")
 	s.Close()
+
+}
+
+func TestSendFreeze(t *testing.T) {
+	serverPrivateKey := newPrivateKey(t)
+	s := NewServer(serverPrivateKey, t.Logf)
+	defer s.Close()
+	s.WriteTimeout = 100 * time.Millisecond
+
+	// We send two streams of messages:
+	//
+	//	alice --> bob
+	//	alice --> cathy
+	//
+	// Then cathy stops processing messsages.
+	// That should not interfere with alice talking to bob.
+
+	newClient := func(name string, k key.Private) (c *Client, clientConn nettest.Conn) {
+		t.Helper()
+		c1, c2 := nettest.NewConn(name, 1024)
+		go s.Accept(c1, bufio.NewReadWriter(bufio.NewReader(c1), bufio.NewWriter(c1)), name)
+
+		brw := bufio.NewReadWriter(bufio.NewReader(c2), bufio.NewWriter(c2))
+		c, err := NewClient(k, c2, brw, t.Logf)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return c, c2
+	}
+
+	aliceKey := newPrivateKey(t)
+	aliceClient, aliceConn := newClient("alice", aliceKey)
+
+	bobKey := newPrivateKey(t)
+	bobClient, bobConn := newClient("bob", bobKey)
+
+	cathyKey := newPrivateKey(t)
+	cathyClient, cathyConn := newClient("cathy", cathyKey)
+
+	var (
+		aliceCh = make(chan struct{}, 32)
+		bobCh   = make(chan struct{}, 32)
+		cathyCh = make(chan struct{}, 32)
+	)
+	chs := func(name string) chan struct{} {
+		switch name {
+		case "alice":
+			return aliceCh
+		case "bob":
+			return bobCh
+		case "cathy":
+			return cathyCh
+		default:
+			panic("unknown ch: " + name)
+		}
+	}
+
+	errCh := make(chan error, 4)
+	recv := func(name string, client *Client) {
+		ch := chs(name)
+		for {
+			m, err := client.Recv()
+			if err != nil {
+				errCh <- fmt.Errorf("%s: %w", name, err)
+				return
+			}
+			switch m := m.(type) {
+			default:
+				errCh <- fmt.Errorf("%s: unexpected message type %T", name, m)
+				return
+			case ReceivedPacket:
+				if m.Source.IsZero() {
+					errCh <- fmt.Errorf("%s: zero Source address in ReceivedPacket", name)
+					return
+				}
+				select {
+				case ch <- struct{}{}:
+				default:
+				}
+			}
+		}
+	}
+	go recv("alice", aliceClient)
+	go recv("bob", bobClient)
+	go recv("cathy", cathyClient)
+
+	var cancel func()
+	go func() {
+		t := time.NewTicker(2 * time.Millisecond)
+		defer t.Stop()
+		var ctx context.Context
+		ctx, cancel = context.WithCancel(context.Background())
+		for {
+			select {
+			case <-t.C:
+			case <-ctx.Done():
+				errCh <- nil
+				return
+			}
+
+			msg1 := []byte("hello alice->bob\n")
+			if err := aliceClient.Send(bobKey.Public(), msg1); err != nil {
+				errCh <- fmt.Errorf("alice send to bob: %w", err)
+				return
+			}
+			msg2 := []byte("hello alice->cathy\n")
+
+			// TODO: an error is expected here.
+			// We ignore it, maybe we should log it somehow?
+			aliceClient.Send(cathyKey.Public(), msg2)
+		}
+	}()
+
+	drainAny := func(ch chan struct{}) {
+		// We are draining potentially infinite sources,
+		// so place some reasonable upper limit.
+		//
+		// The important thing here is to make sure that
+		// if any tokens remain in the channel, they
+		// must have been generated after drainAny was
+		// called.
+		for i := 0; i < cap(ch); i++ {
+			select {
+			case <-ch:
+			default:
+				return
+			}
+		}
+	}
+	drain := func(t *testing.T, name string) bool {
+		t.Helper()
+		timer := time.NewTimer(1 * time.Second)
+		defer timer.Stop()
+
+		// Ensure ch has at least one element.
+		ch := chs(name)
+		select {
+		case <-ch:
+		case <-timer.C:
+			t.Errorf("no packet received by %s", name)
+			return false
+		}
+		// Drain remaining.
+		drainAny(ch)
+		return true
+	}
+	isEmpty := func(t *testing.T, name string) {
+		t.Helper()
+		select {
+		case <-chs(name):
+			t.Errorf("packet received by %s, want none", name)
+		default:
+		}
+	}
+
+	t.Run("initial send", func(t *testing.T) {
+		drain(t, "bob")
+		drain(t, "cathy")
+		isEmpty(t, "alice")
+	})
+
+	t.Run("block cathy", func(t *testing.T) {
+		// Block cathy. Now the cathyConn buffer will fill up quickly,
+		// and the derp server will back up.
+		cathyConn.SetReadBlock(true)
+		time.Sleep(2 * s.WriteTimeout)
+
+		drain(t, "bob")
+		drainAny(chs("cathy"))
+		isEmpty(t, "alice")
+
+		// Now wait a little longer, and ensure packets still flow to bob
+		if !drain(t, "bob") {
+			t.Errorf("connection alice->bob frozen by alice->cathy")
+		}
+	})
+
+	// Cleanup, make sure we process all errors.
+	t.Logf("TEST COMPLETE, cancelling sender")
+	cancel()
+	t.Logf("closing connections")
+	aliceConn.Close()
+	bobConn.Close()
+	cathyConn.Close()
+
+	for i := 0; i < cap(errCh); i++ {
+		err := <-errCh
+		if err != nil {
+			if errors.Is(err, io.EOF) {
+				continue
+			}
+			t.Error(err)
+		}
+	}
+}
+
+type testServer struct {
+	s    *Server
+	ln   net.Listener
+	logf logger.Logf
+
+	mu      sync.Mutex
+	pubName map[key.Public]string
+	clients map[*testClient]bool
+}
+
+func (ts *testServer) addTestClient(c *testClient) {
+	ts.mu.Lock()
+	defer ts.mu.Unlock()
+	ts.clients[c] = true
+}
+
+func (ts *testServer) addKeyName(k key.Public, name string) {
+	ts.mu.Lock()
+	defer ts.mu.Unlock()
+	ts.pubName[k] = name
+	ts.logf("test adding named key %q for %x", name, k)
+}
+
+func (ts *testServer) keyName(k key.Public) string {
+	ts.mu.Lock()
+	defer ts.mu.Unlock()
+	if name, ok := ts.pubName[k]; ok {
+		return name
+	}
+	return k.ShortString()
+}
+
+func (ts *testServer) close(t *testing.T) error {
+	ts.ln.Close()
+	ts.s.Close()
+	for c := range ts.clients {
+		c.close(t)
+	}
+	return nil
+}
+
+func newTestServer(t *testing.T) *testServer {
+	t.Helper()
+	logf := logger.WithPrefix(t.Logf, "derp-server: ")
+	s := NewServer(newPrivateKey(t), logf)
+	s.SetMeshKey("mesh-key")
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	go func() {
+		i := 0
+		for {
+			i++
+			c, err := ln.Accept()
+			if err != nil {
+				return
+			}
+			// TODO: register c in ts so Close also closes it?
+			go func(i int) {
+				brwServer := bufio.NewReadWriter(bufio.NewReader(c), bufio.NewWriter(c))
+				go s.Accept(c, brwServer, fmt.Sprintf("test-client-%d", i))
+			}(i)
+		}
+	}()
+	return &testServer{
+		s:       s,
+		ln:      ln,
+		logf:    logf,
+		clients: map[*testClient]bool{},
+		pubName: map[key.Public]string{},
+	}
+}
+
+type testClient struct {
+	name   string
+	c      *Client
+	nc     net.Conn
+	pub    key.Public
+	ts     *testServer
+	closed bool
+}
+
+func newTestClient(t *testing.T, ts *testServer, name string, newClient func(net.Conn, key.Private, logger.Logf) (*Client, error)) *testClient {
+	t.Helper()
+	nc, err := net.Dial("tcp", ts.ln.Addr().String())
+	if err != nil {
+		t.Fatal(err)
+	}
+	key := newPrivateKey(t)
+	ts.addKeyName(key.Public(), name)
+	c, err := newClient(nc, key, logger.WithPrefix(t.Logf, "client-"+name+": "))
+	if err != nil {
+		t.Fatal(err)
+	}
+	tc := &testClient{
+		name: name,
+		nc:   nc,
+		c:    c,
+		ts:   ts,
+		pub:  key.Public(),
+	}
+	ts.addTestClient(tc)
+	return tc
+}
+
+func newRegularClient(t *testing.T, ts *testServer, name string) *testClient {
+	return newTestClient(t, ts, name, func(nc net.Conn, priv key.Private, logf logger.Logf) (*Client, error) {
+		brw := bufio.NewReadWriter(bufio.NewReader(nc), bufio.NewWriter(nc))
+		return NewClient(priv, nc, brw, logf)
+	})
+}
+
+func newTestWatcher(t *testing.T, ts *testServer, name string) *testClient {
+	return newTestClient(t, ts, name, func(nc net.Conn, priv key.Private, logf logger.Logf) (*Client, error) {
+		brw := bufio.NewReadWriter(bufio.NewReader(nc), bufio.NewWriter(nc))
+		c, err := NewClient(priv, nc, brw, logf, MeshKey("mesh-key"))
+		if err != nil {
+			return nil, err
+		}
+		if err := c.WatchConnectionChanges(); err != nil {
+			return nil, err
+		}
+		return c, nil
+	})
+}
+
+func (tc *testClient) wantPresent(t *testing.T, peers ...key.Public) {
+	t.Helper()
+	want := map[key.Public]bool{}
+	for _, k := range peers {
+		want[k] = true
+	}
+
+	for {
+		m, err := tc.c.recvTimeout(time.Second)
+		if err != nil {
+			t.Fatal(err)
+		}
+		switch m := m.(type) {
+		case PeerPresentMessage:
+			got := key.Public(m)
+			if !want[got] {
+				t.Fatalf("got peer present for %v; want present for %v", tc.ts.keyName(got), logger.ArgWriter(func(bw *bufio.Writer) {
+					for _, pub := range peers {
+						fmt.Fprintf(bw, "%s ", tc.ts.keyName(pub))
+					}
+				}))
+			}
+			delete(want, got)
+			if len(want) == 0 {
+				return
+			}
+		default:
+			t.Fatalf("unexpected message type %T", m)
+		}
+	}
+}
+
+func (tc *testClient) wantGone(t *testing.T, peer key.Public) {
+	t.Helper()
+	m, err := tc.c.recvTimeout(time.Second)
+	if err != nil {
+		t.Fatal(err)
+	}
+	switch m := m.(type) {
+	case PeerGoneMessage:
+		got := key.Public(m)
+		if peer != got {
+			t.Errorf("got gone message for %v; want gone for %v", tc.ts.keyName(got), tc.ts.keyName(peer))
+		}
+	default:
+		t.Fatalf("unexpected message type %T", m)
+	}
+}
+
+func (c *testClient) close(t *testing.T) {
+	t.Helper()
+	if c.closed {
+		return
+	}
+	c.closed = true
+	t.Logf("closing client %q (%x)", c.name, c.pub)
+	c.nc.Close()
+}
+
+// TestWatch tests the connection watcher mechanism used by regional
+// DERP nodes to mesh up with each other.
+func TestWatch(t *testing.T) {
+	ts := newTestServer(t)
+	defer ts.close(t)
+
+	w1 := newTestWatcher(t, ts, "w1")
+	w1.wantPresent(t, w1.pub)
+
+	c1 := newRegularClient(t, ts, "c1")
+	w1.wantPresent(t, c1.pub)
+
+	c2 := newRegularClient(t, ts, "c2")
+	w1.wantPresent(t, c2.pub)
+
+	w2 := newTestWatcher(t, ts, "w2")
+	w1.wantPresent(t, w2.pub)
+	w2.wantPresent(t, w1.pub, w2.pub, c1.pub, c2.pub)
+
+	c3 := newRegularClient(t, ts, "c3")
+	w1.wantPresent(t, c3.pub)
+	w2.wantPresent(t, c3.pub)
+
+	c2.close(t)
+	w1.wantGone(t, c2.pub)
+	w2.wantGone(t, c2.pub)
+
+	w3 := newTestWatcher(t, ts, "w3")
+	w1.wantPresent(t, w3.pub)
+	w2.wantPresent(t, w3.pub)
+	w3.wantPresent(t, c1.pub, c3.pub, w1.pub, w2.pub, w3.pub)
+
+	c1.close(t)
+	w1.wantGone(t, c1.pub)
+	w2.wantGone(t, c1.pub)
+	w3.wantGone(t, c1.pub)
+}
+
+type testFwd int
+
+func (testFwd) ForwardPacket(key.Public, key.Public, []byte) error { panic("not called in tests") }
+
+func pubAll(b byte) (ret key.Public) {
+	for i := range ret {
+		ret[i] = b
+	}
+	return
+}
+
+func TestForwarderRegistration(t *testing.T) {
+	s := &Server{
+		clients:     make(map[key.Public]*sclient),
+		clientsMesh: map[key.Public]PacketForwarder{},
+	}
+	want := func(want map[key.Public]PacketForwarder) {
+		t.Helper()
+		if got := s.clientsMesh; !reflect.DeepEqual(got, want) {
+			t.Fatalf("mismatch\n got: %v\nwant: %v\n", got, want)
+		}
+	}
+	wantCounter := func(c *expvar.Int, want int) {
+		t.Helper()
+		if got := c.Value(); got != int64(want) {
+			t.Errorf("counter = %v; want %v", got, want)
+		}
+	}
+
+	u1 := pubAll(1)
+	u2 := pubAll(2)
+	u3 := pubAll(3)
+
+	s.AddPacketForwarder(u1, testFwd(1))
+	s.AddPacketForwarder(u2, testFwd(2))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(1),
+		u2: testFwd(2),
+	})
+
+	// Verify a remove of non-registered forwarder is no-op.
+	s.RemovePacketForwarder(u2, testFwd(999))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(1),
+		u2: testFwd(2),
+	})
+
+	// Verify a remove of non-registered user is no-op.
+	s.RemovePacketForwarder(u3, testFwd(1))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(1),
+		u2: testFwd(2),
+	})
+
+	// Actual removal.
+	s.RemovePacketForwarder(u2, testFwd(2))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(1),
+	})
+
+	// Adding a dup for a user.
+	wantCounter(&s.multiForwarderCreated, 0)
+	s.AddPacketForwarder(u1, testFwd(100))
+	want(map[key.Public]PacketForwarder{
+		u1: multiForwarder{
+			testFwd(1):   1,
+			testFwd(100): 2,
+		},
+	})
+	wantCounter(&s.multiForwarderCreated, 1)
+
+	// Removing a forwarder in a multi set that doesn't exist; does nothing.
+	s.RemovePacketForwarder(u1, testFwd(55))
+	want(map[key.Public]PacketForwarder{
+		u1: multiForwarder{
+			testFwd(1):   1,
+			testFwd(100): 2,
+		},
+	})
+
+	// Removing a forwarder in a multi set that does exist should collapse it away
+	// from being a multiForwarder.
+	wantCounter(&s.multiForwarderDeleted, 0)
+	s.RemovePacketForwarder(u1, testFwd(1))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(100),
+	})
+	wantCounter(&s.multiForwarderDeleted, 1)
+
+	// Removing an entry for a client that's still connected locally should result
+	// in a nil forwarder.
+	u1c := &sclient{
+		key:  u1,
+		logf: logger.Discard,
+	}
+	s.clients[u1] = u1c
+	s.RemovePacketForwarder(u1, testFwd(100))
+	want(map[key.Public]PacketForwarder{
+		u1: nil,
+	})
+
+	// But once that client disconnects, it should go away.
+	s.unregisterClient(u1c)
+	want(map[key.Public]PacketForwarder{})
+
+	// But if it already has a forwarder, it's not removed.
+	s.AddPacketForwarder(u1, testFwd(2))
+	s.unregisterClient(u1c)
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(2),
+	})
+
+	// Now pretend u1 was already connected locally (so clientsMesh[u1] is nil), and then we heard
+	// that they're also connected to a peer of ours. That sholdn't transition the forwarder
+	// from nil to the new one, not a multiForwarder.
+	s.clients[u1] = u1c
+	s.clientsMesh[u1] = nil
+	want(map[key.Public]PacketForwarder{
+		u1: nil,
+	})
+	s.AddPacketForwarder(u1, testFwd(3))
+	want(map[key.Public]PacketForwarder{
+		u1: testFwd(3),
+	})
 }

derp/derphttp/derphttp_client.go

@@ -12,18 +12,24 @@ package derphttp
 
 import (
 	"bufio"
-	"bytes"
 	"context"
 	"crypto/tls"
 	"errors"
 	"fmt"
+	"io"
 	"io/ioutil"
 	"net"
 	"net/http"
 	"net/url"
 	"sync"
+	"time"
 
+	"inet.af/netaddr"
 	"tailscale.com/derp"
+	"tailscale.com/net/dnscache"
+	"tailscale.com/net/netns"
+	"tailscale.com/net/tlsdial"
+	"tailscale.com/tailcfg"
 	"tailscale.com/types/key"
 	"tailscale.com/types/logger"
 )
@@ -35,32 +41,67 @@ import (
 // Send/Recv will completely re-establish the connection (unless Close
 // has been called).
 type Client struct {
+	TLSConfig *tls.Config        // optional; nil means default
+	DNSCache  *dnscache.Resolver // optional; nil means no caching
+	MeshKey   string             // optional; for trusted clients
+
 	privateKey key.Private
 	logf       logger.Logf
-	closed     chan struct{}
-	url        *url.URL
-	resp       *http.Response
 
-	netConnMu sync.Mutex
-	netConn   net.Conn
+	// Either url or getRegion is non-nil:
+	url       *url.URL
+	getRegion func() *tailcfg.DERPRegion
 
-	clientMu sync.Mutex
-	client   *derp.Client
+	ctx       context.Context // closed via cancelCtx in Client.Close
+	cancelCtx context.CancelFunc
+
+	mu           sync.Mutex
+	preferred    bool
+	closed       bool
+	netConn      io.Closer
+	client       *derp.Client
+	connGen      int // incremented once per new connection; valid values are >0
+	serverPubKey key.Public
+}
+
+// NewRegionClient returns a new DERP-over-HTTP client. It connects lazily.
+// To trigger a connection, use Connect.
+func NewRegionClient(privateKey key.Private, logf logger.Logf, getRegion func() *tailcfg.DERPRegion) *Client {
+	ctx, cancel := context.WithCancel(context.Background())
+	c := &Client{
+		privateKey: privateKey,
+		logf:       logf,
+		getRegion:  getRegion,
+		ctx:        ctx,
+		cancelCtx:  cancel,
+	}
+	return c
+}
+
+// NewNetcheckClient returns a Client that's only able to have its DialRegion method called.
+// It's used by the netcheck package.
+func NewNetcheckClient(logf logger.Logf) *Client {
+	return &Client{logf: logf}
 }
 
 // NewClient returns a new DERP-over-HTTP client. It connects lazily.
-// To trigger a connection use Connect.
+// To trigger a connection, use Connect.
 func NewClient(privateKey key.Private, serverURL string, logf logger.Logf) (*Client, error) {
 	u, err := url.Parse(serverURL)
 	if err != nil {
 		return nil, fmt.Errorf("derphttp.NewClient: %v", err)
 	}
+	if urlPort(u) == "" {
+		return nil, fmt.Errorf("derphttp.NewClient: invalid URL scheme %q", u.Scheme)
+	}
 
+	ctx, cancel := context.WithCancel(context.Background())
 	c := &Client{
 		privateKey: privateKey,
 		logf:       logf,
 		url:        u,
-		closed:     make(chan struct{}),
+		ctx:        ctx,
+		cancelCtx:  cancel,
 	}
 	return c, nil
 }
@@ -68,149 +109,508 @@ func NewClient(privateKey key.Private, serverURL string, logf logger.Logf) (*Cli
 // Connect connects or reconnects to the server, unless already connected.
 // It returns nil if there was already a good connection, or if one was made.
 func (c *Client) Connect(ctx context.Context) error {
-	_, err := c.connect(ctx, "derphttp.Client.Connect")
+	_, _, err := c.connect(ctx, "derphttp.Client.Connect")
 	return err
 }
 
-func (c *Client) connect(ctx context.Context, caller string) (client *derp.Client, err error) {
-	// TODO: use ctx for TCP+TLS+HTTP below
-	select {
-	case <-c.closed:
-		return nil, ErrClientClosed
-	default:
+// ServerPublicKey returns the server's public key.
+//
+// It only returns a non-zero value once a connection has succeeded
+// from an earlier call.
+func (c *Client) ServerPublicKey() key.Public {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.serverPubKey
+}
+
+func urlPort(u *url.URL) string {
+	if p := u.Port(); p != "" {
+		return p
 	}
+	switch u.Scheme {
+	case "https":
+		return "443"
+	case "http":
+		return "80"
+	}
+	return ""
+}
 
-	c.clientMu.Lock()
-	defer c.clientMu.Unlock()
+func (c *Client) targetString(reg *tailcfg.DERPRegion) string {
+	if c.url != nil {
+		return c.url.String()
+	}
+	return fmt.Sprintf("region %d (%v)", reg.RegionID, reg.RegionCode)
+}
 
+func (c *Client) useHTTPS() bool {
+	if c.url != nil && c.url.Scheme == "http" {
+		return false
+	}
+	return true
+}
+
+// tlsServerName returns the tls.Config.ServerName value (for the TLS ClientHello).
+func (c *Client) tlsServerName(node *tailcfg.DERPNode) string {
+	if c.url != nil {
+		return c.url.Host
+	}
+	return node.HostName
+}
+
+func (c *Client) urlString(node *tailcfg.DERPNode) string {
+	if c.url != nil {
+		return c.url.String()
+	}
+	return fmt.Sprintf("https://%s/derp", node.HostName)
+}
+
+func (c *Client) connect(ctx context.Context, caller string) (client *derp.Client, connGen int, err error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.closed {
+		return nil, 0, ErrClientClosed
+	}
 	if c.client != nil {
-		return c.client, nil
+		return c.client, c.connGen, nil
 	}
 
-	c.logf("%s: connecting", caller)
+	// timeout is the fallback maximum time (if ctx doesn't limit
+	// it further) to do all of: DNS + TCP + TLS + HTTP Upgrade +
+	// DERP upgrade.
+	const timeout = 10 * time.Second
+	ctx, cancel := context.WithTimeout(ctx, timeout)
+	go func() {
+		select {
+		case <-ctx.Done():
+			// Either timeout fired (handled below), or
+			// we're returning via the defer cancel()
+			// below.
+		case <-c.ctx.Done():
+			// Propagate a Client.Close call into
+			// cancelling this context.
+			cancel()
+		}
+	}()
+	defer cancel()
+
+	var reg *tailcfg.DERPRegion // nil when using c.url to dial
+	if c.getRegion != nil {
+		reg = c.getRegion()
+		if reg == nil {
+			return nil, 0, errors.New("DERP region not available")
+		}
+	}
+
+	var tcpConn net.Conn
 
-	var netConn net.Conn
 	defer func() {
 		if err != nil {
-			err = fmt.Errorf("%s connect: %v", caller, err)
-			if netConn != nil {
-				netConn.Close()
+			if ctx.Err() != nil {
+				err = fmt.Errorf("%v: %v", ctx.Err(), err)
+			}
+			err = fmt.Errorf("%s connect to %v: %v", caller, c.targetString(reg), err)
+			if tcpConn != nil {
+				go tcpConn.Close()
 			}
 		}
 	}()
 
-	if c.url.Scheme == "https" {
-		port := c.url.Port()
-		if port == "" {
-			port = "443"
-		}
-		config := &tls.Config{}
-		var tlsConn *tls.Conn
-		tlsConn, err = tls.Dial("tcp", net.JoinHostPort(c.url.Host, port), config)
-		if tlsConn != nil {
-			netConn = tlsConn
-		}
+	var node *tailcfg.DERPNode // nil when using c.url to dial
+	if c.url != nil {
+		c.logf("%s: connecting to %v", caller, c.url)
+		tcpConn, err = c.dialURL(ctx)
 	} else {
-		netConn, err = net.Dial("tcp", c.url.Host)
+		c.logf("%s: connecting to derp-%d (%v)", caller, reg.RegionID, reg.RegionCode)
+		tcpConn, node, err = c.dialRegion(ctx, reg)
 	}
 	if err != nil {
-		return nil, err
+		return nil, 0, err
 	}
 
-	c.netConnMu.Lock()
-	c.netConn = netConn
-	c.netConnMu.Unlock()
+	// Now that we have a TCP connection, force close it if the
+	// TLS handshake + DERP setup takes too long.
+	done := make(chan struct{})
+	defer close(done)
+	go func() {
+		select {
+		case <-done:
+			// Normal path. Upgrade occurred in time.
+		case <-ctx.Done():
+			select {
+			case <-done:
+				// Normal path. Upgrade occurred in time.
+				// But the ctx.Done() is also done because
+				// the "defer cancel()" above scheduled
+				// before this goroutine.
+			default:
+				// The TLS or HTTP or DERP exchanges didn't complete
+				// in time. Force close the TCP connection to force
+				// them to fail quickly.
+				tcpConn.Close()
+			}
+		}
+	}()
 
-	conn := bufio.NewReadWriter(bufio.NewReader(netConn), bufio.NewWriter(netConn))
+	var httpConn net.Conn // a TCP conn or a TLS conn; what we speak HTTP to
+	if c.useHTTPS() {
+		httpConn = c.tlsClient(tcpConn, node)
+	} else {
+		httpConn = tcpConn
+	}
 
-	req, err := http.NewRequest("GET", c.url.String(), nil)
+	brw := bufio.NewReadWriter(bufio.NewReader(httpConn), bufio.NewWriter(httpConn))
+
+	req, err := http.NewRequest("GET", c.urlString(node), nil)
 	if err != nil {
-		return nil, err
+		return nil, 0, err
 	}
-	req.Header.Set("Upgrade", "WebSocket")
+	req.Header.Set("Upgrade", "DERP")
 	req.Header.Set("Connection", "Upgrade")
-	if err := req.Write(conn); err != nil {
-		return nil, err
+
+	if err := req.Write(brw); err != nil {
+		return nil, 0, err
 	}
-	if err := conn.Flush(); err != nil {
-		return nil, err
+	if err := brw.Flush(); err != nil {
+		return nil, 0, err
 	}
 
-	resp, err := http.ReadResponse(conn.Reader, req)
+	resp, err := http.ReadResponse(brw.Reader, req)
 	if err != nil {
-		return nil, err
+		return nil, 0, err
 	}
 	if resp.StatusCode != http.StatusSwitchingProtocols {
 		b, _ := ioutil.ReadAll(resp.Body)
 		resp.Body.Close()
-		return nil, fmt.Errorf("GET failed: %v: %s", err, b)
+		return nil, 0, fmt.Errorf("GET failed: %v: %s", err, b)
 	}
-	resp.Body = ioutil.NopCloser(bytes.NewReader([]byte{}))
 
-	derpClient, err := derp.NewClient(c.privateKey, netConn, conn, c.logf)
+	derpClient, err := derp.NewClient(c.privateKey, httpConn, brw, c.logf, derp.MeshKey(c.MeshKey))
 	if err != nil {
-		return nil, err
+		return nil, 0, err
 	}
-	c.resp = resp
+	if c.preferred {
+		if err := derpClient.NotePreferred(true); err != nil {
+			go httpConn.Close()
+			return nil, 0, err
+		}
+	}
+
+	c.serverPubKey = derpClient.ServerPublicKey()
 	c.client = derpClient
-	return c.client, nil
+	c.netConn = tcpConn
+	c.connGen++
+	return c.client, c.connGen, nil
+}
+
+func (c *Client) dialURL(ctx context.Context) (net.Conn, error) {
+	host := c.url.Hostname()
+	hostOrIP := host
+
+	dialer := netns.NewDialer()
+
+	if c.DNSCache != nil {
+		ip, err := c.DNSCache.LookupIP(ctx, host)
+		if err == nil {
+			hostOrIP = ip.String()
+		}
+		if err != nil && netns.IsSOCKSDialer(dialer) {
+			// Return an error if we're not using a dial
+			// proxy that can do DNS lookups for us.
+			return nil, err
+		}
+	}
+
+	tcpConn, err := dialer.DialContext(ctx, "tcp", net.JoinHostPort(hostOrIP, urlPort(c.url)))
+	if err != nil {
+		return nil, fmt.Errorf("dial of %v: %v", host, err)
+	}
+	return tcpConn, nil
+}
+
+// dialRegion returns a TCP connection to the provided region, trying
+// each node in order (with dialNode) until one connects or ctx is
+// done.
+func (c *Client) dialRegion(ctx context.Context, reg *tailcfg.DERPRegion) (net.Conn, *tailcfg.DERPNode, error) {
+	if len(reg.Nodes) == 0 {
+		return nil, nil, fmt.Errorf("no nodes for %s", c.targetString(reg))
+	}
+	var firstErr error
+	for _, n := range reg.Nodes {
+		if n.STUNOnly {
+			if firstErr == nil {
+				firstErr = fmt.Errorf("no non-STUNOnly nodes for %s", c.targetString(reg))
+			}
+			continue
+		}
+		c, err := c.dialNode(ctx, n)
+		if err == nil {
+			return c, n, nil
+		}
+		if firstErr == nil {
+			firstErr = err
+		}
+	}
+	return nil, nil, firstErr
+}
+
+func (c *Client) tlsClient(nc net.Conn, node *tailcfg.DERPNode) *tls.Conn {
+	tlsConf := tlsdial.Config(c.tlsServerName(node), c.TLSConfig)
+	if node != nil {
+		if node.DERPTestPort != 0 {
+			tlsConf.InsecureSkipVerify = true
+		}
+		if node.CertName != "" {
+			tlsdial.SetConfigExpectedCert(tlsConf, node.CertName)
+		}
+	}
+	return tls.Client(nc, tlsConf)
+}
+
+func (c *Client) DialRegionTLS(ctx context.Context, reg *tailcfg.DERPRegion) (tlsConn *tls.Conn, connClose io.Closer, err error) {
+	tcpConn, node, err := c.dialRegion(ctx, reg)
+	if err != nil {
+		return nil, nil, err
+	}
+	done := make(chan bool) // unbufferd
+	defer close(done)
+
+	tlsConn = c.tlsClient(tcpConn, node)
+	go func() {
+		select {
+		case <-done:
+		case <-ctx.Done():
+			tcpConn.Close()
+		}
+	}()
+	err = tlsConn.Handshake()
+	if err != nil {
+		return nil, nil, err
+	}
+	select {
+	case done <- true:
+		return tlsConn, tcpConn, nil
+	case <-ctx.Done():
+		return nil, nil, ctx.Err()
+	}
+}
+
+func (c *Client) dialContext(ctx context.Context, proto, addr string) (net.Conn, error) {
+	return netns.NewDialer().DialContext(ctx, proto, addr)
+}
+
+// shouldDialProto reports whether an explicitly provided IPv4 or IPv6
+// address (given in s) is valid. An empty value means to dial, but to
+// use DNS. The predicate function reports whether the non-empty
+// string s contained a valid IP address of the right family.
+func shouldDialProto(s string, pred func(netaddr.IP) bool) bool {
+	if s == "" {
+		return true
+	}
+	ip, _ := netaddr.ParseIP(s)
+	return pred(ip)
+}
+
+const dialNodeTimeout = 1500 * time.Millisecond
+
+// dialNode returns a TCP connection to node n, racing IPv4 and IPv6
+// (both as applicable) against each other.
+// A node is only given dialNodeTimeout to connect.
+//
+// TODO(bradfitz): longer if no options remain perhaps? ...  Or longer
+// overall but have dialRegion start overlapping races?
+func (c *Client) dialNode(ctx context.Context, n *tailcfg.DERPNode) (net.Conn, error) {
+	type res struct {
+		c   net.Conn
+		err error
+	}
+	resc := make(chan res) // must be unbuffered
+	ctx, cancel := context.WithTimeout(ctx, dialNodeTimeout)
+	defer cancel()
+
+	nwait := 0
+	startDial := func(dstPrimary, proto string) {
+		nwait++
+		go func() {
+			dst := dstPrimary
+			if dst == "" {
+				dst = n.HostName
+			}
+			port := "443"
+			if n.DERPTestPort != 0 {
+				port = fmt.Sprint(n.DERPTestPort)
+			}
+			c, err := c.dialContext(ctx, proto, net.JoinHostPort(dst, port))
+			select {
+			case resc <- res{c, err}:
+			case <-ctx.Done():
+				if c != nil {
+					c.Close()
+				}
+			}
+		}()
+	}
+	if shouldDialProto(n.IPv4, netaddr.IP.Is4) {
+		startDial(n.IPv4, "tcp4")
+	}
+	if shouldDialProto(n.IPv6, netaddr.IP.Is6) {
+		startDial(n.IPv6, "tcp6")
+	}
+	if nwait == 0 {
+		return nil, errors.New("both IPv4 and IPv6 are explicitly disabled for node")
+	}
+
+	var firstErr error
+	for {
+		select {
+		case res := <-resc:
+			nwait--
+			if res.err == nil {
+				return res.c, nil
+			}
+			if firstErr == nil {
+				firstErr = res.err
+			}
+			if nwait == 0 {
+				return nil, firstErr
+			}
+		case <-ctx.Done():
+			return nil, ctx.Err()
+		}
+	}
 }
 
 func (c *Client) Send(dstKey key.Public, b []byte) error {
-	client, err := c.connect(context.TODO(), "derphttp.Client.Send")
+	client, _, err := c.connect(context.TODO(), "derphttp.Client.Send")
 	if err != nil {
 		return err
 	}
 	if err := client.Send(dstKey, b); err != nil {
-		c.close()
+		c.closeForReconnect(client)
 	}
 	return err
 }
 
-func (c *Client) Recv(b []byte) (derp.ReceivedMessage, error) {
-	client, err := c.connect(context.TODO(), "derphttp.Client.Recv")
+func (c *Client) ForwardPacket(from, to key.Public, b []byte) error {
+	client, _, err := c.connect(context.TODO(), "derphttp.Client.ForwardPacket")
 	if err != nil {
-		return nil, err
+		return err
 	}
-	m, err := client.Recv(b)
+	if err := client.ForwardPacket(from, to, b); err != nil {
+		c.closeForReconnect(client)
+	}
+	return err
+}
+
+// NotePreferred notes whether this Client is the caller's preferred
+// (home) DERP node. It's only used for stats.
+func (c *Client) NotePreferred(v bool) {
+	c.mu.Lock()
+	if c.preferred == v {
+		c.mu.Unlock()
+		return
+	}
+	c.preferred = v
+	client := c.client
+	c.mu.Unlock()
+
+	if client != nil {
+		if err := client.NotePreferred(v); err != nil {
+			c.closeForReconnect(client)
+		}
+	}
+}
+
+// WatchConnectionChanges sends a request to subscribe to
+// notifications about clients connecting & disconnecting.
+//
+// Only trusted connections (using MeshKey) are allowed to use this.
+func (c *Client) WatchConnectionChanges() error {
+	client, _, err := c.connect(context.TODO(), "derphttp.Client.WatchConnectionChanges")
 	if err != nil {
-		c.close()
+		return err
 	}
+	err = client.WatchConnectionChanges()
+	if err != nil {
+		c.closeForReconnect(client)
+	}
+	return err
+}
+
+// ClosePeer asks the server to close target's TCP connection.
+//
+// Only trusted connections (using MeshKey) are allowed to use this.
+func (c *Client) ClosePeer(target key.Public) error {
+	client, _, err := c.connect(context.TODO(), "derphttp.Client.ClosePeer")
+	if err != nil {
+		return err
+	}
+	err = client.ClosePeer(target)
+	if err != nil {
+		c.closeForReconnect(client)
+	}
+	return err
+}
+
+// Recv reads a message from c. The returned message may alias memory from Client.
+// The message should only be used until the next Client call.
+func (c *Client) Recv() (derp.ReceivedMessage, error) {
+	m, _, err := c.RecvDetail()
 	return m, err
 }
 
+// RecvDetail is like Recv, but additional returns the connection generation on each message.
+// The connGen value is incremented every time the derphttp.Client reconnects to the server.
+func (c *Client) RecvDetail() (m derp.ReceivedMessage, connGen int, err error) {
+	client, connGen, err := c.connect(context.TODO(), "derphttp.Client.Recv")
+	if err != nil {
+		return nil, 0, err
+	}
+	m, err = client.Recv()
+	if err != nil {
+		c.closeForReconnect(client)
+	}
+	return m, connGen, err
+}
+
 // Close closes the client. It will not automatically reconnect after
 // being closed.
 func (c *Client) Close() error {
-	select {
-	case <-c.closed:
+	c.cancelCtx() // not in lock, so it can cancel Connect, which holds mu
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.closed {
 		return ErrClientClosed
-	default:
 	}
-	close(c.closed)
-	c.close()
+	c.closed = true
+	if c.netConn != nil {
+		c.netConn.Close()
+	}
 	return nil
 }
 
-func (c *Client) close() {
-	c.netConnMu.Lock()
-	netConn := c.netConn
-	c.netConnMu.Unlock()
-
-	if netConn != nil {
-		netConn.Close()
-	}
-
-	c.clientMu.Lock()
-	defer c.clientMu.Unlock()
-	if c.client == nil {
+// closeForReconnect closes the underlying network connection and
+// zeros out the client field so future calls to Connect will
+// reconnect.
+//
+// The provided brokenClient is the client to forget. If current
+// client is not brokenClient, closeForReconnect does nothing. (This
+// prevents a send and receive goroutine from failing at the ~same
+// time and both calling closeForReconnect and the caller goroutines
+// forever calling closeForReconnect in lockstep endlessly;
+// https://github.com/tailscale/tailscale/pull/264)
+func (c *Client) closeForReconnect(brokenClient *derp.Client) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.client != brokenClient {
 		return
 	}
-	c.resp = nil
+	if c.netConn != nil {
+		c.netConn.Close()
+		c.netConn = nil
+	}
 	c.client = nil
-	c.netConnMu.Lock()
-	c.netConn = nil
-	c.netConnMu.Unlock()
 }
 
 var ErrClientClosed = errors.New("derphttp.Client closed")

derp/derphttp/derphttp_server.go

@@ -5,6 +5,7 @@
 package derphttp
 
 import (
+	"log"
 	"net/http"
 
 	"tailscale.com/derp"
@@ -12,11 +13,11 @@ import (
 
 func Handler(s *derp.Server) http.Handler {
 	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.Header.Get("Upgrade") != "WebSocket" {
+		if p := r.Header.Get("Upgrade"); p != "WebSocket" && p != "DERP" {
 			http.Error(w, "DERP requires connection upgrade", http.StatusUpgradeRequired)
 			return
 		}
-		w.Header().Set("Upgrade", "WebSocket")
+		w.Header().Set("Upgrade", "DERP")
 		w.Header().Set("Connection", "Upgrade")
 		w.WriteHeader(http.StatusSwitchingProtocols)
 
@@ -27,9 +28,10 @@ func Handler(s *derp.Server) http.Handler {
 		}
 		netConn, conn, err := h.Hijack()
 		if err != nil {
+			log.Printf("Hijack failed: %v", err)
 			http.Error(w, "HTTP does not support general TCP support", 500)
 			return
 		}
-		s.Accept(netConn, conn)
+		s.Accept(netConn, conn, netConn.RemoteAddr().String())
 	})
 }

derp/derphttp/derphttp_test.go

@@ -93,8 +93,7 @@ func TestSendRecv(t *testing.T) {
 					return
 				default:
 				}
-				b := make([]byte, 1<<16)
-				m, err := c.Recv(b)
+				m, err := c.Recv()
 				if err != nil {
 					t.Logf("client%d: %v", i, err)
 					break
@@ -103,8 +102,10 @@ func TestSendRecv(t *testing.T) {
 				default:
 					t.Errorf("unexpected message type %T", m)
 					continue
+				case derp.PeerGoneMessage:
+					// Ignore.
 				case derp.ReceivedPacket:
-					recvChs[i] <- []byte(m)
+					recvChs[i] <- append([]byte(nil), m.Data...)
 				}
 			}
 		}(i)

derp/derphttp/mesh_client.go

@@ -0,0 +1,122 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package derphttp
+
+import (
+	"sync"
+	"time"
+
+	"tailscale.com/derp"
+	"tailscale.com/types/key"
+)
+
+// RunWatchConnectionLoop loops forever, sending WatchConnectionChanges and subscribing to
+// connection changes.
+//
+// If the server's public key is ignoreServerKey, RunWatchConnectionLoop returns.
+//
+// Otherwise, the add and remove funcs are called as clients come & go.
+func (c *Client) RunWatchConnectionLoop(ignoreServerKey key.Public, add, remove func(key.Public)) {
+	logf := c.logf
+	const retryInterval = 5 * time.Second
+	const statusInterval = 10 * time.Second
+	var (
+		mu              sync.Mutex
+		present         = map[key.Public]bool{}
+		loggedConnected = false
+	)
+	clear := func() {
+		mu.Lock()
+		defer mu.Unlock()
+		if len(present) == 0 {
+			return
+		}
+		logf("reconnected; clearing %d forwarding mappings", len(present))
+		for k := range present {
+			remove(k)
+		}
+		present = map[key.Public]bool{}
+	}
+	lastConnGen := 0
+	lastStatus := time.Now()
+	logConnectedLocked := func() {
+		if loggedConnected {
+			return
+		}
+		logf("connected; %d peers", len(present))
+		loggedConnected = true
+	}
+
+	const logConnectedDelay = 200 * time.Millisecond
+	timer := time.AfterFunc(2*time.Second, func() {
+		mu.Lock()
+		defer mu.Unlock()
+		logConnectedLocked()
+	})
+	defer timer.Stop()
+
+	updatePeer := func(k key.Public, isPresent bool) {
+		if isPresent {
+			add(k)
+		} else {
+			remove(k)
+		}
+
+		mu.Lock()
+		defer mu.Unlock()
+		if isPresent {
+			present[k] = true
+			if !loggedConnected {
+				timer.Reset(logConnectedDelay)
+			}
+		} else {
+			// If we got a peerGone message, that means the initial connection's
+			// flood of peerPresent messages is done, so we can log already:
+			logConnectedLocked()
+			delete(present, k)
+		}
+	}
+
+	for {
+		err := c.WatchConnectionChanges()
+		if err != nil {
+			clear()
+			logf("WatchConnectionChanges: %v", err)
+			time.Sleep(retryInterval)
+			continue
+		}
+
+		if c.ServerPublicKey() == ignoreServerKey {
+			logf("detected self-connect; ignoring host")
+			return
+		}
+		for {
+			m, connGen, err := c.RecvDetail()
+			if err != nil {
+				clear()
+				logf("Recv: %v", err)
+				time.Sleep(retryInterval)
+				break
+			}
+			if connGen != lastConnGen {
+				lastConnGen = connGen
+				clear()
+			}
+			switch m := m.(type) {
+			case derp.PeerPresentMessage:
+				updatePeer(key.Public(m), true)
+			case derp.PeerGoneMessage:
+				updatePeer(key.Public(m), false)
+			default:
+				continue
+			}
+			if now := time.Now(); now.Sub(lastStatus) > statusInterval {
+				lastStatus = now
+				logf("%d peers", len(present))
+			}
+		}
+	}
+
+}

derp/derpmap/derpmap.go

@@ -0,0 +1,65 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package derpmap contains information about Tailscale.com's production DERP nodes.
+package derpmap
+
+import (
+	"fmt"
+	"strings"
+
+	"tailscale.com/tailcfg"
+)
+
+func derpNode(suffix, v4, v6 string) *tailcfg.DERPNode {
+	return &tailcfg.DERPNode{
+		Name:     suffix, // updated later
+		RegionID: 0,      // updated later
+		IPv4:     v4,
+		IPv6:     v6,
+	}
+}
+
+func derpRegion(id int, code string, nodes ...*tailcfg.DERPNode) *tailcfg.DERPRegion {
+	region := &tailcfg.DERPRegion{
+		RegionID:   id,
+		RegionCode: code,
+		Nodes:      nodes,
+	}
+	for _, n := range nodes {
+		n.Name = fmt.Sprintf("%d%s", id, n.Name)
+		n.RegionID = id
+		n.HostName = fmt.Sprintf("derp%s.tailscale.com", strings.TrimSuffix(n.Name, "a"))
+	}
+	return region
+}
+
+// Prod returns Tailscale's map of relay servers.
+//
+// This list is only used by cmd/tailscale's netcheck subcommand. In
+// normal operation the Tailscale nodes get this sent to them from the
+// control server.
+//
+// This list is subject to change and should not be relied on.
+func Prod() *tailcfg.DERPMap {
+	return &tailcfg.DERPMap{
+		Regions: map[int]*tailcfg.DERPRegion{
+			1: derpRegion(1, "nyc",
+				derpNode("a", "159.89.225.99", "2604:a880:400:d1::828:b001"),
+			),
+			2: derpRegion(2, "sfo",
+				derpNode("a", "167.172.206.31", "2604:a880:2:d1::c5:7001"),
+			),
+			3: derpRegion(3, "sin",
+				derpNode("a", "68.183.179.66", "2400:6180:0:d1::67d:8001"),
+			),
+			4: derpRegion(4, "fra",
+				derpNode("a", "167.172.182.26", "2a03:b0c0:3:e0::36e:9001"),
+			),
+			5: derpRegion(5, "syd",
+				derpNode("a", "103.43.75.49", "2001:19f0:5801:10b7:5400:2ff:feaa:284c"),
+			),
+		},
+	}
+}

disco/disco.go

@@ -0,0 +1,179 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package disco contains the discovery message types.
+//
+// A discovery message is:
+//
+// Header:
+//     magic          [6]byte  // “TS💬” (0x54 53 f0 9f 92 ac)
+//     senderDiscoPub [32]byte // nacl public key
+//     nonce          [24]byte
+//
+// The recipient then decrypts the bytes following (the nacl secretbox)
+// and then the inner payload structure is:
+//
+//     messageType    byte  (the MessageType constants below)
+//     messageVersion byte  (0 for now; but always ignore bytes at the end)
+//     message-paylod [...]byte
+package disco
+
+import (
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"net"
+
+	"inet.af/netaddr"
+)
+
+// Magic is the 6 byte header of all discovery messages.
+const Magic = "TS💬" // 6 bytes: 0x54 53 f0 9f 92 ac
+
+const keyLen = 32
+
+// NonceLen is the length of the nonces used by nacl secretboxes.
+const NonceLen = 24
+
+type MessageType byte
+
+const (
+	TypePing        = MessageType(0x01)
+	TypePong        = MessageType(0x02)
+	TypeCallMeMaybe = MessageType(0x03)
+)
+
+const v0 = byte(0)
+
+var errShort = errors.New("short message")
+
+// LooksLikeDiscoWrapper reports whether p looks like it's a packet
+// containing an encrypted disco message.
+func LooksLikeDiscoWrapper(p []byte) bool {
+	if len(p) < len(Magic)+keyLen+NonceLen {
+		return false
+	}
+	return string(p[:len(Magic)]) == Magic
+}
+
+// Parse parses the encrypted part of the message from inside the
+// nacl secretbox.
+func Parse(p []byte) (Message, error) {
+	if len(p) < 2 {
+		return nil, errShort
+	}
+	t, ver, p := MessageType(p[0]), p[1], p[2:]
+	switch t {
+	case TypePing:
+		return parsePing(ver, p)
+	case TypePong:
+		return parsePong(ver, p)
+	case TypeCallMeMaybe:
+		return CallMeMaybe{}, nil
+	default:
+		return nil, fmt.Errorf("unknown message type 0x%02x", byte(t))
+	}
+}
+
+// Message a discovery message.
+type Message interface {
+	// AppendMarshal appends the message's marshaled representation.
+	AppendMarshal([]byte) []byte
+}
+
+// appendMsgHeader appends two bytes (for t and ver) and then also
+// dataLen bytes to b, returning the appended slice in all. The
+// returned data slice is a subslice of all with just dataLen bytes of
+// where the caller will fill in the data.
+func appendMsgHeader(b []byte, t MessageType, ver uint8, dataLen int) (all, data []byte) {
+	// TODO: optimize this?
+	all = append(b, make([]byte, dataLen+2)...)
+	all[len(b)] = byte(t)
+	all[len(b)+1] = ver
+	data = all[len(b)+2:]
+	return
+}
+
+type Ping struct {
+	TxID [12]byte
+}
+
+func (m *Ping) AppendMarshal(b []byte) []byte {
+	ret, d := appendMsgHeader(b, TypePing, v0, 12)
+	copy(d, m.TxID[:])
+	return ret
+}
+
+func parsePing(ver uint8, p []byte) (m *Ping, err error) {
+	if len(p) < 12 {
+		return nil, errShort
+	}
+	m = new(Ping)
+	copy(m.TxID[:], p)
+	return m, nil
+}
+
+// CallMeMaybe is a message sent only over DERP to request that the recipient try
+// to open up a magicsock path back to the sender.
+//
+// The sender should've already sent UDP packets to the peer to open
+// up the stateful firewall mappings inbound.
+//
+// The recipient may choose to not open a path back, if it's already
+// happy with its path. But usually it will.
+type CallMeMaybe struct{}
+
+func (CallMeMaybe) AppendMarshal(b []byte) []byte {
+	ret, _ := appendMsgHeader(b, TypeCallMeMaybe, v0, 0)
+	return ret
+}
+
+// Pong is a response a Ping.
+//
+// It includes the sender's source IP + port, so it's effectively a
+// STUN response.
+type Pong struct {
+	TxID [12]byte
+	Src  netaddr.IPPort // 18 bytes (16+2) on the wire; v4-mapped ipv6 for IPv4
+}
+
+const pongLen = 12 + 16 + 2
+
+func (m *Pong) AppendMarshal(b []byte) []byte {
+	ret, d := appendMsgHeader(b, TypePong, v0, pongLen)
+	d = d[copy(d, m.TxID[:]):]
+	ip16 := m.Src.IP.As16()
+	d = d[copy(d, ip16[:]):]
+	binary.BigEndian.PutUint16(d, m.Src.Port)
+	return ret
+}
+
+func parsePong(ver uint8, p []byte) (m *Pong, err error) {
+	if len(p) < pongLen {
+		return nil, errShort
+	}
+	m = new(Pong)
+	copy(m.TxID[:], p)
+	p = p[12:]
+
+	m.Src.IP, _ = netaddr.FromStdIP(net.IP(p[:16]))
+	p = p[16:]
+
+	m.Src.Port = binary.BigEndian.Uint16(p)
+	return m, nil
+}
+
+// MessageSummary returns a short summary of m for logging purposes.
+func MessageSummary(m Message) string {
+	switch m := m.(type) {
+	case *Ping:
+		return fmt.Sprintf("ping tx=%x", m.TxID[:6])
+	case *Pong:
+		return fmt.Sprintf("pong tx=%x", m.TxID[:6])
+	case CallMeMaybe:
+		return "call-me-maybe"
+	default:
+		return fmt.Sprintf("%#v", m)
+	}
+}

disco/disco_test.go

@@ -0,0 +1,82 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package disco
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+
+	"inet.af/netaddr"
+)
+
+func TestMarshalAndParse(t *testing.T) {
+	tests := []struct {
+		name string
+		want string
+		m    Message
+	}{
+		{
+			name: "ping",
+			m: &Ping{
+				TxID: [12]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12},
+			},
+			want: "01 00 01 02 03 04 05 06 07 08 09 0a 0b 0c",
+		},
+		{
+			name: "pong",
+			m: &Pong{
+				TxID: [12]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12},
+				Src:  mustIPPort("2.3.4.5:1234"),
+			},
+			want: "02 00 01 02 03 04 05 06 07 08 09 0a 0b 0c 00 00 00 00 00 00 00 00 00 00 ff ff 02 03 04 05 04 d2",
+		},
+		{
+			name: "pongv6",
+			m: &Pong{
+				TxID: [12]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12},
+				Src:  mustIPPort("[fed0::12]:6666"),
+			},
+			want: "02 00 01 02 03 04 05 06 07 08 09 0a 0b 0c fe d0 00 00 00 00 00 00 00 00 00 00 00 00 00 12 1a 0a",
+		},
+		{
+			name: "call_me_maybe",
+			m:    CallMeMaybe{},
+			want: "03 00",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			foo := []byte("foo")
+			got := string(tt.m.AppendMarshal(foo))
+			if !strings.HasPrefix(got, "foo") {
+				t.Fatalf("didn't start with foo: got %q", got)
+			}
+			got = strings.TrimPrefix(got, "foo")
+
+			gotHex := fmt.Sprintf("% x", got)
+			if gotHex != tt.want {
+				t.Fatalf("wrong marshal\n got: %s\nwant: %s\n", gotHex, tt.want)
+			}
+
+			back, err := Parse([]byte(got))
+			if err != nil {
+				t.Fatalf("parse back: %v", err)
+			}
+			if !reflect.DeepEqual(back, tt.m) {
+				t.Errorf("message in %+v doesn't match Parse back result %+v", tt.m, back)
+			}
+		})
+	}
+}
+
+func mustIPPort(s string) netaddr.IPPort {
+	ipp, err := netaddr.ParseIPPort(s)
+	if err != nil {
+		panic(err)
+	}
+	return ipp
+}

go.mod

@@ -1,29 +1,38 @@
 module tailscale.com
 
-go 1.13
+go 1.14
 
 require (
 	github.com/anmitsu/go-shlex v0.0.0-20161002113705-648efa622239 // indirect
 	github.com/apenwarr/fixconsole v0.0.0-20191012055117-5a9f6489cc29
+	github.com/coreos/go-iptables v0.4.5
 	github.com/flynn/go-shlex v0.0.0-20150515145356-3f9db97f8568 // indirect
 	github.com/gliderlabs/ssh v0.2.2
 	github.com/go-ole/go-ole v1.2.4
+	github.com/godbus/dbus/v5 v5.0.3
 	github.com/golang/groupcache v0.0.0-20200121045136-8c9f03a8e57e
 	github.com/google/go-cmp v0.4.0
 	github.com/goreleaser/nfpm v1.1.10
-	github.com/klauspost/compress v1.9.8
+	github.com/jsimonetti/rtnetlink v0.0.0-20200117123717-f846d4f6c1f4
+	github.com/klauspost/compress v1.10.10
 	github.com/kr/pty v1.1.1
 	github.com/mdlayher/netlink v1.1.0
+	github.com/miekg/dns v1.1.30
 	github.com/pborman/getopt v0.0.0-20190409184431-ee0cd42419d3
-	github.com/tailscale/hujson v0.0.0-20190930033718-5098e564d9b3
-	github.com/tailscale/winipcfg-go v0.0.0-20200213045944-185b07f8233f
-	github.com/tailscale/wireguard-go v0.0.0-20200224122332-ad79bbddc844
-	golang.org/x/crypto v0.0.0-20200210222208-86ce3cb69678
+	github.com/peterbourgon/ff/v2 v2.0.0
+	github.com/tailscale/winipcfg-go v0.0.0-20200413171540-609dcf2df55f
+	github.com/tailscale/wireguard-go v0.0.0-20200724155040-d554a2a5e7e1
+	github.com/tcnksm/go-httpstat v0.2.0
+	github.com/toqueteos/webbrowser v1.2.0
+	go4.org/mem v0.0.0-20200706164138-185c595c3ecc
+	golang.org/x/crypto v0.0.0-20200317142112-1b76d66859c6
+	golang.org/x/net v0.0.0-20200301022130-244492dfa37a
 	golang.org/x/oauth2 v0.0.0-20200107190931-bf48bf16ab8d
-	golang.org/x/sys v0.0.0-20200217220822-9197077df867
+	golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e
+	golang.org/x/sys v0.0.0-20200501052902-10377860bb8e
 	golang.org/x/time v0.0.0-20191024005414-555d28b269f0
-	gortc.io/stun v1.22.1
-	honnef.co/go/tools v0.0.1-2019.2.3 // indirect
+	golang.org/x/tools v0.0.0-20191216052735-49a3e744a425
+	honnef.co/go/tools v0.0.1-2020.1.4
+	inet.af/netaddr v0.0.0-20200718043157-99321d6ad24c
 	rsc.io/goversion v1.2.0
 )
-

go.sum

@@ -19,6 +19,8 @@ github.com/blakesmith/ar v0.0.0-20190502131153-809d4375e1fb h1:m935MPodAbYS46DG4
 github.com/blakesmith/ar v0.0.0-20190502131153-809d4375e1fb/go.mod h1:PkYb9DJNAwrSvRx5DYA+gUcOIgTGVMNkfSCbZM8cWpI=
 github.com/cavaliercoder/go-cpio v0.0.0-20180626203310-925f9528c45e h1:hHg27A0RSSp2Om9lubZpiMgVbvn39bsUmW9U5h0twqc=
 github.com/cavaliercoder/go-cpio v0.0.0-20180626203310-925f9528c45e/go.mod h1:oDpT4efm8tSYHXV5tHSdRvBet/b/QzxZ+XyyPehvm3A=
+github.com/coreos/go-iptables v0.4.5 h1:DpHb9vJrZQEFMcVLFKAAGMUVX0XoRC0ptCthinRYm38=
+github.com/coreos/go-iptables v0.4.5/go.mod h1:/mVI274lEDI2ns62jHCDnCyBF9Iwsmekav8Dbxlm1MU=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -28,6 +30,8 @@ github.com/gliderlabs/ssh v0.2.2 h1:6zsha5zo/TWhRhwqCD3+EarCAgZ2yN28ipRnGPnwkI0=
 github.com/gliderlabs/ssh v0.2.2/go.mod h1:U7qILu1NlMHj9FlMhZLlkCdDnU1DBEAqr0aevW3Awn0=
 github.com/go-ole/go-ole v1.2.4 h1:nNBDSCOigTSiarFpYE9J/KtEA1IOW4CNeqT9TQDqCxI=
 github.com/go-ole/go-ole v1.2.4/go.mod h1:XCwSNxSkXRo4vlyPy93sltvi/qJq0jqQhjqQNIwKuxM=
+github.com/godbus/dbus/v5 v5.0.3 h1:ZqHaoEF7TBzh4jzPmqVhE/5A1z9of6orkAe5uHoAeME=
+github.com/godbus/dbus/v5 v5.0.3/go.mod h1:xhWf0FNVPg57R7Z0UbKHbJfkEywrmjJnf7w5xrFpKfA=
 github.com/golang/groupcache v0.0.0-20200121045136-8c9f03a8e57e h1:1r7pUrabqp18hOBcwBwiTsbnFeTZHV9eER/QT5JVZxY=
 github.com/golang/groupcache v0.0.0-20200121045136-8c9f03a8e57e/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc=
 github.com/golang/protobuf v1.2.0 h1:P3YflyNX/ehuJFLhxviNdFxQPkGK5cDcApsge1SqnvM=
@@ -47,11 +51,13 @@ github.com/jsimonetti/rtnetlink v0.0.0-20190606172950-9527aa82566a/go.mod h1:Oz+
 github.com/jsimonetti/rtnetlink v0.0.0-20200117123717-f846d4f6c1f4 h1:nwOc1YaOrYJ37sEBrtWZrdqzK22hiJs3GpDmP3sR2Yw=
 github.com/jsimonetti/rtnetlink v0.0.0-20200117123717-f846d4f6c1f4/go.mod h1:WGuG/smIU4J/54PblvSbh+xvCZmpJnFgr3ds6Z55XMQ=
 github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck=
-github.com/klauspost/compress v1.9.8 h1:VMAMUUOh+gaxKTMk+zqbjsSjsIcUcL/LF4o63i82QyA=
-github.com/klauspost/compress v1.9.8/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A=
+github.com/klauspost/compress v1.10.10 h1:a/y8CglcM7gLGYmlbP/stPE5sR3hbhFRUjCBfd/0B3I=
+github.com/klauspost/compress v1.10.10/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs=
+github.com/kr/pretty v0.1.0 h1:L/CwN0zerZDmRFUapSPitk6f+Q3+0za1rQkzVuMiMFI=
 github.com/kr/pretty v0.1.0/go.mod h1:dAy3ld7l9f0ibDNOQOHHMYYIIbhfbHSm3C4ZsoJORNo=
 github.com/kr/pty v1.1.1 h1:VkoXIwSboBpnk99O/KFauAEILuNHv5DVFKZMBN/gUgw=
 github.com/kr/pty v1.1.1/go.mod h1:pFQYn66WHrOpPYNljwOMqo10TkYh1fy3cYio2l3bCsQ=
+github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
 github.com/mattn/go-zglob v0.0.1 h1:xsEx/XUoVlI6yXjqBK062zYhRTZltCNmYPx6v+8DNaY=
 github.com/mattn/go-zglob v0.0.1/go.mod h1:9fxibJccNxU2cnpIKLRRFA7zX7qhkJIQWBb449FYHOo=
@@ -59,10 +65,15 @@ github.com/mdlayher/netlink v0.0.0-20190409211403-11939a169225/go.mod h1:eQB3mZE
 github.com/mdlayher/netlink v1.0.0/go.mod h1:KxeJAFOFLG6AjpyDkQ/iIhxygIUKD+vcwqcnu43w/+M=
 github.com/mdlayher/netlink v1.1.0 h1:mpdLgm+brq10nI9zM1BpX1kpDbh3NLl3RSnVq6ZSkfg=
 github.com/mdlayher/netlink v1.1.0/go.mod h1:H4WCitaheIsdF9yOYu8CFmCgQthAPIWZmcKp9uZHgmY=
+github.com/miekg/dns v1.1.30 h1:Qww6FseFn8PRfw07jueqIXqodm0JKiiKuK0DeXSqfyo=
+github.com/miekg/dns v1.1.30/go.mod h1:KNUDUusw/aVsxyTYZM1oqvCicbwhgbNgztCETuNZ7xM=
 github.com/op/go-logging v0.0.0-20160315200505-970db520ece7 h1:lDH9UUVJtmYCjyT0CI4q8xvlXPxeZ0gYCVvWbmPlp88=
 github.com/op/go-logging v0.0.0-20160315200505-970db520ece7/go.mod h1:HzydrMdWErDVzsI23lYNej1Htcns9BCg93Dk0bBINWk=
 github.com/pborman/getopt v0.0.0-20190409184431-ee0cd42419d3 h1:YtFkrqsMEj7YqpIhRteVxJxCeC3jJBieuLr0d4C4rSA=
 github.com/pborman/getopt v0.0.0-20190409184431-ee0cd42419d3/go.mod h1:85jBQOZwpVEaDAr341tbn15RS4fCAsIst0qp7i8ex1o=
+github.com/pelletier/go-toml v1.6.0/go.mod h1:5N711Q9dKgbdkxHL+MEfF31hpT7l0S0s/t2kKREewys=
+github.com/peterbourgon/ff/v2 v2.0.0 h1:lx0oYI5qr/FU1xnpNhQ+EZM04gKgn46jyYvGEEqBBbY=
+github.com/peterbourgon/ff/v2 v2.0.0/go.mod h1:xjwr+t+SjWm4L46fcj/D+Ap+6ME7+HqFzaP22pP5Ggk=
 github.com/pkg/errors v0.8.1 h1:iURUrRGxPUNPdy5/HRSm+Yj6okJ6UtLINN0Q9M4+h3I=
 github.com/pkg/errors v0.8.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
@@ -73,54 +84,46 @@ github.com/sassoftware/go-rpmutils v0.0.0-20190420191620-a8f1baeba37b/go.mod h1:
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.4.0 h1:2E4SXV/wtOkTonXsotYi4li6zVWxYlZuYNCXe9XRJyk=
 github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
-github.com/tailscale/hujson v0.0.0-20190930033718-5098e564d9b3 h1:rdtXEo9yffOjh4vZQJw3heaY+ggXKp+zvMX5fihh6lI=
-github.com/tailscale/hujson v0.0.0-20190930033718-5098e564d9b3/go.mod h1:STqf+YV0ADdzk4ejtXFsGqDpATP9JoL0OB+hiFQbkdE=
-github.com/tailscale/winipcfg-go v0.0.0-20200213045944-185b07f8233f h1:q2ynfOHxHaaMnkZ1YHswWeO6wEk7IyOnkFozytZ1ztc=
-github.com/tailscale/winipcfg-go v0.0.0-20200213045944-185b07f8233f/go.mod h1:x880GWw5fvrl2DVTQ04ttXQD4DuppTt1Yz6wLibbjNE=
-github.com/tailscale/wireguard-go v0.0.0-20191108062213-b93cdd0582db h1:oP0crfwOb3WZSVrMVm/o51NXN2JirDlcdlNEIPTmgI0=
-github.com/tailscale/wireguard-go v0.0.0-20200207221558-a158079b156a h1:5TWA3nl2QUfL9OiE3tlBpqJd4GYd4hbGtDNkWQQ2fyc=
-github.com/tailscale/wireguard-go v0.0.0-20200207221558-a158079b156a/go.mod h1:QPS8HjBzzAXoQNndUNx2efJaQbCCz8nI2Cv1ksTUHyY=
-github.com/tailscale/wireguard-go v0.0.0-20200208161837-3cd0a483944a h1:vIyObUBvnXB1XTKTBM4AgoUFR9RHiz/kslGHClkXQVg=
-github.com/tailscale/wireguard-go v0.0.0-20200208161837-3cd0a483944a/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
-github.com/tailscale/wireguard-go v0.0.0-20200208214841-2981baf46731 h1:sNmny/5pHqHdm081Fx8rcNFnwt0zTGuee/0+Jz+tXCA=
-github.com/tailscale/wireguard-go v0.0.0-20200208214841-2981baf46731/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
-github.com/tailscale/wireguard-go v0.0.0-20200211020303-f39bc8eeee1b h1:99LOgoPy0PcmT+yg1HiPJXk/aV/KWeGGh4WtSVB7noM=
-github.com/tailscale/wireguard-go v0.0.0-20200211020303-f39bc8eeee1b/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
-github.com/tailscale/wireguard-go v0.0.0-20200213180345-a7c4b7719b1d h1:LVJovgZxbmPxtY6kJm4vwMtk0HpcNeI+vU2jB3T8M40=
-github.com/tailscale/wireguard-go v0.0.0-20200213180345-a7c4b7719b1d/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
-github.com/tailscale/wireguard-go v0.0.0-20200219153015-ff5ce9e55c86 h1:yVWApjTol2CR3dVbiiSOI78KnElu3AlHFvntYM2egBU=
-github.com/tailscale/wireguard-go v0.0.0-20200219153015-ff5ce9e55c86/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
-github.com/tailscale/wireguard-go v0.0.0-20200224122332-ad79bbddc844 h1:CChfZok8JbY2dT+BNLHAcCLcGuCTBEhrkYtUwOk5rVY=
-github.com/tailscale/wireguard-go v0.0.0-20200224122332-ad79bbddc844/go.mod h1:JPm5cTfu1K+qDFRbiHy0sOlHUylYQbpl356sdYFD8V4=
+github.com/tailscale/winipcfg-go v0.0.0-20200413171540-609dcf2df55f h1:uFj5bslHsMzxIM8UTjAhq4VXeo6GfNW91rpoh/WMJaY=
+github.com/tailscale/winipcfg-go v0.0.0-20200413171540-609dcf2df55f/go.mod h1:x880GWw5fvrl2DVTQ04ttXQD4DuppTt1Yz6wLibbjNE=
+github.com/tcnksm/go-httpstat v0.2.0 h1:rP7T5e5U2HfmOBmZzGgGZjBQ5/GluWUylujl0tJ04I0=
+github.com/tcnksm/go-httpstat v0.2.0/go.mod h1:s3JVJFtQxtBEBC9dwcdTTXS9xFnM3SXAZwPG41aurT8=
+github.com/toqueteos/webbrowser v1.2.0 h1:tVP/gpK69Fx+qMJKsLE7TD8LuGWPnEV71wBN9rrstGQ=
+github.com/toqueteos/webbrowser v1.2.0/go.mod h1:XWoZq4cyp9WeUeak7w7LXRUQf1F1ATJMir8RTqb4ayM=
 github.com/ulikunitz/xz v0.5.6 h1:jGHAfXawEGZQ3blwU5wnWKQJvAraT7Ftq9EXjnXYgt8=
 github.com/ulikunitz/xz v0.5.6/go.mod h1:2bypXElzHzzJZwzH67Y6wb67pO62Rzfn7BSiF4ABRW8=
 github.com/xi2/xz v0.0.0-20171230120015-48954b6210f8 h1:nIPpBwaJSVYIxUFsDv3M8ofmx9yWTog9BfvIu0q41lo=
 github.com/xi2/xz v0.0.0-20171230120015-48954b6210f8/go.mod h1:HUYIGzjTL3rfEspMxjDjgmT5uz5wzYJKVo23qUhYTos=
+go4.org/mem v0.0.0-20200706164138-185c595c3ecc h1:paujszgN6SpsO/UsXC7xax3gQAKz/XQKCYZLQdU34Tw=
+go4.org/mem v0.0.0-20200706164138-185c595c3ecc/go.mod h1:NEYvpHWemiG/E5UWfaN5QAIGZeT1sa0Z2UNk6oeMb/k=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20190510104115-cbcb75029529/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/crypto v0.0.0-20191002192127-34f69633bfdc/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
-golang.org/x/crypto v0.0.0-20200206161412-a0c6ece9d31a h1:aczoJ0HPNE92XKa7DrIzkNN6esOKO2TBwiiYoKcINhA=
-golang.org/x/crypto v0.0.0-20200206161412-a0c6ece9d31a/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/crypto v0.0.0-20200208060501-ecb85df21340 h1:KOcEaR10tFr7gdJV2GCKw8Os5yED1u1aOqHjOAb6d2Y=
-golang.org/x/crypto v0.0.0-20200208060501-ecb85df21340/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
-golang.org/x/crypto v0.0.0-20200210222208-86ce3cb69678 h1:wCWoJcFExDgyYx2m2hpHgwz8W3+FPdfldvIgzqDIhyg=
-golang.org/x/crypto v0.0.0-20200210222208-86ce3cb69678/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
+golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
+golang.org/x/crypto v0.0.0-20200317142112-1b76d66859c6 h1:TjszyFsQsyZNHwdVdZ5m7bjmreu0znc2kRYsEml9/Ww=
+golang.org/x/crypto v0.0.0-20200317142112-1b76d66859c6/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
 golang.org/x/mod v0.0.0-20190513183733-4bf6d317e70e/go.mod h1:mXi4GBBbnImb6dmsKGUJ2LatrhH/nqhxcFungHvyanc=
+golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190311183353-d8887717615a/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20190827160401-ba9fcec4b297/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20190923162816-aa69164e4478/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20191003171128-d98b1b443823/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20191007182048-72f939374954/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/net v0.0.0-20200202094626-16171245cfb2 h1:CCH4IOTTfewWjGOlSp+zGcjutRKlBEZQ6wTn8ozI/nI=
 golang.org/x/net v0.0.0-20200202094626-16171245cfb2/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20200301022130-244492dfa37a h1:GuSPYbZzB5/dcLNCwLQLsg3obCJtX9IJhpXkvY7kzk0=
+golang.org/x/net v0.0.0-20200301022130-244492dfa37a/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
 golang.org/x/oauth2 v0.0.0-20200107190931-bf48bf16ab8d h1:TzXSXBo42m9gQenoE3b9BGiEpg5IG2JkU5FkPIawgtw=
 golang.org/x/oauth2 v0.0.0-20200107190931-bf48bf16ab8d/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
 golang.org/x/sync v0.0.0-20181221193216-37e7f081c4d4 h1:YUO/7uOKsKeq9UokNS62b8FYywz3ker1l1vDZRCRefw=
 golang.org/x/sync v0.0.0-20181221193216-37e7f081c4d4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
-golang.org/x/sync v0.0.0-20190423024810-112230192c58 h1:8gQV6CLnAEikrhgkHFbMAEhagSSnXWGV915qUMm9mrU=
 golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e h1:vcxGaoTs7kV8m5Np9uUNQin4BrLOthgV7252N8V+FwY=
+golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190310054646-10058d7d4faa/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20190312061237-fead79001313/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -128,38 +131,41 @@ golang.org/x/sys v0.0.0-20190405154228-4b34438f7a67/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20190411185658-b44545bcd369/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20190826190057-c7b8b68b1456/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20190924154521-2837fb4f24fe/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20191003212358-c178f38b412c/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20191008105621-543471e840be/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200202164722-d101bd2416d5 h1:LfCXLvNmTYH9kEmVgqbnsWfruoXZIrh4YBgqVHtDvw0=
 golang.org/x/sys v0.0.0-20200202164722-d101bd2416d5/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20200212091648-12a6c2dcc1e4 h1:sfkvUWPNGwSV+8/fNqctR5lS2AqCSqYwXdrjCxp/dXo=
-golang.org/x/sys v0.0.0-20200212091648-12a6c2dcc1e4/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20200217220822-9197077df867 h1:JoRuNIf+rpHl+VhScRQQvzbHed86tKkqwPMV34T8myw=
-golang.org/x/sys v0.0.0-20200217220822-9197077df867/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20200501052902-10377860bb8e h1:hq86ru83GdWTlfQFZGO4nZJTU4Bs2wfHl8oFHRaXsfc=
+golang.org/x/sys v0.0.0-20200501052902-10377860bb8e/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2 h1:tW2bmiBqwgJj/UpqtC8EpXEZVYOwU0yG4iWbprSVAcs=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
 golang.org/x/time v0.0.0-20191024005414-555d28b269f0 h1:/5xXl8Y5W96D+TtHSlonuFqGHIWVuyCkGJLwGh9JJFs=
 golang.org/x/time v0.0.0-20191024005414-555d28b269f0/go.mod h1:tRJNPiyCQ0inRvYxbN9jk5I+vvW/OXSQhTDSoE431IQ=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
-golang.org/x/tools v0.0.0-20190621195816-6e04913cbbac h1:MQEvx39qSf8vyrx3XRaOe+j1UDIzKwkYOVObRgGPVqI=
-golang.org/x/tools v0.0.0-20190621195816-6e04913cbbac/go.mod h1:/rFqwRUd4F7ZHNgwSSTFct+R/Kf4OFW1sUzUTQQTgfc=
+golang.org/x/tools v0.0.0-20191130070609-6e064ea0cf2d h1:/iIZNFGxc/a7C3yWjGcnboV+Tkc7mxr+p6fDztwoxuM=
+golang.org/x/tools v0.0.0-20191130070609-6e064ea0cf2d/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.0.0-20191216052735-49a3e744a425 h1:VvQyQJN0tSuecqgcIxMWnnfG5kSmgy9KZR9sW3W5QeA=
+golang.org/x/tools v0.0.0-20191216052735-49a3e744a425/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543 h1:E7g+9GITq07hpfrRu66IVDexMakfv52eLZ2CXBWiKr4=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-golang.zx2c4.com/wireguard v0.0.20200121 h1:vcswa5Q6f+sylDfjqyrVNNrjsFUUbPsgAQTBCAg/Qf8=
-golang.zx2c4.com/wireguard v0.0.20200121/go.mod h1:P2HsVp8SKwZEufsnezXZA4GRX/T49/HlU7DGuelXsU4=
 google.golang.org/appengine v1.4.0 h1:/wp5JvzpHIxhs/dumFmF7BXTf3Z+dd4uXta4kVyO508=
 google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20180628173108-788fd7840127/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15 h1:YR8cESwS4TdDjEe65xsg0ogRM/Nc3DYOhEAlW+xobZo=
 gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/errgo.v2 v2.1.0/go.mod h1:hNsd1EY+bozCKY1Ytp96fpM3vjJbqLJn88ws8XvfDNI=
 gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
+gopkg.in/yaml.v2 v2.2.4/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 gopkg.in/yaml.v2 v2.2.7 h1:VUgggvou5XRW9mHwD/yXxIYSMtY0zoKQf/v226p2nyo=
 gopkg.in/yaml.v2 v2.2.7/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
-gortc.io/stun v1.22.1 h1:96mOdDATYRqhYB+TZdenWBg4CzL2Ye5kPyBXQ8KAB+8=
-gortc.io/stun v1.22.1/go.mod h1:XD5lpONVyjvV3BgOyJFNo0iv6R2oZB4L+weMqxts+zg=
-honnef.co/go/tools v0.0.1-2019.2.3 h1:3JgtbtFHMiCmsznwGVTUWbgGov+pVqnlf1dEJTNAXeM=
-honnef.co/go/tools v0.0.1-2019.2.3/go.mod h1:a3bituU0lyd329TUQxRnasdCoJDkEUEAqEt0JzvZhAg=
+honnef.co/go/tools v0.0.1-2020.1.4 h1:UoveltGrhghAA7ePc+e+QYDHXrBps2PqFZiHkGR/xK8=
+honnef.co/go/tools v0.0.1-2020.1.4/go.mod h1:X/FiERA/W4tHapMX5mGpAtMSVEeEUOyHaw9vFzvIQ3k=
+inet.af/netaddr v0.0.0-20200718043157-99321d6ad24c h1:si3Owrfem175Ry6gKqnh59eOXxDojyBTIHxUKuvK/Eo=
+inet.af/netaddr v0.0.0-20200718043157-99321d6ad24c/go.mod h1:qqYzz/2whtrbWJvt+DNWQyvekNN4ePQZcg2xc2/Yjww=
 rsc.io/goversion v1.2.0 h1:SPn+NLTiAG7w30IRK/DKp1BjvpWabYgxlLp/+kx5J8w=
 rsc.io/goversion v1.2.0/go.mod h1:Eih9y/uIBS3ulggl7KNJ09xGSLcuNaLgmvvqa07sgfo=

interfaces/interfaces.go

@@ -1,58 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package interfaces contains helpers for looking up system network interfaces.
-package interfaces
-
-import (
-	"net"
-	"strings"
-)
-
-// Tailscale returns the current machine's Tailscale interface, if any.
-// If none is found, all zero values are returned.
-// A non-nil error is only returned on a problem listing the system interfaces.
-func Tailscale() (net.IP, *net.Interface, error) {
-	ifs, err := net.Interfaces()
-	if err != nil {
-		return nil, nil, err
-	}
-	for _, iface := range ifs {
-		if !maybeTailscaleInterfaceName(iface.Name) {
-			continue
-		}
-		addrs, err := iface.Addrs()
-		if err != nil {
-			continue
-		}
-		for _, a := range addrs {
-			if ipnet, ok := a.(*net.IPNet); ok && IsTailscaleIP(ipnet.IP) {
-				return ipnet.IP, &iface, nil
-			}
-		}
-	}
-	return nil, nil, nil
-}
-
-// maybeTailscaleInterfaceName reports whether s is an interface
-// name that might be used by Tailscale.
-func maybeTailscaleInterfaceName(s string) bool {
-	return strings.HasPrefix(s, "wg") ||
-		strings.HasPrefix(s, "ts") ||
-		strings.HasPrefix(s, "tailscale")
-}
-
-// IsTailscaleIP reports whether ip is an IP in a range used by
-// Tailscale virtual network interfaces.
-func IsTailscaleIP(ip net.IP) bool {
-	return cgNAT.Contains(ip)
-}
-
-var cgNAT = func() *net.IPNet {
-	_, ipNet, err := net.ParseCIDR("100.64.0.0/10")
-	if err != nil {
-		panic(err)
-	}
-	return ipNet
-}()

internal/deepprint/deepprint.go

@@ -0,0 +1,93 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package deepprint walks a Go value recursively, in a predictable
+// order, without looping, and prints each value out to a given
+// Writer, which is assumed to be a hash.Hash, as this package doesn't
+// format things nicely.
+//
+// This is intended as a lighter version of go-spew, etc. We don't need its
+// features when our writer is just a hash.
+package deepprint
+
+import (
+	"crypto/sha256"
+	"fmt"
+	"io"
+	"reflect"
+)
+
+func Hash(v interface{}) string {
+	h := sha256.New()
+	Print(h, v)
+	return fmt.Sprintf("%x", h.Sum(nil))
+}
+
+func Print(w io.Writer, v interface{}) {
+	print(w, reflect.ValueOf(v), make(map[uintptr]bool))
+}
+
+func print(w io.Writer, v reflect.Value, visited map[uintptr]bool) {
+	if !v.IsValid() {
+		return
+	}
+	switch v.Kind() {
+	default:
+		panic(fmt.Sprintf("unhandled kind %v for type %v", v.Kind(), v.Type()))
+	case reflect.Ptr:
+		ptr := v.Pointer()
+		if visited[ptr] {
+			return
+		}
+		visited[ptr] = true
+		print(w, v.Elem(), visited)
+		return
+	case reflect.Struct:
+		fmt.Fprintf(w, "struct{\n")
+		t := v.Type()
+		for i, n := 0, v.NumField(); i < n; i++ {
+			sf := t.Field(i)
+			fmt.Fprintf(w, "%s: ", sf.Name)
+			print(w, v.Field(i), visited)
+			fmt.Fprintf(w, "\n")
+		}
+	case reflect.Slice, reflect.Array:
+		if v.Type().Elem().Kind() == reflect.Uint8 && v.CanInterface() {
+			fmt.Fprintf(w, "%q", v.Interface())
+			return
+		}
+		fmt.Fprintf(w, "[%d]{\n", v.Len())
+		for i, ln := 0, v.Len(); i < ln; i++ {
+			fmt.Fprintf(w, " [%d]: ", i)
+			print(w, v.Index(i), visited)
+			fmt.Fprintf(w, "\n")
+		}
+		fmt.Fprintf(w, "}\n")
+	case reflect.Interface:
+		print(w, v.Elem(), visited)
+	case reflect.Map:
+		sm := newSortedMap(v)
+		fmt.Fprintf(w, "map[%d]{\n", len(sm.Key))
+		for i, k := range sm.Key {
+			print(w, k, visited)
+			fmt.Fprintf(w, ": ")
+			print(w, sm.Value[i], visited)
+			fmt.Fprintf(w, "\n")
+		}
+		fmt.Fprintf(w, "}\n")
+
+	case reflect.String:
+		fmt.Fprintf(w, "%s", v.String())
+	case reflect.Bool:
+		fmt.Fprintf(w, "%v", v.Bool())
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		fmt.Fprintf(w, "%v", v.Int())
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		fmt.Fprintf(w, "%v", v.Uint())
+	case reflect.Float32, reflect.Float64:
+		fmt.Fprintf(w, "%v", v.Float())
+	case reflect.Complex64, reflect.Complex128:
+		fmt.Fprintf(w, "%v", v.Complex())
+	}
+}

internal/deepprint/deepprint_test.go

@@ -0,0 +1,70 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package deepprint
+
+import (
+	"bytes"
+	"testing"
+
+	"github.com/tailscale/wireguard-go/wgcfg"
+	"inet.af/netaddr"
+	"tailscale.com/wgengine/router"
+)
+
+func TestDeepPrint(t *testing.T) {
+	// v contains the types of values we care about for our current callers.
+	// Mostly we're just testing that we don't panic on handled types.
+	v := getVal()
+
+	var buf bytes.Buffer
+	Print(&buf, v)
+	t.Logf("Got: %s", buf.Bytes())
+
+	hash1 := Hash(v)
+	t.Logf("hash: %v", hash1)
+	for i := 0; i < 20; i++ {
+		hash2 := Hash(getVal())
+		if hash1 != hash2 {
+			t.Error("second hash didn't match")
+		}
+	}
+}
+
+func getVal() []interface{} {
+	return []interface{}{
+		&wgcfg.Config{
+			Name:       "foo",
+			Addresses:  []wgcfg.CIDR{{Mask: 5, IP: wgcfg.IP{Addr: [16]byte{3: 3}}}},
+			ListenPort: 5,
+			Peers: []wgcfg.Peer{
+				{
+					Endpoints: []wgcfg.Endpoint{
+						{
+							Host: "foo",
+							Port: 5,
+						},
+					},
+				},
+			},
+		},
+		&router.Config{
+			DNSConfig: router.DNSConfig{
+				Nameservers: []netaddr.IP{netaddr.IPv4(8, 8, 8, 8)},
+				Domains:     []string{"tailscale.net"},
+			},
+		},
+		map[string]string{
+			"key1": "val1",
+			"key2": "val2",
+			"key3": "val3",
+			"key4": "val4",
+			"key5": "val5",
+			"key6": "val6",
+			"key7": "val7",
+			"key8": "val8",
+			"key9": "val9",
+		},
+	}
+}

internal/deepprint/fmtsort.go

@@ -0,0 +1,224 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// and
+
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This is a slightly modified fork of Go's src/internal/fmtsort/sort.go
+
+package deepprint
+
+import (
+	"reflect"
+	"sort"
+)
+
+// Note: Throughout this package we avoid calling reflect.Value.Interface as
+// it is not always legal to do so and it's easier to avoid the issue than to face it.
+
+// sortedMap represents a map's keys and values. The keys and values are
+// aligned in index order: Value[i] is the value in the map corresponding to Key[i].
+type sortedMap struct {
+	Key   []reflect.Value
+	Value []reflect.Value
+}
+
+func (o *sortedMap) Len() int           { return len(o.Key) }
+func (o *sortedMap) Less(i, j int) bool { return compare(o.Key[i], o.Key[j]) < 0 }
+func (o *sortedMap) Swap(i, j int) {
+	o.Key[i], o.Key[j] = o.Key[j], o.Key[i]
+	o.Value[i], o.Value[j] = o.Value[j], o.Value[i]
+}
+
+// Sort accepts a map and returns a sortedMap that has the same keys and
+// values but in a stable sorted order according to the keys, modulo issues
+// raised by unorderable key values such as NaNs.
+//
+// The ordering rules are more general than with Go's < operator:
+//
+//  - when applicable, nil compares low
+//  - ints, floats, and strings order by <
+//  - NaN compares less than non-NaN floats
+//  - bool compares false before true
+//  - complex compares real, then imag
+//  - pointers compare by machine address
+//  - channel values compare by machine address
+//  - structs compare each field in turn
+//  - arrays compare each element in turn.
+//    Otherwise identical arrays compare by length.
+//  - interface values compare first by reflect.Type describing the concrete type
+//    and then by concrete value as described in the previous rules.
+//
+func newSortedMap(mapValue reflect.Value) *sortedMap {
+	if mapValue.Type().Kind() != reflect.Map {
+		return nil
+	}
+	// Note: this code is arranged to not panic even in the presence
+	// of a concurrent map update. The runtime is responsible for
+	// yelling loudly if that happens. See issue 33275.
+	n := mapValue.Len()
+	key := make([]reflect.Value, 0, n)
+	value := make([]reflect.Value, 0, n)
+	iter := mapValue.MapRange()
+	for iter.Next() {
+		key = append(key, iter.Key())
+		value = append(value, iter.Value())
+	}
+	sorted := &sortedMap{
+		Key:   key,
+		Value: value,
+	}
+	sort.Stable(sorted)
+	return sorted
+}
+
+// compare compares two values of the same type. It returns -1, 0, 1
+// according to whether a > b (1), a == b (0), or a < b (-1).
+// If the types differ, it returns -1.
+// See the comment on Sort for the comparison rules.
+func compare(aVal, bVal reflect.Value) int {
+	aType, bType := aVal.Type(), bVal.Type()
+	if aType != bType {
+		return -1 // No good answer possible, but don't return 0: they're not equal.
+	}
+	switch aVal.Kind() {
+	case reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64:
+		a, b := aVal.Int(), bVal.Int()
+		switch {
+		case a < b:
+			return -1
+		case a > b:
+			return 1
+		default:
+			return 0
+		}
+	case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
+		a, b := aVal.Uint(), bVal.Uint()
+		switch {
+		case a < b:
+			return -1
+		case a > b:
+			return 1
+		default:
+			return 0
+		}
+	case reflect.String:
+		a, b := aVal.String(), bVal.String()
+		switch {
+		case a < b:
+			return -1
+		case a > b:
+			return 1
+		default:
+			return 0
+		}
+	case reflect.Float32, reflect.Float64:
+		return floatCompare(aVal.Float(), bVal.Float())
+	case reflect.Complex64, reflect.Complex128:
+		a, b := aVal.Complex(), bVal.Complex()
+		if c := floatCompare(real(a), real(b)); c != 0 {
+			return c
+		}
+		return floatCompare(imag(a), imag(b))
+	case reflect.Bool:
+		a, b := aVal.Bool(), bVal.Bool()
+		switch {
+		case a == b:
+			return 0
+		case a:
+			return 1
+		default:
+			return -1
+		}
+	case reflect.Ptr:
+		a, b := aVal.Pointer(), bVal.Pointer()
+		switch {
+		case a < b:
+			return -1
+		case a > b:
+			return 1
+		default:
+			return 0
+		}
+	case reflect.Chan:
+		if c, ok := nilCompare(aVal, bVal); ok {
+			return c
+		}
+		ap, bp := aVal.Pointer(), bVal.Pointer()
+		switch {
+		case ap < bp:
+			return -1
+		case ap > bp:
+			return 1
+		default:
+			return 0
+		}
+	case reflect.Struct:
+		for i := 0; i < aVal.NumField(); i++ {
+			if c := compare(aVal.Field(i), bVal.Field(i)); c != 0 {
+				return c
+			}
+		}
+		return 0
+	case reflect.Array:
+		for i := 0; i < aVal.Len(); i++ {
+			if c := compare(aVal.Index(i), bVal.Index(i)); c != 0 {
+				return c
+			}
+		}
+		return 0
+	case reflect.Interface:
+		if c, ok := nilCompare(aVal, bVal); ok {
+			return c
+		}
+		c := compare(reflect.ValueOf(aVal.Elem().Type()), reflect.ValueOf(bVal.Elem().Type()))
+		if c != 0 {
+			return c
+		}
+		return compare(aVal.Elem(), bVal.Elem())
+	default:
+		// Certain types cannot appear as keys (maps, funcs, slices), but be explicit.
+		panic("bad type in compare: " + aType.String())
+	}
+}
+
+// nilCompare checks whether either value is nil. If not, the boolean is false.
+// If either value is nil, the boolean is true and the integer is the comparison
+// value. The comparison is defined to be 0 if both are nil, otherwise the one
+// nil value compares low. Both arguments must represent a chan, func,
+// interface, map, pointer, or slice.
+func nilCompare(aVal, bVal reflect.Value) (int, bool) {
+	if aVal.IsNil() {
+		if bVal.IsNil() {
+			return 0, true
+		}
+		return -1, true
+	}
+	if bVal.IsNil() {
+		return 1, true
+	}
+	return 0, false
+}
+
+// floatCompare compares two floating-point values. NaNs compare low.
+func floatCompare(a, b float64) int {
+	switch {
+	case isNaN(a):
+		return -1 // No good answer if b is a NaN so don't bother checking.
+	case isNaN(b):
+		return 1
+	case a < b:
+		return -1
+	case a > b:
+		return 1
+	}
+	return 0
+}
+
+func isNaN(a float64) bool {
+	return a != a
+}

ipn/backend.go

@@ -5,11 +5,15 @@
 package ipn
 
 import (
+	"net/http"
 	"time"
 
+	"golang.org/x/oauth2"
 	"tailscale.com/control/controlclient"
+	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/empty"
+	"tailscale.com/types/structs"
 	"tailscale.com/wgengine"
 )
 
@@ -24,31 +28,48 @@ const (
 	Running
 )
 
+// GoogleIDToken Type is the oauth2.Token.TokenType for the Google
+// ID tokens used by the Android client.
+const GoogleIDTokenType = "ts_android_google_login"
+
 func (s State) String() string {
 	return [...]string{"NoState", "NeedsLogin", "NeedsMachineAuth",
 		"Stopped", "Starting", "Running"}[s]
 }
 
+// EngineStatus contains WireGuard engine stats.
 type EngineStatus struct {
 	RBytes, WBytes wgengine.ByteCount
 	NumLive        int
+	LiveDERPs      int // number of active DERP connections
 	LivePeers      map[tailcfg.NodeKey]wgengine.PeerStatus
 }
 
-type NetworkMap = controlclient.NetworkMap
-
+// Notify is a communication from a backend (e.g. tailscaled) to a frontend
+// (cmd/tailscale, iOS, macOS, Win Tasktray).
 // In any given notification, any or all of these may be nil, meaning
 // that they have not changed.
+// They are JSON-encoded on the wire, despite the lack of struct tags.
 type Notify struct {
-	Version       string         // version number of IPN backend
-	ErrMessage    *string        // critical error message, if any
-	LoginFinished *empty.Message // event: non-nil when login process succeeded
-	State         *State         // current IPN state has changed
-	Prefs         *Prefs         // preferences were changed
-	NetMap        *NetworkMap    // new netmap received
-	Engine        *EngineStatus  // wireguard engine stats
-	BrowseToURL   *string        // UI should open a browser right now
-	BackendLogID  *string        // public logtail id used by backend
+	_             structs.Incomparable
+	Version       string                    // version number of IPN backend
+	ErrMessage    *string                   // critical error message, if any
+	LoginFinished *empty.Message            // event: non-nil when login process succeeded
+	State         *State                    // current IPN state has changed
+	Prefs         *Prefs                    // preferences were changed
+	NetMap        *controlclient.NetworkMap // new netmap received
+	Engine        *EngineStatus             // wireguard engine stats
+	Status        *ipnstate.Status          // full status
+	BrowseToURL   *string                   // UI should open a browser right now
+	BackendLogID  *string                   // public logtail id used by backend
+
+	// LocalTCPPort, if non-nil, informs the UI frontend which
+	// (non-zero) localhost TCP port it's listening on.
+	// This is currently only used by Tailscale when run in the
+	// macOS Network Extension.
+	LocalTCPPort *uint16 `json:",omitempty"`
+
+	// type is mirrored in xcode/Shared/IPN.swift
 }
 
 // StateKey is an opaque identifier for a set of LocalBackend state
@@ -82,6 +103,9 @@ type Options struct {
 	//    an initial overwrite of backend state with Prefs.
 	StateKey StateKey
 	Prefs    *Prefs
+	// AuthKey is an optional node auth key used to authorize a
+	// new node key without user interaction.
+	AuthKey string
 	// LegacyConfigPath optionally specifies the old-style relaynode
 	// relay.conf location. If both LegacyConfigPath and StateKey are
 	// specified and the requested state doesn't exist in the backend
@@ -92,27 +116,41 @@ type Options struct {
 	LegacyConfigPath string
 	// Notify is called when backend events happen.
 	Notify func(Notify) `json:"-"`
+	// HTTPTestClient is an optional HTTP client to pass to controlclient
+	// (for tests only).
+	HTTPTestClient *http.Client
 }
 
+// Backend is the interface between Tailscale frontends
+// (e.g. cmd/tailscale, iOS/MacOS/Windows GUIs) and the tailscale
+// backend (e.g. cmd/tailscaled) running on the same machine.
+// (It has nothing to do with the interface between the backends
+// and the cloud control plane.)
 type Backend interface {
-	// Start or restart the backend, because a new Handle has connected.
-	Start(opts Options) error
+	// Start starts or restarts the backend, typically when a
+	// frontend client connects.
+	Start(Options) error
 	// StartLoginInteractive requests to start a new interactive login
 	// flow. This should trigger a new BrowseToURL notification
 	// eventually.
 	StartLoginInteractive()
-	// Logout terminates the current login session and stop the
+	// Login logs in with an OAuth2 token.
+	Login(token *oauth2.Token)
+	// Logout terminates the current login session and stops the
 	// wireguard engine.
 	Logout()
-	// SetPrefs install a new set of user preferences, including
-	// WantRunning.  This may cause the wireguard engine to
+	// SetPrefs installs a new set of user preferences, including
+	// WantRunning. This may cause the wireguard engine to
 	// reconfigure or stop.
-	SetPrefs(new *Prefs)
+	SetPrefs(*Prefs)
 	// RequestEngineStatus polls for an update from the wireguard
 	// engine. Only needed if you want to display byte
 	// counts. Connection events are emitted automatically without
 	// polling.
 	RequestEngineStatus()
+	// RequestStatus requests that a full Status update
+	// notification is sent.
+	RequestStatus()
 	// FakeExpireAfter pretends that the current key is going to
 	// expire after duration x. This is useful for testing GUIs to
 	// make sure they react properly with keys that are going to

ipn/e2e_test.go

@@ -1,205 +0,0 @@
-// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build depends_on_currently_unreleased
-
-package ipn
-
-import (
-	"bytes"
-	"io/ioutil"
-	"net/http"
-	"net/http/httptest"
-	"testing"
-	"time"
-
-	"github.com/tailscale/wireguard-go/tun/tuntest"
-	"tailscale.com/control/controlclient"
-	"tailscale.com/tailcfg"
-	"tailscale.com/testy"
-	"tailscale.com/wgengine"
-	"tailscale.com/wgengine/magicsock"
-	"tailscale.io/control" // not yet released
-)
-
-func TestIPN(t *testing.T) {
-	testy.FixLogs(t)
-	defer testy.UnfixLogs(t)
-
-	// Turn off STUN for the test to make it hermitic.
-	// TODO(crawshaw): add a test that runs against a local STUN server.
-	origDefaultSTUN := magicsock.DefaultSTUN
-	magicsock.DefaultSTUN = nil
-	defer func() {
-		magicsock.DefaultSTUN = origDefaultSTUN
-	}()
-
-	// TODO(apenwarr): Make resource checks actually pass.
-	// They don't right now, because (at least) wgengine doesn't fully
-	// shut down.
-	//	rc := testy.NewResourceCheck()
-	//	defer rc.Assert(t)
-
-	var ctl *control.Server
-
-	ctlHandler := func(w http.ResponseWriter, r *http.Request) {
-		ctl.ServeHTTP(w, r)
-	}
-	https := httptest.NewServer(http.HandlerFunc(ctlHandler))
-	serverURL := https.URL
-	defer https.Close()
-	defer https.CloseClientConnections()
-
-	tmpdir, err := ioutil.TempDir("", "ipntest")
-	if err != nil {
-		t.Fatalf("create tempdir: %v\n", err)
-	}
-	ctl, err = control.New(tmpdir, serverURL, true)
-	if err != nil {
-		t.Fatalf("create control server: %v\n", ctl)
-	}
-
-	n1 := newNode(t, "n1", https)
-	defer n1.Backend.Shutdown()
-	n1.Backend.StartLoginInteractive()
-
-	n2 := newNode(t, "n2", https)
-	defer n2.Backend.Shutdown()
-	n2.Backend.StartLoginInteractive()
-
-	var s1, s2 State
-	for {
-		t.Logf("\n\nn1.state=%v n2.state=%v\n\n", s1, s2)
-
-		// TODO(crawshaw): switch from || to &&. To do this we need to
-		// transmit some data so that the handshake completes on both
-		// sides. (Because handshakes are 1RTT, it is the data
-		// transmission that completes the handshake.)
-		if s1 == Running || s2 == Running {
-			// TODO(apenwarr): ensure state sequence.
-			// Right now we'll just exit as soon as
-			// state==Running, even if the backend is lying or
-			// something. Not a great test.
-			break
-		}
-
-		select {
-		case n := <-n1.NotifyCh:
-			t.Logf("n1n: %v\n", n)
-			if n.State != nil {
-				s1 = *n.State
-				if s1 == NeedsMachineAuth {
-					authNode(t, ctl, n1.Backend)
-				}
-			}
-		case n := <-n2.NotifyCh:
-			t.Logf("n2n: %v\n", n)
-			if n.State != nil {
-				s2 = *n.State
-				if s2 == NeedsMachineAuth {
-					authNode(t, ctl, n2.Backend)
-				}
-			}
-		case <-time.After(3 * time.Second):
-			t.Fatalf("\n\n\nFATAL: timed out waiting for notifications.\n\n\n")
-		}
-	}
-
-	t.Skip("skipping ping tests, they are flaky") // TODO(crawshaw): this exposes a real bug!
-
-	n1addr := n1.Backend.NetMap().Addresses[0].IP
-	n2addr := n2.Backend.NetMap().Addresses[0].IP
-	t.Run("ping n2", func(t *testing.T) {
-		msg := tuntest.Ping(n2addr.IP(), n1addr.IP())
-		n1.ChannelTUN.Outbound <- msg
-		select {
-		case msgRecv := <-n2.ChannelTUN.Inbound:
-			if !bytes.Equal(msg, msgRecv) {
-				t.Error("bad ping")
-			}
-		case <-time.After(1 * time.Second):
-			t.Error("no ping seen")
-		}
-	})
-	t.Run("ping n1", func(t *testing.T) {
-		msg := tuntest.Ping(n1addr.IP(), n2addr.IP())
-		n2.ChannelTUN.Outbound <- msg
-		select {
-		case msgRecv := <-n1.ChannelTUN.Inbound:
-			if !bytes.Equal(msg, msgRecv) {
-				t.Error("bad ping")
-			}
-		case <-time.After(1 * time.Second):
-			t.Error("no ping seen")
-		}
-	})
-}
-
-type testNode struct {
-	Backend    *LocalBackend
-	ChannelTUN *tuntest.ChannelTUN
-	NotifyCh   <-chan Notify
-}
-
-// Create a new IPN node.
-func newNode(t *testing.T, prefix string, https *httptest.Server) testNode {
-	t.Helper()
-	logfe := func(fmt string, args ...interface{}) {
-		t.Logf(prefix+".e: "+fmt, args...)
-	}
-	logf := func(fmt string, args ...interface{}) {
-		t.Logf(prefix+": "+fmt, args...)
-	}
-
-	tun := tuntest.NewChannelTUN()
-	e1, err := wgengine.NewUserspaceEngineAdvanced(logfe, tun.TUN(), wgengine.NewFakeRouter, 0)
-	if err != nil {
-		t.Fatalf("NewFakeEngine: %v\n", err)
-	}
-	n, err := NewLocalBackend(logf, prefix, &MemoryStore{}, e1)
-	if err != nil {
-		t.Fatalf("NewLocalBackend: %v\n", err)
-	}
-	nch := make(chan Notify, 1000)
-	c := controlclient.Persist{
-		Provider:  "google",
-		LoginName: "test1@tailscale.com",
-	}
-	n.Start(Options{
-		FrontendLogID: prefix + "-f",
-		Prefs: &Prefs{
-			ControlURL:       https.URL,
-			RouteAll:         true,
-			AllowSingleHosts: true,
-			CorpDNS:          true,
-			WantRunning:      true,
-			Persist:          &c,
-		},
-		Notify: func(n Notify) {
-			// Automatically visit auth URLs
-			if n.BrowseToURL != nil {
-				t.Logf("\n\n\nURL! %vv\n", *n.BrowseToURL)
-				hc := https.Client()
-				_, err := hc.Get(*n.BrowseToURL)
-				if err != nil {
-					t.Logf("BrowseToURL: %v\n", err)
-				}
-			}
-			nch <- n
-		},
-	})
-
-	return testNode{
-		Backend:    n,
-		ChannelTUN: tun,
-		NotifyCh:   nch,
-	}
-}
-
-// Tell the control server to authorize the given node.
-func authNode(t *testing.T, ctl *control.Server, n *LocalBackend) {
-	mk := n.prefs.Persist.PrivateMachineKey.Public()
-	nk := n.prefs.Persist.PrivateNodeKey.Public()
-	ctl.AuthorizeMachine(tailcfg.MachineKey(mk), tailcfg.NodeKey(nk))
-}

ipn/fake_test.go

@@ -7,6 +7,10 @@ package ipn
 import (
 	"log"
 	"time"
+
+	"golang.org/x/oauth2"
+	"tailscale.com/control/controlclient"
+	"tailscale.com/ipn/ipnstate"
 )
 
 type FakeBackend struct {
@@ -39,10 +43,18 @@ func (b *FakeBackend) newState(s State) {
 func (b *FakeBackend) StartLoginInteractive() {
 	u := b.serverURL + "/this/is/fake"
 	b.notify(Notify{BrowseToURL: &u})
+	b.login()
+}
+
+func (b *FakeBackend) Login(token *oauth2.Token) {
+	b.login()
+}
+
+func (b *FakeBackend) login() {
 	b.newState(NeedsMachineAuth)
 	b.newState(Stopped)
 	// TODO(apenwarr): Fill in a more interesting netmap here.
-	b.notify(Notify{NetMap: &NetworkMap{}})
+	b.notify(Notify{NetMap: &controlclient.NetworkMap{}})
 	b.newState(Starting)
 	// TODO(apenwarr): Fill in a more interesting status.
 	b.notify(Notify{Engine: &EngineStatus{}})
@@ -58,7 +70,7 @@ func (b *FakeBackend) SetPrefs(new *Prefs) {
 		panic("FakeBackend.SetPrefs got nil prefs")
 	}
 
-	b.notify(Notify{Prefs: new.Copy()})
+	b.notify(Notify{Prefs: new.Clone()})
 	if new.WantRunning && !b.live {
 		b.newState(Starting)
 		b.newState(Running)
@@ -71,6 +83,10 @@ func (b *FakeBackend) RequestEngineStatus() {
 	b.notify(Notify{Engine: &EngineStatus{}})
 }
 
-func (b *FakeBackend) FakeExpireAfter(x time.Duration) {
-	b.notify(Notify{NetMap: &NetworkMap{}})
+func (b *FakeBackend) RequestStatus() {
+	b.notify(Notify{Status: &ipnstate.Status{}})
+}
+
+func (b *FakeBackend) FakeExpireAfter(x time.Duration) {
+	b.notify(Notify{NetMap: &controlclient.NetworkMap{}})
 }

ipn/handle.go

@@ -9,18 +9,20 @@ import (
 	"time"
 
 	"github.com/tailscale/wireguard-go/wgcfg"
+	"golang.org/x/oauth2"
+	"tailscale.com/control/controlclient"
 	"tailscale.com/types/logger"
 )
 
 type Handle struct {
 	frontendLogID string
 	b             Backend
-	xnotify       func(n Notify)
+	xnotify       func(Notify)
 	logf          logger.Logf
 
 	// Mutex protects everything below
 	mu                sync.Mutex
-	netmapCache       *NetworkMap
+	netmapCache       *controlclient.NetworkMap
 	engineStatusCache EngineStatus
 	stateCache        State
 	prefsCache        *Prefs
@@ -47,7 +49,7 @@ func (h *Handle) Start(opts Options) error {
 	h.engineStatusCache = EngineStatus{}
 	h.stateCache = NoState
 	if opts.Prefs != nil {
-		h.prefsCache = opts.Prefs.Copy()
+		h.prefsCache = opts.Prefs.Clone()
 	}
 	xopts := opts
 	xopts.Notify = h.notify
@@ -62,14 +64,14 @@ func (h *Handle) Reset() {
 func (h *Handle) notify(n Notify) {
 	h.mu.Lock()
 	if n.BackendLogID != nil {
-		h.logf("Handle: logs: be:%v fe:%v\n",
+		h.logf("Handle: logs: be:%v fe:%v",
 			*n.BackendLogID, h.frontendLogID)
 	}
 	if n.State != nil {
 		h.stateCache = *n.State
 	}
 	if n.Prefs != nil {
-		h.prefsCache = n.Prefs.Copy()
+		h.prefsCache = n.Prefs.Clone()
 	}
 	if n.NetMap != nil {
 		h.netmapCache = n.NetMap
@@ -89,14 +91,14 @@ func (h *Handle) Prefs() *Prefs {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
-	return h.prefsCache.Copy()
+	return h.prefsCache.Clone()
 }
 
 func (h *Handle) UpdatePrefs(updateFn func(p *Prefs)) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
-	new := h.prefsCache.Copy()
+	new := h.prefsCache.Clone()
 	updateFn(new)
 	h.prefsCache = new
 	h.b.SetPrefs(new)
@@ -127,7 +129,7 @@ func (h *Handle) LocalAddrs() []wgcfg.CIDR {
 	return []wgcfg.CIDR{}
 }
 
-func (h *Handle) NetMap() *NetworkMap {
+func (h *Handle) NetMap() *controlclient.NetworkMap {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
@@ -153,6 +155,10 @@ func (h *Handle) StartLoginInteractive() {
 	h.b.StartLoginInteractive()
 }
 
+func (h *Handle) Login(token *oauth2.Token) {
+	h.b.Login(token)
+}
+
 func (h *Handle) Logout() {
 	h.b.Logout()
 }
@@ -161,6 +167,10 @@ func (h *Handle) RequestEngineStatus() {
 	h.b.RequestEngineStatus()
 }
 
+func (h *Handle) RequestStatus() {
+	h.b.RequestStatus()
+}
+
 func (h *Handle) FakeExpireAfter(x time.Duration) {
 	h.b.FakeExpireAfter(x)
 }

ipn/ipnserver/server.go

@@ -10,20 +10,19 @@ import (
 	"fmt"
 	"log"
 	"net"
+	"net/http"
 	"os"
 	"os/exec"
 	"os/signal"
-	"strings"
 	"sync"
 	"syscall"
 	"time"
 
-	"github.com/google/go-cmp/cmp"
-	"github.com/klauspost/compress/zstd"
 	"tailscale.com/control/controlclient"
 	"tailscale.com/ipn"
 	"tailscale.com/logtail/backoff"
 	"tailscale.com/safesocket"
+	"tailscale.com/smallzstd"
 	"tailscale.com/types/logger"
 	"tailscale.com/version"
 	"tailscale.com/wgengine"
@@ -34,15 +33,19 @@ type Options struct {
 	// SocketPath, on unix systems, is the unix socket path to listen
 	// on for frontend connections.
 	SocketPath string
+
 	// Port, on windows, is the localhost TCP port to listen on for
 	// frontend connections.
 	Port int
+
 	// StatePath is the path to the stored agent state.
 	StatePath string
+
 	// AutostartStateKey, if non-empty, immediately starts the agent
 	// using the given StateKey. If empty, the agent stays idle and
 	// waits for a frontend to start it.
 	AutostartStateKey ipn.StateKey
+
 	// LegacyConfigPath optionally specifies the old-style relaynode
 	// relay.conf location. If both LegacyConfigPath and
 	// AutostartStateKey are specified and the requested state doesn't
@@ -52,43 +55,150 @@ type Options struct {
 	// TODO(danderson): remove some time after the transition to
 	// tailscaled is done.
 	LegacyConfigPath string
+
 	// SurviveDisconnects specifies how the server reacts to its
 	// frontend disconnecting. If true, the server keeps running on
 	// its existing state, and accepts new frontend connections. If
 	// false, the server dumps its state and becomes idle.
+	//
+	// To support CLI connections (notably, "tailscale status"),
+	// the actual definition of "disconnect" is when the
+	// connection count transitions from 1 to 0.
 	SurviveDisconnects bool
+
+	// DebugMux, if non-nil, specifies an HTTP ServeMux in which
+	// to register a debug handler.
+	DebugMux *http.ServeMux
+
+	// ErrorMessage, if not empty, signals that the server will exist
+	// only to relay the provided critical error message to the user.
+	ErrorMessage string
 }
 
-func pump(logf logger.Logf, ctx context.Context, bs *ipn.BackendServer, s net.Conn) {
-	defer logf("Control connection done.\n")
+// server is an IPN backend and its set of 0 or more active connections
+// talking to an IPN backend.
+type server struct {
+	resetOnZero bool // call bs.Reset on transition from 1->0 connections
 
-	for ctx.Err() == nil && !bs.GotQuit {
-		msg, err := ipn.ReadMsg(s)
+	bsMu sync.Mutex // lock order: bsMu, then mu
+	bs   *ipn.BackendServer
+
+	mu      sync.Mutex
+	clients map[net.Conn]bool
+}
+
+func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
+	s.addConn(c)
+	logf("incoming control connection")
+	defer s.removeAndCloseConn(c)
+	for ctx.Err() == nil {
+		msg, err := ipn.ReadMsg(c)
 		if err != nil {
-			logf("ReadMsg: %v\n", err)
-			break
+			if ctx.Err() == nil {
+				logf("ReadMsg: %v", err)
+			}
+			return
 		}
-		err = bs.GotCommandMsg(msg)
-		if err != nil {
-			logf("GotCommandMsg: %v\n", err)
-			break
+		s.bsMu.Lock()
+		if err := s.bs.GotCommandMsg(msg); err != nil {
+			logf("GotCommandMsg: %v", err)
+		}
+		gotQuit := s.bs.GotQuit
+		s.bsMu.Unlock()
+		if gotQuit {
+			return
 		}
 	}
 }
 
-func Run(rctx context.Context, logf logger.Logf, logid string, opts Options, e wgengine.Engine) error {
-	bo := backoff.Backoff{Name: "ipnserver"}
+func (s *server) addConn(c net.Conn) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.clients == nil {
+		s.clients = map[net.Conn]bool{}
+	}
+	s.clients[c] = true
+}
+
+func (s *server) removeAndCloseConn(c net.Conn) {
+	s.mu.Lock()
+	delete(s.clients, c)
+	remain := len(s.clients)
+	s.mu.Unlock()
+
+	if remain == 0 && s.resetOnZero {
+		s.bsMu.Lock()
+		s.bs.Reset()
+		s.bsMu.Unlock()
+	}
+	c.Close()
+}
+
+func (s *server) stopAll() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for c := range s.clients {
+		safesocket.ConnCloseRead(c)
+		safesocket.ConnCloseWrite(c)
+	}
+	s.clients = nil
+}
+
+func (s *server) writeToClients(b []byte) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for c := range s.clients {
+		ipn.WriteMsg(c, b)
+	}
+}
+
+func Run(ctx context.Context, logf logger.Logf, logid string, opts Options, e wgengine.Engine) error {
+	runDone := make(chan struct{})
+	defer close(runDone)
 
 	listen, _, err := safesocket.Listen(opts.SocketPath, uint16(opts.Port))
 	if err != nil {
 		return fmt.Errorf("safesocket.Listen: %v", err)
 	}
-	// Go listeners can't take a context, close it instead.
+
+	server := &server{
+		resetOnZero: !opts.SurviveDisconnects,
+	}
+
+	// When the context is closed or when we return, whichever is first, close our listner
+	// and all open connections.
 	go func() {
-		<-rctx.Done()
+		select {
+		case <-ctx.Done():
+		case <-runDone:
+		}
+		server.stopAll()
 		listen.Close()
 	}()
-	logf("Listening on %v\n", listen.Addr())
+	logf("Listening on %v", listen.Addr())
+
+	bo := backoff.NewBackoff("ipnserver", logf)
+
+	if opts.ErrorMessage != "" {
+		for i := 1; ctx.Err() == nil; i++ {
+			s, err := listen.Accept()
+			if err != nil {
+				logf("%d: Accept: %v", i, err)
+				bo.BackOff(ctx, err)
+				continue
+			}
+			serverToClient := func(b []byte) {
+				ipn.WriteMsg(s, b)
+			}
+			go func() {
+				defer s.Close()
+				bs := ipn.NewBackendServer(logf, nil, serverToClient)
+				bs.SendErrorMessage(opts.ErrorMessage)
+				s.Read(make([]byte, 1))
+			}()
+		}
+		return ctx.Err()
+	}
 
 	var store ipn.StateStore
 	if opts.StatePath != "" {
@@ -104,22 +214,24 @@ func Run(rctx context.Context, logf logger.Logf, logid string, opts Options, e w
 	if err != nil {
 		return fmt.Errorf("NewLocalBackend: %v", err)
 	}
+	defer b.Shutdown()
 	b.SetDecompressor(func() (controlclient.Decompressor, error) {
-		return zstd.NewReader(nil)
+		return smallzstd.NewDecoder(nil)
 	})
-	b.SetCmpDiff(func(x, y interface{}) string { return cmp.Diff(x, y) })
 
-	var s net.Conn
-	serverToClient := func(b []byte) {
-		if s != nil {
-			ipn.WriteMsg(s, b)
-		}
+	if opts.DebugMux != nil {
+		opts.DebugMux.HandleFunc("/debug/ipn", func(w http.ResponseWriter, r *http.Request) {
+			w.Header().Set("Content-Type", "text/html; charset=utf-8")
+			st := b.Status()
+			// TODO(bradfitz): add LogID and opts to st?
+			st.WriteHTML(w)
+		})
 	}
 
-	bs := ipn.NewBackendServer(logf, b, serverToClient)
+	server.bs = ipn.NewBackendServer(logf, b, server.writeToClients)
 
 	if opts.AutostartStateKey != "" {
-		bs.GotCommand(&ipn.Command{
+		server.bs.GotCommand(&ipn.Command{
 			Version: version.LONG,
 			Start: &ipn.StartArgs{
 				Opts: ipn.Options{
@@ -130,57 +242,18 @@ func Run(rctx context.Context, logf logger.Logf, logid string, opts Options, e w
 		})
 	}
 
-	var oldS net.Conn
-	//lint:ignore SA4006 ctx is never used, but has to be defined so
-	// that it can be assigned to in the following for loop. It's a
-	// bit of necessary code convolution to work around Go's variable
-	// shadowing rules.
-	ctx, cancel := context.WithCancel(rctx)
-
-	stopAll := func() {
-		// Currently we only support one client connection at a time.
-		// Theoretically we could allow multiple clients, by passing
-		// notifications to all of them and accepting commands from
-		// any of them, but there doesn't seem to be much need for
-		// that right now.
-		if oldS != nil {
-			cancel()
-			safesocket.ConnCloseRead(oldS)
-			safesocket.ConnCloseWrite(oldS)
-		}
-	}
-
-	for i := 1; rctx.Err() == nil; i++ {
-		s, err = listen.Accept()
+	for i := 1; ctx.Err() == nil; i++ {
+		c, err := listen.Accept()
 		if err != nil {
-			logf("%d: Accept: %v\n", i, err)
-			bo.BackOff(rctx, err)
+			if ctx.Err() == nil {
+				logf("ipnserver: Accept: %v", err)
+				bo.BackOff(ctx, err)
+			}
 			continue
 		}
-		logf("%d: Incoming control connection.\n", i)
-		stopAll()
-
-		ctx, cancel = context.WithCancel(context.Background())
-		oldS = s
-
-		go func(ctx context.Context, bs *ipn.BackendServer, s net.Conn, i int) {
-			si := fmt.Sprintf("%d: ", i)
-			pump(func(fmt string, args ...interface{}) {
-				logf(si+fmt, args...)
-			}, ctx, bs, s)
-			if !opts.SurviveDisconnects || bs.GotQuit {
-				bs.Reset()
-				s.Close()
-			}
-			// Quitting not allowed, just keep going.
-			bs.GotQuit = false
-		}(ctx, bs, s, i)
-
-		bo.BackOff(ctx, nil)
+		go server.serveConn(ctx, c, logger.WithPrefix(logf, fmt.Sprintf("ipnserver: conn%d: ", i)))
 	}
-	stopAll()
-
-	return rctx.Err()
+	return ctx.Err()
 }
 
 func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
@@ -202,10 +275,10 @@ func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
 		var sig os.Signal
 		select {
 		case sig = <-interrupt:
-			logf("BabysitProc: got signal: %v\n", sig)
+			logf("BabysitProc: got signal: %v", sig)
 			close(done)
 		case <-ctx.Done():
-			logf("BabysitProc: context done\n")
+			logf("BabysitProc: context done")
 			sig = os.Kill
 			close(done)
 		}
@@ -215,11 +288,11 @@ func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
 		proc.mu.Unlock()
 	}()
 
-	bo := backoff.Backoff{Name: "BabysitProc"}
+	bo := backoff.NewBackoff("BabysitProc", logf)
 
 	for {
 		startTime := time.Now()
-		log.Printf("exec: %#v %v\n", executable, args)
+		log.Printf("exec: %#v %v", executable, args)
 		cmd := exec.Command(executable, args...)
 
 		// Create a pipe object to use as the subproc's stdin.
@@ -230,7 +303,7 @@ func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
 		// We never need to actually write to wStdin.
 		rStdin, wStdin, err := os.Pipe()
 		if err != nil {
-			log.Printf("os.Pipe 1: %v\n", err)
+			log.Printf("os.Pipe 1: %v", err)
 			return
 		}
 
@@ -242,7 +315,7 @@ func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
 		// logf() calls. bufio is more appropriate.
 		rStdout, wStdout, err := os.Pipe()
 		if err != nil {
-			log.Printf("os.Pipe 2: %v\n", err)
+			log.Printf("os.Pipe 2: %v", err)
 		}
 		go func(r *os.File) {
 			defer r.Close()
@@ -250,7 +323,7 @@ func BabysitProc(ctx context.Context, args []string, logf logger.Logf) {
 			for {
 				s, err := rb.ReadString('\n')
 				if s != "" {
-					logf("%s\n", strings.TrimSuffix(s, "\n"))
+					logf("%s", s)
 				}
 				if err != nil {
 					break

ipn/ipnserver/server_test.go

@@ -0,0 +1,77 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package ipnserver_test
+
+import (
+	"context"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"tailscale.com/ipn"
+	"tailscale.com/ipn/ipnserver"
+	"tailscale.com/safesocket"
+	"tailscale.com/wgengine"
+)
+
+func TestRunMultipleAccepts(t *testing.T) {
+	t.Skipf("TODO(bradfitz): finish this test, once other fires are out")
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	td, err := ioutil.TempDir("", "TestRunMultipleAccepts")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer os.RemoveAll(td)
+	socketPath := filepath.Join(td, "tailscale.sock")
+
+	logf := func(format string, args ...interface{}) {
+		format = strings.TrimRight(format, "\n")
+		println(fmt.Sprintf(format, args...))
+		t.Logf(format, args...)
+	}
+
+	connect := func() {
+		for i := 1; i <= 2; i++ {
+			logf("connect %d ...", i)
+			c, err := safesocket.Connect(socketPath, 0)
+			if err != nil {
+				t.Fatalf("safesocket.Connect: %v\n", err)
+			}
+			clientToServer := func(b []byte) {
+				ipn.WriteMsg(c, b)
+			}
+			bc := ipn.NewBackendClient(logf, clientToServer)
+			prefs := ipn.NewPrefs()
+			bc.SetPrefs(prefs)
+			c.Close()
+		}
+	}
+
+	logTriggerTestf := func(format string, args ...interface{}) {
+		logf(format, args...)
+		if strings.HasPrefix(format, "Listening on ") {
+			connect()
+		}
+	}
+
+	eng, err := wgengine.NewFakeUserspaceEngine(logf, 0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer eng.Close()
+
+	opts := ipnserver.Options{
+		SocketPath: socketPath,
+	}
+	t.Logf("pre-Run")
+	err = ipnserver.Run(ctx, logTriggerTestf, "dummy_logid", opts, eng)
+	t.Logf("ipnserver.Run = %v", err)
+}

ipn/ipnstate/ipnstate.go

@@ -0,0 +1,304 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package ipnstate captures the entire state of the Tailscale network.
+//
+// It's a leaf package so ipn, wgengine, and magicsock can all depend on it.
+package ipnstate
+
+import (
+	"bytes"
+	"fmt"
+	"html"
+	"io"
+	"log"
+	"sort"
+	"strings"
+	"sync"
+	"time"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/key"
+)
+
+// Status represents the entire state of the IPN network.
+type Status struct {
+	BackendState string
+	Peer         map[key.Public]*PeerStatus
+	User         map[tailcfg.UserID]tailcfg.UserProfile
+}
+
+func (s *Status) Peers() []key.Public {
+	kk := make([]key.Public, 0, len(s.Peer))
+	for k := range s.Peer {
+		kk = append(kk, k)
+	}
+	sort.Slice(kk, func(i, j int) bool { return bytes.Compare(kk[i][:], kk[j][:]) < 0 })
+	return kk
+}
+
+type PeerStatus struct {
+	PublicKey key.Public
+	HostName  string // HostInfo's Hostname (not a DNS name or necessarily unique)
+	OS        string // HostInfo.OS
+	UserID    tailcfg.UserID
+
+	TailAddr string // Tailscale IP
+
+	// Endpoints:
+	Addrs   []string
+	CurAddr string // one of Addrs, or unique if roaming
+	Relay   string // DERP region
+
+	RxBytes       int64
+	TxBytes       int64
+	Created       time.Time // time registered with tailcontrol
+	LastWrite     time.Time // time last packet sent
+	LastSeen      time.Time // last seen to tailcontrol
+	LastHandshake time.Time // with local wireguard
+	KeepAlive     bool
+
+	// InNetworkMap means that this peer was seen in our latest network map.
+	// In theory, all of InNetworkMap and InMagicSock and InEngine should all be true.
+	InNetworkMap bool
+
+	// InMagicSock means that this peer is being tracked by magicsock.
+	// In theory, all of InNetworkMap and InMagicSock and InEngine should all be true.
+	InMagicSock bool
+
+	// InEngine means that this peer is tracked by the wireguard engine.
+	// In theory, all of InNetworkMap and InMagicSock and InEngine should all be true.
+	InEngine bool
+}
+
+// SimpleHostName returns a potentially simplified version of ps.HostName for display purposes.
+func (ps *PeerStatus) SimpleHostName() string {
+	n := ps.HostName
+	n = strings.TrimSuffix(n, ".local")
+	n = strings.TrimSuffix(n, ".localdomain")
+	return n
+}
+
+type StatusBuilder struct {
+	mu     sync.Mutex
+	locked bool
+	st     Status
+}
+
+func (sb *StatusBuilder) Status() *Status {
+	sb.mu.Lock()
+	defer sb.mu.Unlock()
+	sb.locked = true
+	return &sb.st
+}
+
+// AddUser adds a user profile to the status.
+func (sb *StatusBuilder) AddUser(id tailcfg.UserID, up tailcfg.UserProfile) {
+	sb.mu.Lock()
+	defer sb.mu.Unlock()
+	if sb.locked {
+		log.Printf("[unexpected] ipnstate: AddUser after Locked")
+		return
+	}
+
+	if sb.st.User == nil {
+		sb.st.User = make(map[tailcfg.UserID]tailcfg.UserProfile)
+	}
+
+	sb.st.User[id] = up
+}
+
+// AddPeer adds a peer node to the status.
+//
+// Its PeerStatus is mixed with any previous status already added.
+func (sb *StatusBuilder) AddPeer(peer key.Public, st *PeerStatus) {
+	if st == nil {
+		panic("nil PeerStatus")
+	}
+
+	sb.mu.Lock()
+	defer sb.mu.Unlock()
+	if sb.locked {
+		log.Printf("[unexpected] ipnstate: AddPeer after Locked")
+		return
+	}
+
+	if sb.st.Peer == nil {
+		sb.st.Peer = make(map[key.Public]*PeerStatus)
+	}
+	e, ok := sb.st.Peer[peer]
+	if !ok {
+		sb.st.Peer[peer] = st
+		st.PublicKey = peer
+		return
+	}
+
+	if v := st.HostName; v != "" {
+		e.HostName = v
+	}
+	if v := st.Relay; v != "" {
+		e.Relay = v
+	}
+	if v := st.UserID; v != 0 {
+		e.UserID = v
+	}
+	if v := st.TailAddr; v != "" {
+		e.TailAddr = v
+	}
+	if v := st.OS; v != "" {
+		e.OS = st.OS
+	}
+	if v := st.Addrs; v != nil {
+		e.Addrs = v
+	}
+	if v := st.CurAddr; v != "" {
+		e.CurAddr = v
+	}
+	if v := st.RxBytes; v != 0 {
+		e.RxBytes = v
+	}
+	if v := st.TxBytes; v != 0 {
+		e.TxBytes = v
+	}
+	if v := st.LastHandshake; !v.IsZero() {
+		e.LastHandshake = v
+	}
+	if v := st.Created; !v.IsZero() {
+		e.Created = v
+	}
+	if v := st.LastSeen; !v.IsZero() {
+		e.LastSeen = v
+	}
+	if v := st.LastWrite; !v.IsZero() {
+		e.LastWrite = v
+	}
+	if st.InNetworkMap {
+		e.InNetworkMap = true
+	}
+	if st.InMagicSock {
+		e.InMagicSock = true
+	}
+	if st.InEngine {
+		e.InEngine = true
+	}
+	if st.KeepAlive {
+		e.KeepAlive = true
+	}
+}
+
+type StatusUpdater interface {
+	UpdateStatus(*StatusBuilder)
+}
+
+func (st *Status) WriteHTML(w io.Writer) {
+	f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
+
+	f(`<!DOCTYPE html>
+<html lang="en">
+<head>
+<title>Tailscale State</title>
+<style>
+body { font-family: monospace; }
+.owner { text-decoration: underline; }
+.tailaddr { font-style: italic; }
+.acenter { text-align: center; }
+.aright { text-align: right; }
+table, th, td { border: 1px solid black; border-spacing : 0; border-collapse : collapse; }
+thead { background-color: #FFA500; }
+th, td { padding: 5px; }
+td { vertical-align: top; }
+table tbody tr:nth-child(even) td { background-color: #f5f5f5; }
+</style>
+</head>
+<body>
+<h1>Tailscale State</h1>
+`)
+
+	//f("<p><b>logid:</b> %s</p>\n", logid)
+	//f("<p><b>opts:</b> <code>%s</code></p>\n", html.EscapeString(fmt.Sprintf("%+v", opts)))
+
+	f("<table>\n<thead>\n")
+	f("<tr><th>Peer</th><th>Node</th><th>Owner</th><th>Rx</th><th>Tx</th><th>Activity</th><th>Endpoints</th></tr>\n")
+	f("</thead>\n<tbody>\n")
+
+	now := time.Now()
+
+	for _, peer := range st.Peers() {
+		ps := st.Peer[peer]
+		var actAgo string
+		if !ps.LastWrite.IsZero() {
+			ago := now.Sub(ps.LastWrite)
+			actAgo = ago.Round(time.Second).String() + " ago"
+			if ago < 5*time.Minute {
+				actAgo = "<b>" + actAgo + "</b>"
+			}
+		}
+		var owner string
+		if up, ok := st.User[ps.UserID]; ok {
+			owner = up.LoginName
+			if i := strings.Index(owner, "@"); i != -1 {
+				owner = owner[:i]
+			}
+		}
+		f("<tr><td>%s</td><td>%s %s<br><span class=\"tailaddr\">%s</span></td><td class=\"acenter owner\">%s</td><td class=\"aright\">%v</td><td class=\"aright\">%v</td><td class=\"aright\">%v</td>",
+			peer.ShortString(),
+			html.EscapeString(ps.SimpleHostName()),
+			osEmoji(ps.OS),
+			ps.TailAddr,
+			html.EscapeString(owner),
+			ps.RxBytes,
+			ps.TxBytes,
+			actAgo,
+		)
+		f("<td class=\"aright\">")
+		// TODO: let server report this active bool instead
+		active := !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
+		relay := ps.Relay
+		if relay != "" {
+			if active && ps.CurAddr == "" {
+				f("🔗 <b>derp-%v</b><br>", html.EscapeString(relay))
+			} else {
+				f("derp-%v<br>", html.EscapeString(relay))
+			}
+		}
+
+		match := false
+		for _, addr := range ps.Addrs {
+			if addr == ps.CurAddr {
+				match = true
+				f("🔗 <b>%s</b><br>", addr)
+			} else {
+				f("%s<br>", addr)
+			}
+		}
+		if ps.CurAddr != "" && !match {
+			f("<b>%s</b> \xf0\x9f\xa7\xb3<br>", ps.CurAddr)
+		}
+		f("</td>") // end Addrs
+
+		f("</tr>\n")
+	}
+	f("</tbody>\n</table>\n")
+	f("</body>\n</html>\n")
+}
+
+func osEmoji(os string) string {
+	switch os {
+	case "linux":
+		return "🐧"
+	case "macOS":
+		return "🍎"
+	case "windows":
+		return "🖥️"
+	case "iOS":
+		return "📱"
+	case "android":
+		return "🤖"
+	case "freebsd":
+		return "👿"
+	case "openbsd":
+		return "🐡"
+	}
+	return "👽"
+}

ipn/message.go

@@ -13,7 +13,9 @@ import (
 	"log"
 	"time"
 
+	"golang.org/x/oauth2"
 	"tailscale.com/types/logger"
+	"tailscale.com/types/structs"
 	"tailscale.com/version"
 )
 
@@ -34,23 +36,33 @@ type FakeExpireAfterArgs struct {
 // Command is a command message that is JSON encoded and sent by a
 // frontend to a backend.
 type Command struct {
+	_ structs.Incomparable
+
+	// Version is the binary version of the frontend (the client).
 	Version string
 
+	// AllowVersionSkew controls whether it's permitted for the
+	// client and server to have a different version. The default
+	// (false) means to be strict.
+	AllowVersionSkew bool
+
 	// Exactly one of the following must be non-nil.
 	Quit                  *NoArgs
 	Start                 *StartArgs
 	StartLoginInteractive *NoArgs
+	Login                 *oauth2.Token
 	Logout                *NoArgs
 	SetPrefs              *SetPrefsArgs
 	RequestEngineStatus   *NoArgs
+	RequestStatus         *NoArgs
 	FakeExpireAfter       *FakeExpireAfterArgs
 }
 
 type BackendServer struct {
 	logf          logger.Logf
-	b             Backend        // the Backend we are serving up
-	sendNotifyMsg func(b []byte) // send a notification message
-	GotQuit       bool           // a Quit command was received
+	b             Backend              // the Backend we are serving up
+	sendNotifyMsg func(jsonMsg []byte) // send a notification message
+	GotQuit       bool                 // a Quit command was received
 }
 
 func NewBackendServer(logf logger.Logf, b Backend, sendNotifyMsg func(b []byte)) *BackendServer {
@@ -65,25 +77,38 @@ func (bs *BackendServer) send(n Notify) {
 	n.Version = version.LONG
 	b, err := json.Marshal(n)
 	if err != nil {
-		log.Fatalf("Failed json.Marshal(notify): %v\n%#v\n", err, n)
+		log.Fatalf("Failed json.Marshal(notify): %v\n%#v", err, n)
 	}
 	bs.sendNotifyMsg(b)
 }
 
-// Inform the BackendServer of an incoming message.
+func (bs *BackendServer) SendErrorMessage(msg string) {
+	bs.send(Notify{ErrMessage: &msg})
+}
+
+// GotCommandMsg parses the incoming message b as a JSON Command and
+// calls GotCommand with it.
 func (bs *BackendServer) GotCommandMsg(b []byte) error {
-	cmd := Command{}
-	if err := json.Unmarshal(b, &cmd); err != nil {
+	cmd := &Command{}
+	if len(b) == 0 {
+		return nil
+	}
+	if err := json.Unmarshal(b, cmd); err != nil {
 		return err
 	}
-	return bs.GotCommand(&cmd)
+	return bs.GotCommand(cmd)
+}
+
+func (bs *BackendServer) GotFakeCommand(cmd *Command) error {
+	cmd.Version = version.LONG
+	return bs.GotCommand(cmd)
 }
 
 func (bs *BackendServer) GotCommand(cmd *Command) error {
-	if cmd.Version != version.LONG {
-		vs := fmt.Sprintf("Version mismatch! frontend=%#v backend=%#v\n",
+	if cmd.Version != version.LONG && !cmd.AllowVersionSkew {
+		vs := fmt.Sprintf("GotCommand: Version mismatch! frontend=%#v backend=%#v",
 			cmd.Version, version.LONG)
-		bs.logf("%s\n", vs)
+		bs.logf("%s", vs)
 		// ignore the command, but send a message back to the
 		// caller so it can realize the version mismatch too.
 		// We don't want to exit because it might cause a crash
@@ -105,6 +130,9 @@ func (bs *BackendServer) GotCommand(cmd *Command) error {
 	} else if c := cmd.StartLoginInteractive; c != nil {
 		bs.b.StartLoginInteractive()
 		return nil
+	} else if c := cmd.Login; c != nil {
+		bs.b.Login(c)
+		return nil
 	} else if c := cmd.Logout; c != nil {
 		bs.b.Logout()
 		return nil
@@ -114,6 +142,9 @@ func (bs *BackendServer) GotCommand(cmd *Command) error {
 	} else if c := cmd.RequestEngineStatus; c != nil {
 		bs.b.RequestEngineStatus()
 		return nil
+	} else if c := cmd.RequestStatus; c != nil {
+		bs.b.RequestStatus()
+		return nil
 	} else if c := cmd.FakeExpireAfter; c != nil {
 		bs.b.FakeExpireAfter(c.Duration)
 		return nil
@@ -125,16 +156,20 @@ func (bs *BackendServer) GotCommand(cmd *Command) error {
 func (bs *BackendServer) Reset() error {
 	// Tell the backend we got a Logout command, which will cause it
 	// to forget all its authentication information.
-	return bs.GotCommand(&Command{Logout: &NoArgs{}})
+	return bs.GotFakeCommand(&Command{Logout: &NoArgs{}})
 }
 
 type BackendClient struct {
 	logf           logger.Logf
-	sendCommandMsg func(b []byte)
-	notify         func(n Notify)
+	sendCommandMsg func(jsonb []byte)
+	notify         func(Notify)
+
+	// AllowVersionSkew controls whether to allow mismatched
+	// frontend & backend versions.
+	AllowVersionSkew bool
 }
 
-func NewBackendClient(logf logger.Logf, sendCommandMsg func(b []byte)) *BackendClient {
+func NewBackendClient(logf logger.Logf, sendCommandMsg func(jsonb []byte)) *BackendClient {
 	return &BackendClient{
 		logf:           logf,
 		sendCommandMsg: sendCommandMsg,
@@ -142,14 +177,18 @@ func NewBackendClient(logf logger.Logf, sendCommandMsg func(b []byte)) *BackendC
 }
 
 func (bc *BackendClient) GotNotifyMsg(b []byte) {
+	if len(b) == 0 {
+		// not interesting
+		return
+	}
 	n := Notify{}
 	if err := json.Unmarshal(b, &n); err != nil {
-		log.Fatalf("BackendClient.Notify: cannot decode message")
+		log.Fatalf("BackendClient.Notify: cannot decode message (length=%d)\n%#v", len(b), string(b))
 	}
-	if n.Version != version.LONG {
-		vs := fmt.Sprintf("Version mismatch! frontend=%#v backend=%#v",
+	if n.Version != version.LONG && !bc.AllowVersionSkew {
+		vs := fmt.Sprintf("GotNotify: Version mismatch! frontend=%#v backend=%#v",
 			version.LONG, n.Version)
-		bc.logf("%s\n", vs)
+		bc.logf("%s", vs)
 		// delete anything in the notification except the version,
 		// to prevent incorrect operation.
 		n = Notify{
@@ -171,6 +210,10 @@ func (bc *BackendClient) send(cmd Command) {
 	bc.sendCommandMsg(b)
 }
 
+func (bc *BackendClient) SetNotifyCallback(fn func(Notify)) {
+	bc.notify = fn
+}
+
 func (bc *BackendClient) Quit() error {
 	bc.send(Command{Quit: &NoArgs{}})
 	return nil
@@ -187,6 +230,10 @@ func (bc *BackendClient) StartLoginInteractive() {
 	bc.send(Command{StartLoginInteractive: &NoArgs{}})
 }
 
+func (bc *BackendClient) Login(token *oauth2.Token) {
+	bc.send(Command{Login: token})
+}
+
 func (bc *BackendClient) Logout() {
 	bc.send(Command{Logout: &NoArgs{}})
 }
@@ -199,11 +246,16 @@ func (bc *BackendClient) RequestEngineStatus() {
 	bc.send(Command{RequestEngineStatus: &NoArgs{}})
 }
 
+func (bc *BackendClient) RequestStatus() {
+	bc.send(Command{AllowVersionSkew: true, RequestStatus: &NoArgs{}})
+}
+
 func (bc *BackendClient) FakeExpireAfter(x time.Duration) {
 	bc.send(Command{FakeExpireAfter: &FakeExpireAfterArgs{Duration: x}})
 }
 
-const MSG_MAX = 1024 * 1024
+// MaxMessageSize is the maximum message size, in bytes.
+const MaxMessageSize = 1 << 20
 
 // TODO(apenwarr): incremental json decode?
 //  That would let us avoid storing the whole byte array uselessly in RAM.
@@ -214,14 +266,17 @@ func ReadMsg(r io.Reader) ([]byte, error) {
 		return nil, err
 	}
 	n := binary.LittleEndian.Uint32(cb)
-	if n > 1024*1024 {
+	if n > MaxMessageSize {
 		return nil, fmt.Errorf("ipn.Read: message too large: %v bytes", n)
 	}
 	b := make([]byte, n)
-	_, err = io.ReadFull(r, b)
+	nn, err := io.ReadFull(r, b)
 	if err != nil {
 		return nil, err
 	}
+	if nn != int(n) {
+		return nil, fmt.Errorf("ipn.Read: expected %v bytes, got %v", n, nn)
+	}
 	return b, nil
 }
 
@@ -229,8 +284,12 @@ func ReadMsg(r io.Reader) ([]byte, error) {
 //  That would save RAM, at the expense of having to encode once so that
 //  we can produce the initial byte count.
 func WriteMsg(w io.Writer, b []byte) error {
+	// TODO(bradfitz): this does two writes to w, which likely
+	// does two writes on the wire, two frame generations, etc. We
+	// should take a concrete buffered type, or use a sync.Pool to
+	// allocate a buf and do one write.
 	cb := make([]byte, 4)
-	if len(b) > MSG_MAX {
+	if len(b) > MaxMessageSize {
 		return fmt.Errorf("ipn.Write: message too large: %v bytes", len(b))
 	}
 	binary.LittleEndian.PutUint32(cb, uint32(len(b)))

ipn/message_test.go

@@ -9,14 +9,14 @@ import (
 	"testing"
 	"time"
 
-	"tailscale.com/testy"
+	"golang.org/x/oauth2"
+	"tailscale.com/tstest"
 )
 
 func TestReadWrite(t *testing.T) {
-	testy.FixLogs(t)
-	defer testy.UnfixLogs(t)
+	tstest.PanicOnLog()
 
-	rc := testy.NewResourceCheck()
+	rc := tstest.NewResourceCheck()
 	defer rc.Assert(t)
 
 	buf := bytes.Buffer{}
@@ -62,10 +62,9 @@ func TestReadWrite(t *testing.T) {
 }
 
 func TestClientServer(t *testing.T) {
-	testy.FixLogs(t)
-	defer testy.UnfixLogs(t)
+	tstest.PanicOnLog()
 
-	rc := testy.NewResourceCheck()
+	rc := tstest.NewResourceCheck()
 	defer rc.Assert(t)
 
 	b := &FakeBackend{}
@@ -179,4 +178,10 @@ func TestClientServer(t *testing.T) {
 
 	h.Logout()
 	flushUntil(NeedsLogin)
+
+	h.Login(&oauth2.Token{
+		AccessToken: "google_id_token",
+		TokenType:   GoogleIDTokenType,
+	})
+	flushUntil(Running)
 }

ipn/policy/policy.go

@@ -0,0 +1,42 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package policy contains various policy decisions that need to be
+// shared between the node client & control server.
+package policy
+
+import "tailscale.com/tailcfg"
+
+// IsInterestingService reports whether service s on the given operating
+// system (a version.OS value) is an interesting enough port to report
+// to our peer nodes for discovery purposes.
+func IsInterestingService(s tailcfg.Service, os string) bool {
+	if s.Proto != tailcfg.TCP {
+		return false
+	}
+	if os != "windows" {
+		// For non-Windows machines, assume all TCP listeners
+		// are interesting enough. We don't see listener spam
+		// there.
+		return true
+	}
+	// Windows has tons of TCP listeners. We need to move to a blacklist
+	// model later, but for now we just whitelist some common ones:
+	switch s.Port {
+	case 22, // ssh
+		80,    // http
+		443,   // https (but no hostname, so little useless)
+		3389,  // rdp
+		5900,  // vnc
+		32400, // plex
+
+		// And now some arbitary HTTP dev server ports:
+		// Eventually we'll remove this and make all ports
+		// work, once we nicely filter away noisy system
+		// ports.
+		8000, 8080, 8443, 8888:
+		return true
+	}
+	return false
+}