More logging :(

The cyolosecurity fork of certstore did not update its module name and
thus can only be used with a replace directive. This interferes with
installing using `go install` so I created a tailscale fork with an
updated module name.

Signed-off-by: Adrian Dewhurst <adrian@tailscale.com>

.github/ISSUE_TEMPLATE/bug_report.md

@@ -2,36 +2,7 @@
 name: Bug report
 about: Create a bug report
 title: ''
-labels: ''
+labels: 'needs-triage'
 assignees: ''
 
 ---
-
-<!-- Please note, this template is for definite bugs, not requests for
-support. If you need help with Tailscale, please email
-support@tailscale.com. We don't provide support via Github issues. -->
-
-**Describe the bug**
-A clear and concise description of what the bug is.
-
-**To Reproduce**
-Steps to reproduce the behavior:
-1. Go to '...'
-2. Click on '....'
-3. Scroll down to '....'
-4. See error
-
-**Expected behavior**
-A clear and concise description of what you expected to happen.
-
-**Screenshots**
-If applicable, add screenshots to help explain your problem.
-
-**Version information:**
- - Device: [e.g. iPhone X, laptop]
- - OS: [e.g. Windows, MacOS]
- - OS version: [e.g. Windows 10, Ubuntu 18.04]
- - Tailscale version: [e.g. 0.95-0]
-
-**Additional context**
-Add any other context about the problem here.

.github/ISSUE_TEMPLATE/feature_request.md

@@ -2,25 +2,6 @@
 name: Feature request
 about: Suggest an idea for this project
 title: ''
-labels: ''
+labels: 'needs-triage'
 assignees: ''
-
 ---
-
-**Is your feature request related to a problem? Please describe.**
-
-A clear and concise description of what the problem is. Ex. I'm always
-frustrated when [...]
-
-**Describe the solution you'd like**
-
-A clear and concise description of what you want to happen.
-
-**Describe alternatives you've considered**
-
-A clear and concise description of any alternative solutions or
-features you've considered.
-
-**Additional context**
-
-Add any other context or screenshots about the feature request here.

.github/workflows/coverage.yml

@@ -1,48 +0,0 @@
-name: Code Coverage
-
-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: 1.16
-      id: go
-
-    - name: Check out code into the Go module directory
-      uses: actions/checkout@v1
-
-      # https://markphelps.me/2019/11/speed-up-your-go-builds-with-actions-cache/
-    - name: Restore Cache
-      uses: actions/cache@preview
-      id: cache
-      with:
-        path: ~/go/pkg/mod
-        key: ${{ runner.os }}-${{ hashFiles('**/go.sum') }}
-
-    - name: Basic build
-      run: go build ./cmd/...
-
-    - name: Run tests on linux with coverage data
-      run: go test -race -coverprofile=coverage.txt -bench=. -benchtime=1x ./...
-
-    - name: coveralls.io
-      uses: shogo82148/actions-goveralls@v1
-      env:
-        COVERALLS_TOKEN: ${{ secrets.COVERALLS_TOKEN }}
-        GITHUB_TOKEN: ${{ secrets.COVERALLS_BOT_PUBLIC_REPO_TOKEN }}
-      with:
-        path-to-profile: ./coverage.txt

.github/workflows/linux-race.yml

@@ -0,0 +1,48 @@
+name: Linux race
+
+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: 1.16
+      id: go
+
+    - name: Check out code into the Go module directory
+      uses: actions/checkout@v1
+
+    - name: Basic build
+      run: go build ./cmd/...
+
+    - name: Run tests and benchmarks with -race flag on linux
+      run: go test -race -bench=. -benchtime=1x ./...
+
+    - 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

@@ -29,7 +29,7 @@ jobs:
       run: go build ./cmd/...
 
     - name: Run tests on linux
-      run: go test ./...
+      run: go test -bench=. -benchtime=1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/linux32.yml

@@ -29,7 +29,7 @@ jobs:
       run: GOARCH=386 go build ./cmd/...
 
     - name: Run tests on linux
-      run: GOARCH=386 go test ./...
+      run: GOARCH=386 go test -bench=. -benchtime=1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/staticcheck.yml

@@ -24,11 +24,23 @@ jobs:
     - name: Run go vet
       run: go vet ./...
 
-    - name: Print staticcheck version
-      run: go run honnef.co/go/tools/cmd/staticcheck -version
+    - name: Install staticcheck
+      run: "GOBIN=~/.local/bin go install honnef.co/go/tools/cmd/staticcheck"
 
-    - name: Run staticcheck
-      run: "go run honnef.co/go/tools/cmd/staticcheck -- $(go list ./... | grep -v tempfork)"
+    - name: Print staticcheck version
+      run: "staticcheck -version"
+
+    - name: Run staticcheck (linux/amd64)
+      run: "GOOS=linux GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (darwin/amd64)
+      run: "GOOS=darwin GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (windows/amd64)
+      run: "GOOS=windows GOARCH=amd64 staticcheck -- $(go list ./... | grep -v tempfork)"
+
+    - name: Run staticcheck (windows/386)
+      run: "GOOS=windows GOARCH=386 staticcheck -- $(go list ./... | grep -v tempfork)"
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

.github/workflows/windows-race.yml

@@ -0,0 +1,55 @@
+name: Windows race
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - '*'
+
+jobs:
+  test:
+    runs-on: windows-latest
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+
+    - name: Install Go
+      uses: actions/setup-go@v2
+      with:
+        go-version: 1.16.x
+
+    - name: Checkout code
+      uses: actions/checkout@v2
+
+    - name: Restore Cache
+      uses: actions/cache@v2
+      with:
+        path: ~/go/pkg/mod
+        key: ${{ runner.os }}-go-${{ hashFiles('**/go.sum') }}
+        restore-keys: |
+          ${{ runner.os }}-go-
+
+    - name: Test with -race flag
+      # Don't use -bench=. -benchtime=1x.
+      # Somewhere in the layers (powershell?)
+      # the equals signs cause great confusion.
+      run: go test -race -bench . -benchtime 1x ./...
+
+    - uses: k0kubun/action-slack@v2.0.0
+      with:
+        payload: |
+          {
+            "attachments": [{
+              "text": "${{ 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/windows.yml

@@ -33,7 +33,10 @@ jobs:
           ${{ runner.os }}-go-
 
     - name: Test
-      run: go test ./...
+      # Don't use -bench=. -benchtime=1x.
+      # Somewhere in the layers (powershell?)
+      # the equals signs cause great confusion.
+      run: go test -bench . -benchtime 1x ./...
 
     - uses: k0kubun/action-slack@v2.0.0
       with:

VERSION.txt

@@ -1 +1 @@
-1.7.0
+1.9.0

cmd/hello/hello.go

@@ -112,13 +112,13 @@ func tailscaleIP(who *apitype.WhoIsResponse) string {
 		return ""
 	}
 	for _, nodeIP := range who.Node.Addresses {
-		if nodeIP.IP.Is4() && nodeIP.IsSingleIP() {
-			return nodeIP.IP.String()
+		if nodeIP.IP().Is4() && nodeIP.IsSingleIP() {
+			return nodeIP.IP().String()
 		}
 	}
 	for _, nodeIP := range who.Node.Addresses {
 		if nodeIP.IsSingleIP() {
-			return nodeIP.IP.String()
+			return nodeIP.IP().String()
 		}
 	}
 	return ""

cmd/tailscale/cli/cli.go

@@ -8,13 +8,16 @@ package cli
 
 import (
 	"context"
+	"errors"
 	"flag"
 	"fmt"
+	"io"
 	"log"
 	"net"
 	"os"
 	"os/signal"
 	"runtime"
+	"strconv"
 	"strings"
 	"syscall"
 	"text/tabwriter"
@@ -31,15 +34,45 @@ import (
 // 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 {
+	// This function is only used on macOS.
+	if runtime.GOOS != "darwin" {
 		return false
 	}
-	switch os.Args[1] {
-	case "up", "down", "status", "netcheck", "ping", "version",
-		"debug",
-		"-V", "--version", "-h", "--help":
+
+	// Escape hatch to let people force running the macOS
+	// GUI Tailscale binary as the CLI.
+	if v, _ := strconv.ParseBool(os.Getenv("TAILSCALE_BE_CLI")); v {
 		return true
 	}
+
+	// If our parent is launchd, we're definitely not
+	// being run as a CLI.
+	if os.Getppid() == 1 {
+		return false
+	}
+
+	// Xcode adds the -NSDocumentRevisionsDebugMode flag on execution.
+	// If present, we are almost certainly being run as a GUI.
+	for _, arg := range os.Args {
+		if arg == "-NSDocumentRevisionsDebugMode" {
+			return false
+		}
+	}
+
+	// Looking at the environment of the GUI Tailscale app (ps eww
+	// $PID), empirically none of these environment variables are
+	// present. But all or some of these should be present with
+	// Terminal.all and bash or zsh.
+	for _, e := range []string{
+		"SHLVL",
+		"TERM",
+		"TERM_PROGRAM",
+		"PS1",
+	} {
+		if os.Getenv(e) != "" {
+			return true
+		}
+	}
 	return false
 }
 
@@ -72,7 +105,7 @@ change in the future.
 			pingCmd,
 			versionCmd,
 			webCmd,
-			pushCmd,
+			fileCmd,
 			bugReportCmd,
 		},
 		FlagSet:   rootfs,
@@ -146,21 +179,22 @@ func connect(ctx context.Context) (net.Conn, *ipn.BackendClient, context.Context
 }
 
 // pump receives backend messages on conn and pushes them into bc.
-func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) {
+func pump(ctx context.Context, bc *ipn.BackendClient, conn net.Conn) error {
 	defer conn.Close()
 	for ctx.Err() == nil {
 		msg, err := ipn.ReadMsg(conn)
 		if err != nil {
 			if ctx.Err() != nil {
-				return
+				return ctx.Err()
 			}
-			if !gotSignal.Get() {
-				log.Printf("ReadMsg: %v\n", err)
+			if errors.Is(err, io.EOF) || errors.Is(err, net.ErrClosed) {
+				return fmt.Errorf("%w (tailscaled stopped running?)", err)
 			}
-			break
+			return err
 		}
 		bc.GotNotifyMsg(msg)
 	}
+	return ctx.Err()
 }
 
 func strSliceContains(ss []string, s string) bool {

cmd/tailscale/cli/cli_test.go

@@ -5,113 +5,664 @@
 package cli
 
 import (
+	"bytes"
+	"encoding/json"
 	"flag"
+	"fmt"
+	"reflect"
+	"strings"
 	"testing"
 
+	"inet.af/netaddr"
 	"tailscale.com/ipn"
+	"tailscale.com/ipn/ipnstate"
+	"tailscale.com/types/preftype"
 )
 
+// geese is a collection of gooses. It need not be complete.
+// But it should include anything handled specially (e.g. linux, windows)
+// and at least one thing that's not (darwin, freebsd).
+var geese = []string{"linux", "darwin", "windows", "freebsd"}
+
 // Test that checkForAccidentalSettingReverts's updateMaskedPrefsFromUpFlag can handle
 // all flags. This will panic if a new flag creeps in that's unhandled.
+//
+// Also, issue 1880: advertise-exit-node was being ignored. Verify that all flags cause an edit.
 func TestUpdateMaskedPrefsFromUpFlag(t *testing.T) {
-	mp := new(ipn.MaskedPrefs)
-	upFlagSet.VisitAll(func(f *flag.Flag) {
-		updateMaskedPrefsFromUpFlag(mp, f.Name)
-	})
+	for _, goos := range geese {
+		var upArgs upArgsT
+		fs := newUpFlagSet(goos, &upArgs)
+		fs.VisitAll(func(f *flag.Flag) {
+			mp := new(ipn.MaskedPrefs)
+			updateMaskedPrefsFromUpFlag(mp, f.Name)
+			got := mp.Pretty()
+			wantEmpty := preflessFlag(f.Name)
+			isEmpty := got == "MaskedPrefs{}"
+			if isEmpty != wantEmpty {
+				t.Errorf("flag %q created MaskedPrefs %s; want empty=%v", f.Name, got, wantEmpty)
+			}
+		})
+	}
 }
 
 func TestCheckForAccidentalSettingReverts(t *testing.T) {
-	f := func(flags ...string) map[string]bool {
-		m := make(map[string]bool)
-		for _, f := range flags {
-			m[f] = true
-		}
-		return m
-	}
 	tests := []struct {
 		name     string
-		flagSet  map[string]bool
+		flags    []string // argv to be parsed by FlagSet
 		curPrefs *ipn.Prefs
-		mp       *ipn.MaskedPrefs
-		want     string
+
+		curExitNodeIP netaddr.IP
+		curUser       string // os.Getenv("USER") on the client side
+		goos          string // empty means "linux"
+
+		want string
 	}{
 		{
-			name:    "bare_up_means_up",
-			flagSet: f(),
+			name:  "bare_up_means_up",
+			flags: []string{},
 			curPrefs: &ipn.Prefs{
+				ControlURL:  ipn.DefaultControlURL,
 				WantRunning: false,
 				Hostname:    "foo",
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					WantRunning: true,
-				},
-				WantRunningSet: true,
-			},
 			want: "",
 		},
 		{
-			name:    "losing_hostname",
-			flagSet: f("accept-dns"),
+			name:  "losing_hostname",
+			flags: []string{"--accept-dns"},
 			curPrefs: &ipn.Prefs{
-				WantRunning: false,
-				Hostname:    "foo",
-				CorpDNS:     true,
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      false,
+				Hostname:         "foo",
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					WantRunning: true,
-					CorpDNS:     true,
-				},
-				WantRunningSet: true,
-				CorpDNSSet:     true,
-			},
-			want: `'tailscale up' without --reset requires all preferences with changing values to be explicitly mentioned; --hostname is not specified but its default value of "" differs from current value "foo"`,
+			want: accidentalUpPrefix + " --accept-dns --hostname=foo",
 		},
 		{
-			name:    "hostname_changing_explicitly",
-			flagSet: f("hostname"),
+			name:  "hostname_changing_explicitly",
+			flags: []string{"--hostname=bar"},
 			curPrefs: &ipn.Prefs{
-				WantRunning: false,
-				Hostname:    "foo",
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					WantRunning: true,
-					Hostname:    "bar",
-				},
-				WantRunningSet: true,
-				HostnameSet:    true,
+				ControlURL:       ipn.DefaultControlURL,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+				Hostname:         "foo",
 			},
 			want: "",
 		},
 		{
-			name:    "hostname_changing_empty_explicitly",
-			flagSet: f("hostname"),
+			name:  "hostname_changing_empty_explicitly",
+			flags: []string{"--hostname="},
 			curPrefs: &ipn.Prefs{
-				WantRunning: false,
-				Hostname:    "foo",
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					WantRunning: true,
-					Hostname:    "",
-				},
-				WantRunningSet: true,
-				HostnameSet:    true,
+				ControlURL:       ipn.DefaultControlURL,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+				Hostname:         "foo",
 			},
 			want: "",
 		},
+		{
+			// Issue 1725: "tailscale up --authkey=..." (or other non-empty flags) works from
+			// a fresh server's initial prefs.
+			name:     "up_with_default_prefs",
+			flags:    []string{"--authkey=foosdlkfjskdljf"},
+			curPrefs: ipn.NewPrefs(),
+			want:     "",
+		},
+		{
+			name:  "implicit_operator_change",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				OperatorUser:     "alice",
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --hostname=foo --operator=alice",
+		},
+		{
+			name:  "implicit_operator_matches_shell_user",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				OperatorUser:     "alice",
+			},
+			curUser: "alice",
+			want:    "",
+		},
+		{
+			name:  "error_advertised_routes_exit_node_removed",
+			flags: []string{"--advertise-routes=10.0.42.0/24"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-routes=10.0.42.0/24 --advertise-exit-node",
+		},
+		{
+			name:  "advertised_routes_exit_node_removed_explicit",
+			flags: []string{"--advertise-routes=10.0.42.0/24", "--advertise-exit-node=false"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: "",
+		},
+		{
+			name:  "advertised_routes_includes_the_0_routes", // but no --advertise-exit-node
+			flags: []string{"--advertise-routes=11.1.43.0/24,0.0.0.0/0,::/0"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.42.0/24"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+			},
+			want: "",
+		},
+		{
+			name:  "advertise_exit_node", // Issue 1859
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: "",
+		},
+		{
+			name:  "advertise_exit_node_over_existing_routes",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "advertise_exit_node_over_existing_routes_and_exit_node",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "exit_node_clearing", // Issue 1777
+			flags: []string{"--exit-node="},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeID: "fooID",
+			},
+			want: "",
+		},
+		{
+			name:  "remove_all_implicit",
+			flags: []string{"--force-reauth"},
+			curPrefs: &ipn.Prefs{
+				WantRunning:      true,
+				ControlURL:       ipn.DefaultControlURL,
+				RouteAll:         true,
+				AllowSingleHosts: false,
+				ExitNodeIP:       netaddr.MustParseIP("100.64.5.6"),
+				CorpDNS:          false,
+				ShieldsUp:        true,
+				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
+				Hostname:         "myhostname",
+				ForceDaemon:      true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.0.0/16"),
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+				NetfilterMode: preftype.NetfilterNoDivert,
+				OperatorUser:  "alice",
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --force-reauth --accept-dns=false --accept-routes --advertise-exit-node --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --hostname=myhostname --netfilter-mode=nodivert --operator=alice --shields-up",
+		},
+		{
+			name:  "remove_all_implicit_except_hostname",
+			flags: []string{"--hostname=newhostname"},
+			curPrefs: &ipn.Prefs{
+				WantRunning:      true,
+				ControlURL:       ipn.DefaultControlURL,
+				RouteAll:         true,
+				AllowSingleHosts: false,
+				ExitNodeIP:       netaddr.MustParseIP("100.64.5.6"),
+				CorpDNS:          false,
+				ShieldsUp:        true,
+				AdvertiseTags:    []string{"tag:foo", "tag:bar"},
+				Hostname:         "myhostname",
+				ForceDaemon:      true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("10.0.0.0/16"),
+				},
+				NetfilterMode: preftype.NetfilterNoDivert,
+				OperatorUser:  "alice",
+			},
+			curUser: "eve",
+			want:    accidentalUpPrefix + " --hostname=newhostname --accept-dns=false --accept-routes --advertise-routes=10.0.0.0/16 --advertise-tags=tag:foo,tag:bar --exit-node=100.64.5.6 --host-routes=false --netfilter-mode=nodivert --operator=alice --shields-up",
+		},
+		{
+			name:  "loggedout_is_implicit",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				LoggedOut:        true,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			want: "", // not an error. LoggedOut is implicit.
+		},
+		{
+			// Test that a pre-1.8 version of Tailscale with bogus NoSNAT pref
+			// values is able to enable exit nodes without warnings.
+			name:  "make_windows_exit_node",
+			flags: []string{"--advertise-exit-node"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+
+				// And assume this no-op accidental pre-1.8 value:
+				NoSNAT: true,
+			},
+			goos: "windows",
+			want: "", // not an error
+		},
+		{
+			name:  "ignore_netfilter_change_non_linux",
+			flags: []string{"--accept-dns"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+
+				NetfilterMode: preftype.NetfilterNoDivert, // we never had this bug, but pretend it got set non-zero on Windows somehow
+			},
+			goos: "windows",
+			want: "", // not an error
+		},
+		{
+			name:  "operator_losing_routes_step1", // https://twitter.com/EXPbits/status/1390418145047887877
+			flags: []string{"--operator=expbits"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --operator=expbits --advertise-exit-node --advertise-routes=1.2.0.0/16",
+		},
+		{
+			name:  "operator_losing_routes_step2", // https://twitter.com/EXPbits/status/1390418145047887877
+			flags: []string{"--operator=expbits", "--advertise-routes=1.2.0.0/16"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+					netaddr.MustParseIPPrefix("1.2.0.0/16"),
+				},
+			},
+			want: accidentalUpPrefix + " --advertise-routes=1.2.0.0/16 --operator=expbits --advertise-exit-node",
+		},
+		{
+			name:  "errors_preserve_explicit_flags",
+			flags: []string{"--reset", "--force-reauth=false", "--authkey=secretrand"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      false,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+
+				Hostname: "foo",
+			},
+			want: accidentalUpPrefix + " --authkey=secretrand --force-reauth=false --reset --hostname=foo",
+		},
+		{
+			name:  "error_exit_node_omit_with_ip_pref",
+			flags: []string{"--hostname=foo"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeIP: netaddr.MustParseIP("100.64.5.4"),
+			},
+			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.4",
+		},
+		{
+			name:          "error_exit_node_omit_with_id_pref",
+			flags:         []string{"--hostname=foo"},
+			curExitNodeIP: netaddr.MustParseIP("100.64.5.7"),
+			curPrefs: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
+				ExitNodeID: "some_stable_id",
+			},
+			want: accidentalUpPrefix + " --hostname=foo --exit-node=100.64.5.7",
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
+			goos := "linux"
+			if tt.goos != "" {
+				goos = tt.goos
+			}
+			var upArgs upArgsT
+			flagSet := newUpFlagSet(goos, &upArgs)
+			flagSet.Parse(tt.flags)
+			newPrefs, err := prefsFromUpArgs(upArgs, t.Logf, new(ipnstate.Status), goos)
+			if err != nil {
+				t.Fatal(err)
+			}
+			applyImplicitPrefs(newPrefs, tt.curPrefs, tt.curUser)
 			var got string
-			if err := checkForAccidentalSettingReverts(tt.flagSet, tt.curPrefs, tt.mp); err != nil {
+			if err := checkForAccidentalSettingReverts(flagSet, tt.curPrefs, newPrefs, upCheckEnv{
+				goos:          goos,
+				curExitNodeIP: tt.curExitNodeIP,
+			}); err != nil {
 				got = err.Error()
 			}
-			if got != tt.want {
+			if strings.TrimSpace(got) != tt.want {
 				t.Errorf("unexpected result\n got: %s\nwant: %s\n", got, tt.want)
 			}
 		})
 	}
 }
+
+func upArgsFromOSArgs(goos string, flagArgs ...string) (args upArgsT) {
+	fs := newUpFlagSet(goos, &args)
+	fs.Parse(flagArgs) // populates args
+	return
+}
+
+func TestPrefsFromUpArgs(t *testing.T) {
+	tests := []struct {
+		name     string
+		args     upArgsT
+		goos     string           // runtime.GOOS; empty means linux
+		st       *ipnstate.Status // or nil
+		want     *ipn.Prefs
+		wantErr  string
+		wantWarn string
+	}{
+		{
+			name: "default_linux",
+			goos: "linux",
+			args: upArgsFromOSArgs("linux"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				NoSNAT:           false,
+				NetfilterMode:    preftype.NetfilterOn,
+				CorpDNS:          true,
+				AllowSingleHosts: true,
+			},
+		},
+		{
+			name: "default_windows",
+			goos: "windows",
+			args: upArgsFromOSArgs("windows"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				CorpDNS:          true,
+				AllowSingleHosts: true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+		},
+		{
+			name: "advertise_default_route",
+			args: upArgsFromOSArgs("linux", "--advertise-exit-node"),
+			want: &ipn.Prefs{
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      true,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				AdvertiseRoutes: []netaddr.IPPrefix{
+					netaddr.MustParseIPPrefix("0.0.0.0/0"),
+					netaddr.MustParseIPPrefix("::/0"),
+				},
+				NetfilterMode: preftype.NetfilterOn,
+			},
+		},
+		{
+			name: "error_advertise_route_invalid_ip",
+			args: upArgsT{
+				advertiseRoutes: "foo",
+			},
+			wantErr: `"foo" is not a valid IP address or CIDR prefix`,
+		},
+		{
+			name: "error_advertise_route_unmasked_bits",
+			args: upArgsT{
+				advertiseRoutes: "1.2.3.4/16",
+			},
+			wantErr: `1.2.3.4/16 has non-address bits set; expected 1.2.0.0/16`,
+		},
+		{
+			name: "error_exit_node_bad_ip",
+			args: upArgsT{
+				exitNodeIP: "foo",
+			},
+			wantErr: `invalid IP address "foo" for --exit-node: ParseIP("foo"): unable to parse IP`,
+		},
+		{
+			name: "error_exit_node_allow_lan_without_exit_node",
+			args: upArgsT{
+				exitNodeAllowLANAccess: true,
+			},
+			wantErr: `--exit-node-allow-lan-access can only be used with --exit-node`,
+		},
+		{
+			name: "error_tag_prefix",
+			args: upArgsT{
+				advertiseTags: "foo",
+			},
+			wantErr: `tag: "foo": tags must start with 'tag:'`,
+		},
+		{
+			name: "error_long_hostname",
+			args: upArgsT{
+				hostname: strings.Repeat("a", 300),
+			},
+			wantErr: `hostname too long: 300 bytes (max 256)`,
+		},
+		{
+			name: "error_linux_netfilter_empty",
+			args: upArgsT{
+				netfilterMode: "",
+			},
+			wantErr: `invalid value --netfilter-mode=""`,
+		},
+		{
+			name: "error_linux_netfilter_bogus",
+			args: upArgsT{
+				netfilterMode: "bogus",
+			},
+			wantErr: `invalid value --netfilter-mode="bogus"`,
+		},
+		{
+			name: "error_exit_node_ip_is_self_ip",
+			args: upArgsT{
+				exitNodeIP: "100.105.106.107",
+			},
+			st: &ipnstate.Status{
+				TailscaleIPs: []netaddr.IP{netaddr.MustParseIP("100.105.106.107")},
+			},
+			wantErr: `cannot use 100.105.106.107 as the exit node as it is a local IP address to this machine, did you mean --advertise-exit-node?`,
+		},
+		{
+			name: "warn_linux_netfilter_nodivert",
+			goos: "linux",
+			args: upArgsT{
+				netfilterMode: "nodivert",
+			},
+			wantWarn: "netfilter=nodivert; add iptables calls to ts-* chains manually.",
+			want: &ipn.Prefs{
+				WantRunning:   true,
+				NetfilterMode: preftype.NetfilterNoDivert,
+				NoSNAT:        true,
+			},
+		},
+		{
+			name: "warn_linux_netfilter_off",
+			goos: "linux",
+			args: upArgsT{
+				netfilterMode: "off",
+			},
+			wantWarn: "netfilter=off; configure iptables yourself.",
+			want: &ipn.Prefs{
+				WantRunning:   true,
+				NetfilterMode: preftype.NetfilterOff,
+				NoSNAT:        true,
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var warnBuf bytes.Buffer
+			warnf := func(format string, a ...interface{}) {
+				fmt.Fprintf(&warnBuf, format, a...)
+			}
+			goos := tt.goos
+			if goos == "" {
+				goos = "linux"
+			}
+			st := tt.st
+			if st == nil {
+				st = new(ipnstate.Status)
+			}
+			got, err := prefsFromUpArgs(tt.args, warnf, st, goos)
+			gotErr := fmt.Sprint(err)
+			if tt.wantErr != "" {
+				if tt.wantErr != gotErr {
+					t.Errorf("wrong error.\n got error: %v\nwant error: %v\n", gotErr, tt.wantErr)
+				}
+				return
+			}
+			if err != nil {
+				t.Fatal(err)
+			}
+			if tt.want == nil {
+				t.Fatal("tt.want is nil")
+			}
+			if !got.Equals(tt.want) {
+				jgot, _ := json.MarshalIndent(got, "", "\t")
+				jwant, _ := json.MarshalIndent(tt.want, "", "\t")
+				if bytes.Equal(jgot, jwant) {
+					t.Logf("prefs differ only in non-JSON-visible ways (nil/non-nil zero-length arrays)")
+				}
+				t.Errorf("wrong prefs\n got: %s\nwant: %s\n\ngot: %s\nwant: %s\n",
+					got.Pretty(), tt.want.Pretty(),
+					jgot, jwant,
+				)
+
+			}
+		})
+	}
+
+}
+
+func TestPrefFlagMapping(t *testing.T) {
+	prefHasFlag := map[string]bool{}
+	for _, pv := range prefsOfFlag {
+		for _, pref := range pv {
+			prefHasFlag[pref] = true
+		}
+	}
+
+	prefType := reflect.TypeOf(ipn.Prefs{})
+	for i := 0; i < prefType.NumField(); i++ {
+		prefName := prefType.Field(i).Name
+		if prefHasFlag[prefName] {
+			continue
+		}
+		switch prefName {
+		case "WantRunning", "Persist", "LoggedOut":
+			// All explicitly handled (ignored) by checkForAccidentalSettingReverts.
+			continue
+		case "OSVersion", "DeviceModel":
+			// Only used by Android, which doesn't have a CLI mode anyway, so
+			// fine to not map.
+			continue
+		case "NotepadURLs":
+			// TODO(bradfitz): https://github.com/tailscale/tailscale/issues/1830
+			continue
+		}
+		t.Errorf("unexpected new ipn.Pref field %q is not handled by up.go (see addPrefFlagMapping and checkForAccidentalSettingReverts)", prefName)
+	}
+}
+
+func TestFlagAppliesToOS(t *testing.T) {
+	for _, goos := range geese {
+		var upArgs upArgsT
+		fs := newUpFlagSet(goos, &upArgs)
+		fs.VisitAll(func(f *flag.Flag) {
+			if !flagAppliesToOS(f.Name, goos) {
+				t.Errorf("flagAppliesToOS(%q, %q) = false but found in %s set", f.Name, goos, goos)
+			}
+		})
+	}
+}

cmd/tailscale/cli/file.go

@@ -0,0 +1,444 @@
+// Copyright (c) 2021 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"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"log"
+	"mime"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"golang.org/x/time/rate"
+	"inet.af/netaddr"
+	"tailscale.com/client/tailscale"
+	"tailscale.com/client/tailscale/apitype"
+	"tailscale.com/ipn"
+	"tailscale.com/net/tsaddr"
+	"tailscale.com/version"
+)
+
+var fileCmd = &ffcli.Command{
+	Name:       "file",
+	ShortUsage: "file <cp|get> ...",
+	ShortHelp:  "Send or receive files",
+	Subcommands: []*ffcli.Command{
+		fileCpCmd,
+		fileGetCmd,
+	},
+	Exec: func(context.Context, []string) error {
+		// TODO(bradfitz): is there a better ffcli way to
+		// annotate subcommand-required commands that don't
+		// have an exec body of their own?
+		return errors.New("file subcommand required; run 'tailscale file -h' for details")
+	},
+}
+
+var fileCpCmd = &ffcli.Command{
+	Name:       "cp",
+	ShortUsage: "file cp <files...> <target>:",
+	ShortHelp:  "Copy file(s) to a host",
+	Exec:       runCp,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("cp", flag.ExitOnError)
+		fs.StringVar(&cpArgs.name, "name", "", "alternate filename to use, especially useful when <file> is \"-\" (stdin)")
+		fs.BoolVar(&cpArgs.verbose, "verbose", false, "verbose output")
+		fs.BoolVar(&cpArgs.targets, "targets", false, "list possible file cp targets")
+		return fs
+	})(),
+}
+
+var cpArgs struct {
+	name    string
+	verbose bool
+	targets bool
+}
+
+func runCp(ctx context.Context, args []string) error {
+	if cpArgs.targets {
+		return runCpTargets(ctx, args)
+	}
+	if len(args) < 2 {
+		//lint:ignore ST1005 no sorry need that colon at the end
+		return errors.New("usage: tailscale file cp <files...> <target>:")
+	}
+	files, target := args[:len(args)-1], args[len(args)-1]
+	if !strings.HasSuffix(target, ":") {
+		return fmt.Errorf("final argument to 'tailscale file cp' must end in colon")
+	}
+	target = strings.TrimSuffix(target, ":")
+	hadBrackets := false
+	if strings.HasPrefix(target, "[") && strings.HasSuffix(target, "]") {
+		hadBrackets = true
+		target = strings.TrimSuffix(strings.TrimPrefix(target, "["), "]")
+	}
+	if ip, err := netaddr.ParseIP(target); err == nil && ip.Is6() && !hadBrackets {
+		return fmt.Errorf("an IPv6 literal must be written as [%s]", ip)
+	} else if hadBrackets && (err != nil || !ip.Is6()) {
+		return errors.New("unexpected brackets around target")
+	}
+	ip, err := tailscaleIPFromArg(ctx, target)
+	if err != nil {
+		return err
+	}
+
+	peerAPIBase, lastSeen, isOffline, err := discoverPeerAPIBase(ctx, ip)
+	if err != nil {
+		return fmt.Errorf("can't send to %s: %v", target, err)
+	}
+	if isOffline {
+		fmt.Fprintf(os.Stderr, "# warning: %s is offline\n", target)
+	} else if !lastSeen.IsZero() && time.Since(lastSeen) > lastSeenOld {
+		fmt.Fprintf(os.Stderr, "# warning: %s last seen %v ago\n", target, time.Since(lastSeen).Round(time.Minute))
+	}
+
+	if len(files) > 1 {
+		if cpArgs.name != "" {
+			return errors.New("can't use --name= with multiple files")
+		}
+		for _, fileArg := range files {
+			if fileArg == "-" {
+				return errors.New("can't use '-' as STDIN file when providing filename arguments")
+			}
+		}
+	}
+
+	for _, fileArg := range files {
+		var fileContents io.Reader
+		var name = cpArgs.name
+		var contentLength int64 = -1
+		if fileArg == "-" {
+			fileContents = os.Stdin
+			if name == "" {
+				name, fileContents, err = pickStdinFilename()
+				if err != nil {
+					return err
+				}
+			}
+		} else {
+			f, err := os.Open(fileArg)
+			if err != nil {
+				if version.IsSandboxedMacOS() {
+					return errors.New("the GUI version of Tailscale on macOS runs in a macOS sandbox that can't read files")
+				}
+				return err
+			}
+			defer f.Close()
+			fi, err := f.Stat()
+			if err != nil {
+				return err
+			}
+			if fi.IsDir() {
+				return errors.New("directories not supported")
+			}
+			contentLength = fi.Size()
+			fileContents = io.LimitReader(f, contentLength)
+			if name == "" {
+				name = filepath.Base(fileArg)
+			}
+
+			if slow, _ := strconv.ParseBool(os.Getenv("TS_DEBUG_SLOW_PUSH")); slow {
+				fileContents = &slowReader{r: fileContents}
+			}
+		}
+
+		dstURL := peerAPIBase + "/v0/put/" + url.PathEscape(name)
+		req, err := http.NewRequestWithContext(ctx, "PUT", dstURL, fileContents)
+		if err != nil {
+			return err
+		}
+		req.ContentLength = contentLength
+		if cpArgs.verbose {
+			log.Printf("sending to %v ...", dstURL)
+		}
+		res, err := http.DefaultClient.Do(req)
+		if err != nil {
+			return err
+		}
+		if res.StatusCode == 200 {
+			io.Copy(ioutil.Discard, res.Body)
+			res.Body.Close()
+			continue
+		}
+		io.Copy(os.Stdout, res.Body)
+		res.Body.Close()
+		return errors.New(res.Status)
+	}
+	return nil
+}
+
+func discoverPeerAPIBase(ctx context.Context, ipStr string) (base string, lastSeen time.Time, isOffline bool, err error) {
+	ip, err := netaddr.ParseIP(ipStr)
+	if err != nil {
+		return "", time.Time{}, false, err
+	}
+	fts, err := tailscale.FileTargets(ctx)
+	if err != nil {
+		return "", time.Time{}, false, err
+	}
+	for _, ft := range fts {
+		n := ft.Node
+		for _, a := range n.Addresses {
+			if a.IP() != ip {
+				continue
+			}
+			if n.LastSeen != nil {
+				lastSeen = *n.LastSeen
+			}
+			isOffline = n.Online != nil && !*n.Online
+			return ft.PeerAPIURL, lastSeen, isOffline, nil
+		}
+	}
+	return "", time.Time{}, false, fileTargetErrorDetail(ctx, ip)
+}
+
+// fileTargetErrorDetail returns a non-nil error saying why ip is an
+// invalid file sharing target.
+func fileTargetErrorDetail(ctx context.Context, ip netaddr.IP) error {
+	found := false
+	if st, err := tailscale.Status(ctx); err == nil && st.Self != nil {
+		for _, peer := range st.Peer {
+			for _, pip := range peer.TailscaleIPs {
+				if pip == ip {
+					found = true
+					if peer.UserID != st.Self.UserID {
+						return errors.New("owned by different user; can only send files to your own devices")
+					}
+				}
+			}
+		}
+	}
+	if found {
+		return errors.New("target seems to be running an old Tailscale version")
+	}
+	if !tsaddr.IsTailscaleIP(ip) {
+		return fmt.Errorf("unknown target; %v is not a Tailscale IP address", ip)
+	}
+	return errors.New("unknown target; not in your Tailnet")
+}
+
+const maxSniff = 4 << 20
+
+func ext(b []byte) string {
+	if len(b) < maxSniff && utf8.Valid(b) {
+		return ".txt"
+	}
+	if exts, _ := mime.ExtensionsByType(http.DetectContentType(b)); len(exts) > 0 {
+		return exts[0]
+	}
+	return ""
+}
+
+// pickStdinFilename reads a bit of stdin to return a good filename
+// for its contents. The returned Reader is the concatenation of the
+// read and unread bits.
+func pickStdinFilename() (name string, r io.Reader, err error) {
+	sniff, err := io.ReadAll(io.LimitReader(os.Stdin, maxSniff))
+	if err != nil {
+		return "", nil, err
+	}
+	return "stdin" + ext(sniff), io.MultiReader(bytes.NewReader(sniff), os.Stdin), nil
+}
+
+type slowReader struct {
+	r  io.Reader
+	rl *rate.Limiter
+}
+
+func (r *slowReader) Read(p []byte) (n int, err error) {
+	const burst = 4 << 10
+	plen := len(p)
+	if plen > burst {
+		plen = burst
+	}
+	if r.rl == nil {
+		r.rl = rate.NewLimiter(rate.Limit(1<<10), burst)
+	}
+	n, err = r.r.Read(p[:plen])
+	r.rl.WaitN(context.Background(), n)
+	return
+}
+
+const lastSeenOld = 20 * time.Minute
+
+func runCpTargets(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		return errors.New("invalid arguments with --targets")
+	}
+	fts, err := tailscale.FileTargets(ctx)
+	if err != nil {
+		return err
+	}
+	for _, ft := range fts {
+		n := ft.Node
+		var detail string
+		if n.Online != nil {
+			if !*n.Online {
+				detail = "offline"
+			}
+		} else {
+			detail = "unknown-status"
+		}
+		if detail != "" && n.LastSeen != nil {
+			d := time.Since(*n.LastSeen)
+			detail += fmt.Sprintf("; last seen %v ago", d.Round(time.Minute))
+		}
+		if detail != "" {
+			detail = "\t" + detail
+		}
+		fmt.Printf("%s\t%s%s\n", n.Addresses[0].IP(), n.ComputedName, detail)
+	}
+	return nil
+}
+
+var fileGetCmd = &ffcli.Command{
+	Name:       "get",
+	ShortUsage: "file get [--wait] [--verbose] <target-directory>",
+	ShortHelp:  "Move files out of the Tailscale file inbox",
+	Exec:       runFileGet,
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("get", flag.ExitOnError)
+		fs.BoolVar(&getArgs.wait, "wait", false, "wait for a file to arrive if inbox is empty")
+		fs.BoolVar(&getArgs.verbose, "verbose", false, "verbose output")
+		return fs
+	})(),
+}
+
+var getArgs struct {
+	wait    bool
+	verbose bool
+}
+
+func runFileGet(ctx context.Context, args []string) error {
+	if len(args) != 1 {
+		return errors.New("usage: file get <target-directory>")
+	}
+	log.SetFlags(0)
+
+	dir := args[0]
+	if dir == "/dev/null" {
+		return wipeInbox(ctx)
+	}
+
+	if fi, err := os.Stat(dir); err != nil || !fi.IsDir() {
+		return fmt.Errorf("%q is not a directory", dir)
+	}
+
+	var wfs []apitype.WaitingFile
+	var err error
+	for {
+		wfs, err = tailscale.WaitingFiles(ctx)
+		if err != nil {
+			return fmt.Errorf("getting WaitingFiles: %v", err)
+		}
+		if len(wfs) != 0 || !getArgs.wait {
+			break
+		}
+		if getArgs.verbose {
+			log.Printf("waiting for file...")
+		}
+		if err := waitForFile(ctx); err != nil {
+			return err
+		}
+	}
+
+	deleted := 0
+	for _, wf := range wfs {
+		rc, size, err := tailscale.GetWaitingFile(ctx, wf.Name)
+		if err != nil {
+			return fmt.Errorf("opening inbox file %q: %v", wf.Name, err)
+		}
+		targetFile := filepath.Join(dir, wf.Name)
+		of, err := os.OpenFile(targetFile, os.O_RDWR|os.O_CREATE|os.O_EXCL, 0644)
+		if err != nil {
+			if _, err := os.Stat(targetFile); err == nil {
+				return fmt.Errorf("refusing to overwrite %v", targetFile)
+			}
+			return err
+		}
+		_, err = io.Copy(of, rc)
+		rc.Close()
+		if err != nil {
+			return fmt.Errorf("failed to write %v: %v", targetFile, err)
+		}
+		if err := of.Close(); err != nil {
+			return err
+		}
+		if getArgs.verbose {
+			log.Printf("wrote %v (%d bytes)", wf.Name, size)
+		}
+		if err := tailscale.DeleteWaitingFile(ctx, wf.Name); err != nil {
+			return fmt.Errorf("deleting %q from inbox: %v", wf.Name, err)
+		}
+		deleted++
+	}
+	if getArgs.verbose {
+		log.Printf("moved %d files", deleted)
+	}
+	return nil
+}
+
+func wipeInbox(ctx context.Context) error {
+	if getArgs.wait {
+		return errors.New("can't use --wait with /dev/null target")
+	}
+	wfs, err := tailscale.WaitingFiles(ctx)
+	if err != nil {
+		return fmt.Errorf("getting WaitingFiles: %v", err)
+	}
+	deleted := 0
+	for _, wf := range wfs {
+		if getArgs.verbose {
+			log.Printf("deleting %v ...", wf.Name)
+		}
+		if err := tailscale.DeleteWaitingFile(ctx, wf.Name); err != nil {
+			return fmt.Errorf("deleting %q: %v", wf.Name, err)
+		}
+		deleted++
+	}
+	if getArgs.verbose {
+		log.Printf("deleted %d files", deleted)
+	}
+	return nil
+}
+
+func waitForFile(ctx context.Context) error {
+	c, bc, pumpCtx, cancel := connect(ctx)
+	defer cancel()
+	fileWaiting := make(chan bool, 1)
+	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.ErrMessage != nil {
+			log.Fatal(*n.ErrMessage)
+		}
+		if n.FilesWaiting != nil {
+			select {
+			case fileWaiting <- true:
+			default:
+			}
+		}
+	})
+	go pump(pumpCtx, bc, c)
+	select {
+	case <-fileWaiting:
+		return nil
+	case <-pumpCtx.Done():
+		return pumpCtx.Err()
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

cmd/tailscale/cli/ping.go

@@ -64,6 +64,7 @@ var pingArgs struct {
 }
 
 func runPing(ctx context.Context, args []string) error {
+	fmt.Println("runPing")
 	c, bc, ctx, cancel := connect(ctx)
 	defer cancel()
 
@@ -80,7 +81,8 @@ func runPing(ctx context.Context, args []string) error {
 			prc <- pr
 		}
 	})
-	go pump(ctx, bc, c)
+	pumpErr := make(chan error, 1)
+	go func() { pumpErr <- pump(ctx, bc, c) }()
 
 	hostOrIP := args[0]
 	ip, err := tailscaleIPFromArg(ctx, hostOrIP)
@@ -101,6 +103,8 @@ func runPing(ctx context.Context, args []string) error {
 		select {
 		case <-timer.C:
 			fmt.Printf("timeout waiting for ping reply\n")
+		case err := <-pumpErr:
+			return err
 		case pr := <-prc:
 			timer.Stop()
 			if pr.Err != "" {
@@ -136,6 +140,9 @@ func runPing(ctx context.Context, args []string) error {
 			if !anyPong {
 				return errors.New("no reply")
 			}
+			if pingArgs.untilDirect {
+				return errors.New("direct connection not established")
+			}
 			return nil
 		}
 	}

cmd/tailscale/cli/push.go

@@ -1,218 +0,0 @@
-// Copyright (c) 2021 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"
-	"errors"
-	"flag"
-	"fmt"
-	"io"
-	"log"
-	"mime"
-	"net/http"
-	"net/url"
-	"os"
-	"strconv"
-	"time"
-	"unicode/utf8"
-
-	"github.com/peterbourgon/ff/v2/ffcli"
-	"golang.org/x/time/rate"
-	"inet.af/netaddr"
-	"tailscale.com/client/tailscale"
-)
-
-var pushCmd = &ffcli.Command{
-	Name:       "push",
-	ShortUsage: "push [--flags] <hostname-or-IP> <file>",
-	ShortHelp:  "Push a file to a host",
-	Exec:       runPush,
-	FlagSet: (func() *flag.FlagSet {
-		fs := flag.NewFlagSet("push", flag.ExitOnError)
-		fs.StringVar(&pushArgs.name, "name", "", "alternate filename to use, especially useful when <file> is \"-\" (stdin)")
-		fs.BoolVar(&pushArgs.verbose, "verbose", false, "verbose output")
-		fs.BoolVar(&pushArgs.targets, "targets", false, "list possible push targets")
-		return fs
-	})(),
-}
-
-var pushArgs struct {
-	name    string
-	verbose bool
-	targets bool
-}
-
-func runPush(ctx context.Context, args []string) error {
-	if pushArgs.targets {
-		return runPushTargets(ctx, args)
-	}
-	if len(args) != 2 || args[0] == "" {
-		return errors.New("usage: push <hostname-or-IP> <file>\n       push --targets")
-	}
-	var ip string
-
-	hostOrIP, fileArg := args[0], args[1]
-	ip, err := tailscaleIPFromArg(ctx, hostOrIP)
-	if err != nil {
-		return err
-	}
-
-	peerAPIBase, lastSeen, err := discoverPeerAPIBase(ctx, ip)
-	if err != nil {
-		return err
-	}
-
-	if !lastSeen.IsZero() && time.Since(lastSeen) > lastSeenOld {
-		fmt.Fprintf(os.Stderr, "# warning: %s last seen %v ago\n", hostOrIP, time.Since(lastSeen).Round(time.Minute))
-	}
-
-	var fileContents io.Reader
-	var name = pushArgs.name
-	var contentLength int64 = -1
-	if fileArg == "-" {
-		fileContents = os.Stdin
-		if name == "" {
-			name, fileContents, err = pickStdinFilename()
-			if err != nil {
-				return err
-			}
-		}
-	} else {
-		f, err := os.Open(fileArg)
-		if err != nil {
-			return err
-		}
-		defer f.Close()
-		fi, err := f.Stat()
-		if err != nil {
-			return err
-		}
-		if fi.IsDir() {
-			return errors.New("directories not supported")
-		}
-		contentLength = fi.Size()
-		fileContents = io.LimitReader(f, contentLength)
-		if name == "" {
-			name = fileArg
-		}
-
-		if slow, _ := strconv.ParseBool(os.Getenv("TS_DEBUG_SLOW_PUSH")); slow {
-			fileContents = &slowReader{r: fileContents}
-		}
-	}
-
-	dstURL := peerAPIBase + "/v0/put/" + url.PathEscape(name)
-	req, err := http.NewRequestWithContext(ctx, "PUT", dstURL, fileContents)
-	if err != nil {
-		return err
-	}
-	req.ContentLength = contentLength
-	if pushArgs.verbose {
-		log.Printf("sending to %v ...", dstURL)
-	}
-	res, err := http.DefaultClient.Do(req)
-	if err != nil {
-		return err
-	}
-	defer res.Body.Close()
-	if res.StatusCode == 200 {
-		return nil
-	}
-	io.Copy(os.Stdout, res.Body)
-	return errors.New(res.Status)
-}
-
-func discoverPeerAPIBase(ctx context.Context, ipStr string) (base string, lastSeen time.Time, err error) {
-	ip, err := netaddr.ParseIP(ipStr)
-	if err != nil {
-		return "", time.Time{}, err
-	}
-	fts, err := tailscale.FileTargets(ctx)
-	if err != nil {
-		return "", time.Time{}, err
-	}
-	for _, ft := range fts {
-		n := ft.Node
-		for _, a := range n.Addresses {
-			if a.IP != ip {
-				continue
-			}
-			if n.LastSeen != nil {
-				lastSeen = *n.LastSeen
-			}
-			return ft.PeerAPIURL, lastSeen, nil
-		}
-	}
-	return "", time.Time{}, errors.New("target seems to be running an old Tailscale version")
-}
-
-const maxSniff = 4 << 20
-
-func ext(b []byte) string {
-	if len(b) < maxSniff && utf8.Valid(b) {
-		return ".txt"
-	}
-	if exts, _ := mime.ExtensionsByType(http.DetectContentType(b)); len(exts) > 0 {
-		return exts[0]
-	}
-	return ""
-}
-
-// pickStdinFilename reads a bit of stdin to return a good filename
-// for its contents. The returned Reader is the concatenation of the
-// read and unread bits.
-func pickStdinFilename() (name string, r io.Reader, err error) {
-	sniff, err := io.ReadAll(io.LimitReader(os.Stdin, maxSniff))
-	if err != nil {
-		return "", nil, err
-	}
-	return "stdin" + ext(sniff), io.MultiReader(bytes.NewReader(sniff), os.Stdin), nil
-}
-
-type slowReader struct {
-	r  io.Reader
-	rl *rate.Limiter
-}
-
-func (r *slowReader) Read(p []byte) (n int, err error) {
-	const burst = 4 << 10
-	plen := len(p)
-	if plen > burst {
-		plen = burst
-	}
-	if r.rl == nil {
-		r.rl = rate.NewLimiter(rate.Limit(1<<10), burst)
-	}
-	n, err = r.r.Read(p[:plen])
-	r.rl.WaitN(context.Background(), n)
-	return
-}
-
-const lastSeenOld = 20 * time.Minute
-
-func runPushTargets(ctx context.Context, args []string) error {
-	if len(args) > 0 {
-		return errors.New("invalid arguments with --targets")
-	}
-	fts, err := tailscale.FileTargets(ctx)
-	if err != nil {
-		return err
-	}
-	for _, ft := range fts {
-		n := ft.Node
-		var ago string
-		if n.LastSeen == nil {
-			ago = "\tnode never seen"
-		} else {
-			if d := time.Since(*n.LastSeen); d > lastSeenOld {
-				ago = fmt.Sprintf("\tlast seen %v ago", d.Round(time.Minute))
-			}
-		}
-		fmt.Printf("%s\t%s%s\n", n.Addresses[0].IP, n.ComputedName, ago)
-	}
-	return nil
-}

cmd/tailscale/cli/up.go

@@ -13,16 +13,18 @@ import (
 	"reflect"
 	"runtime"
 	"sort"
-	"strconv"
 	"strings"
 	"sync"
 
-	"github.com/go-multierror/multierror"
+	shellquote "github.com/kballard/go-shellquote"
 	"github.com/peterbourgon/ff/v2/ffcli"
 	"inet.af/netaddr"
 	"tailscale.com/client/tailscale"
 	"tailscale.com/ipn"
+	"tailscale.com/ipn/ipnstate"
+	"tailscale.com/safesocket"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/preftype"
 	"tailscale.com/version/distro"
 )
@@ -49,13 +51,15 @@ flag is also used.
 	Exec:    runUp,
 }
 
-var upFlagSet = (func() *flag.FlagSet {
+var upFlagSet = newUpFlagSet(runtime.GOOS, &upArgs)
+
+func newUpFlagSet(goos string, upArgs *upArgsT) *flag.FlagSet {
 	upf := flag.NewFlagSet("up", flag.ExitOnError)
 
 	upf.BoolVar(&upArgs.forceReauth, "force-reauth", false, "force reauthentication")
 	upf.BoolVar(&upArgs.reset, "reset", false, "reset unspecified settings to their default values")
 
-	upf.StringVar(&upArgs.server, "login-server", "https://login.tailscale.com", "base URL of control server")
+	upf.StringVar(&upArgs.server, "login-server", ipn.DefaultControlURL, "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")
@@ -67,15 +71,18 @@ var upFlagSet = (func() *flag.FlagSet {
 	upf.StringVar(&upArgs.hostname, "hostname", "", "hostname to use instead of the one provided by the OS")
 	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\")")
 	upf.BoolVar(&upArgs.advertiseDefaultRoute, "advertise-exit-node", false, "offer to be an exit node for internet traffic for the tailnet")
-	if runtime.GOOS == "linux" {
+	if safesocket.GOOSUsesPeerCreds(goos) {
+		upf.StringVar(&upArgs.opUser, "operator", "", "Unix username to allow to operate on tailscaled without sudo")
+	}
+	switch goos {
+	case "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", defaultNetfilterMode(), "netfilter mode (one of on, nodivert, off)")
-	}
-	if runtime.GOOS == "windows" {
+	case "windows":
 		upf.BoolVar(&upArgs.forceDaemon, "unattended", false, "run in \"Unattended Mode\" where Tailscale keeps running even after the current GUI user logs out (Windows-only)")
 	}
 	return upf
-})()
+}
 
 func defaultNetfilterMode() string {
 	if distro.Get() == distro.Synology {
@@ -84,7 +91,7 @@ func defaultNetfilterMode() string {
 	return "on"
 }
 
-var upArgs struct {
+type upArgsT struct {
 	reset                  bool
 	server                 string
 	acceptRoutes           bool
@@ -102,8 +109,11 @@ var upArgs struct {
 	netfilterMode          string
 	authKey                string
 	hostname               string
+	opUser                 string
 }
 
+var upArgs upArgsT
+
 func warnf(format string, args ...interface{}) {
 	fmt.Printf("Warning: "+format+"\n", args...)
 }
@@ -113,29 +123,13 @@ var (
 	ipv6default = netaddr.MustParseIPPrefix("::/0")
 )
 
-func runUp(ctx context.Context, args []string) error {
-	if len(args) > 0 {
-		fatalf("too many non-flag arguments: %q", args)
-	}
-
-	st, err := tailscale.Status(ctx)
-	if err != nil {
-		fatalf("can't fetch status from tailscaled: %v", err)
-	}
-
-	if distro.Get() == distro.Synology {
-		notSupported := "not yet supported on Synology; see https://github.com/tailscale/tailscale/issues/451"
-		if upArgs.acceptRoutes {
-			return errors.New("--accept-routes is " + notSupported)
-		}
-		if upArgs.exitNodeIP != "" {
-			return errors.New("--exit-node is " + notSupported)
-		}
-		if upArgs.netfilterMode != "off" {
-			return errors.New("--netfilter-mode values besides \"off\" " + notSupported)
-		}
-	}
-
+// prefsFromUpArgs returns the ipn.Prefs for the provided args.
+//
+// Note that the parameters upArgs and warnf are named intentionally
+// to shadow the globals to prevent accidental misuse of them. This
+// function exists for testing and should have no side effects or
+// outside interactions (e.g. no making Tailscale local API calls).
+func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goos string) (*ipn.Prefs, error) {
 	routeMap := map[netaddr.IPPrefix]bool{}
 	var default4, default6 bool
 	if upArgs.advertiseRoutes != "" {
@@ -143,10 +137,10 @@ func runUp(ctx context.Context, args []string) error {
 		for _, s := range advroutes {
 			ipp, err := netaddr.ParseIPPrefix(s)
 			if err != nil {
-				fatalf("%q is not a valid IP address or CIDR prefix", s)
+				return nil, fmt.Errorf("%q is not a valid IP address or CIDR prefix", s)
 			}
 			if ipp != ipp.Masked() {
-				fatalf("%s has non-address bits set; expected %s", ipp, ipp.Masked())
+				return nil, fmt.Errorf("%s has non-address bits set; expected %s", ipp, ipp.Masked())
 			}
 			if ipp == ipv4default {
 				default4 = true
@@ -156,29 +150,24 @@ func runUp(ctx context.Context, args []string) error {
 			routeMap[ipp] = true
 		}
 		if default4 && !default6 {
-			fatalf("%s advertised without its IPv6 counterpart, please also advertise %s", ipv4default, ipv6default)
+			return nil, fmt.Errorf("%s advertised without its IPv6 counterpart, please also advertise %s", ipv4default, ipv6default)
 		} else if default6 && !default4 {
-			fatalf("%s advertised without its IPv6 counterpart, please also advertise %s", ipv6default, ipv4default)
+			return nil, fmt.Errorf("%s advertised without its IPv6 counterpart, please also advertise %s", ipv6default, ipv4default)
 		}
 	}
 	if upArgs.advertiseDefaultRoute {
 		routeMap[netaddr.MustParseIPPrefix("0.0.0.0/0")] = true
 		routeMap[netaddr.MustParseIPPrefix("::/0")] = true
 	}
-	if len(routeMap) > 0 {
-		if err := tailscale.CheckIPForwarding(context.Background()); err != nil {
-			warnf("%v", err)
-		}
-	}
 	routes := make([]netaddr.IPPrefix, 0, len(routeMap))
 	for r := range routeMap {
 		routes = append(routes, r)
 	}
 	sort.Slice(routes, func(i, j int) bool {
-		if routes[i].Bits != routes[j].Bits {
-			return routes[i].Bits < routes[j].Bits
+		if routes[i].Bits() != routes[j].Bits() {
+			return routes[i].Bits() < routes[j].Bits()
 		}
-		return routes[i].IP.Less(routes[j].IP)
+		return routes[i].IP().Less(routes[j].IP())
 	})
 
 	var exitNodeIP netaddr.IP
@@ -186,16 +175,16 @@ func runUp(ctx context.Context, args []string) error {
 		var err error
 		exitNodeIP, err = netaddr.ParseIP(upArgs.exitNodeIP)
 		if err != nil {
-			fatalf("invalid IP address %q for --exit-node: %v", upArgs.exitNodeIP, err)
+			return nil, fmt.Errorf("invalid IP address %q for --exit-node: %v", upArgs.exitNodeIP, err)
 		}
 	} else if upArgs.exitNodeAllowLANAccess {
-		fatalf("--exit-node-allow-lan-access can only be used with --exit-node")
+		return nil, fmt.Errorf("--exit-node-allow-lan-access can only be used with --exit-node")
 	}
 
-	if !exitNodeIP.IsZero() {
+	if upArgs.exitNodeIP != "" {
 		for _, ip := range st.TailscaleIPs {
 			if exitNodeIP == ip {
-				fatalf("cannot use %s as the exit node as it is a local IP address to this machine, did you mean --advertise-exit-node?", exitNodeIP)
+				return nil, fmt.Errorf("cannot use %s as the exit node as it is a local IP address to this machine, did you mean --advertise-exit-node?", upArgs.exitNodeIP)
 			}
 		}
 	}
@@ -206,13 +195,13 @@ func runUp(ctx context.Context, args []string) error {
 		for _, tag := range tags {
 			err := tailcfg.CheckTag(tag)
 			if err != nil {
-				fatalf("tag: %q: %s", tag, err)
+				return nil, fmt.Errorf("tag: %q: %s", tag, err)
 			}
 		}
 	}
 
 	if len(upArgs.hostname) > 256 {
-		fatalf("hostname too long: %d bytes (max 256)", len(upArgs.hostname))
+		return nil, fmt.Errorf("hostname too long: %d bytes (max 256)", len(upArgs.hostname))
 	}
 
 	prefs := ipn.NewPrefs()
@@ -226,11 +215,13 @@ func runUp(ctx context.Context, args []string) error {
 	prefs.ShieldsUp = upArgs.shieldsUp
 	prefs.AdvertiseRoutes = routes
 	prefs.AdvertiseTags = tags
-	prefs.NoSNAT = !upArgs.snat
 	prefs.Hostname = upArgs.hostname
 	prefs.ForceDaemon = upArgs.forceDaemon
+	prefs.OperatorUser = upArgs.opUser
+
+	if goos == "linux" {
+		prefs.NoSNAT = !upArgs.snat
 
-	if runtime.GOOS == "linux" {
 		switch upArgs.netfilterMode {
 		case "on":
 			prefs.NetfilterMode = preftype.NetfilterOn
@@ -239,9 +230,64 @@ func runUp(ctx context.Context, args []string) error {
 			warnf("netfilter=nodivert; add iptables calls to ts-* chains manually.")
 		case "off":
 			prefs.NetfilterMode = preftype.NetfilterOff
-			warnf("netfilter=off; configure iptables yourself.")
+			if defaultNetfilterMode() != "off" {
+				warnf("netfilter=off; configure iptables yourself.")
+			}
 		default:
-			fatalf("invalid value --netfilter-mode: %q", upArgs.netfilterMode)
+			return nil, fmt.Errorf("invalid value --netfilter-mode=%q", upArgs.netfilterMode)
+		}
+	}
+	return prefs, nil
+}
+
+func runUp(ctx context.Context, args []string) error {
+	if len(args) > 0 {
+		fatalf("too many non-flag arguments: %q", args)
+	}
+
+	st, err := tailscale.Status(ctx)
+	if err != nil {
+		fatalf("can't fetch status from tailscaled: %v", err)
+	}
+	origAuthURL := st.AuthURL
+
+	// printAuthURL reports whether we should print out the
+	// provided auth URL from an IPN notify.
+	printAuthURL := func(url string) bool {
+		if upArgs.authKey != "" {
+			// Issue 1755: when using an authkey, don't
+			// show an authURL that might still be pending
+			// from a previous non-completed interactive
+			// login.
+			return false
+		}
+		if upArgs.forceReauth && url == origAuthURL {
+			return false
+		}
+		return true
+	}
+
+	if distro.Get() == distro.Synology {
+		notSupported := "not supported on Synology; see https://github.com/tailscale/tailscale/issues/1995"
+		if upArgs.acceptRoutes {
+			return errors.New("--accept-routes is " + notSupported)
+		}
+		if upArgs.exitNodeIP != "" {
+			return errors.New("--exit-node is " + notSupported)
+		}
+		if upArgs.netfilterMode != "off" {
+			return errors.New("--netfilter-mode values besides \"off\" " + notSupported)
+		}
+	}
+
+	prefs, err := prefsFromUpArgs(upArgs, warnf, st, runtime.GOOS)
+	if err != nil {
+		fatalf("%s", err)
+	}
+
+	if len(prefs.AdvertiseRoutes) > 0 {
+		if err := tailscale.CheckIPForwarding(context.Background()); err != nil {
+			warnf("%v", err)
 		}
 	}
 
@@ -250,17 +296,13 @@ func runUp(ctx context.Context, args []string) error {
 		return err
 	}
 
-	flagSet := map[string]bool{}
-	mp := new(ipn.MaskedPrefs)
-	mp.WantRunningSet = true
-	mp.Prefs = *prefs
-	upFlagSet.Visit(func(f *flag.Flag) {
-		updateMaskedPrefsFromUpFlag(mp, f.Name)
-		flagSet[f.Name] = true
-	})
-
 	if !upArgs.reset {
-		if err := checkForAccidentalSettingReverts(flagSet, curPrefs, mp); err != nil {
+		applyImplicitPrefs(prefs, curPrefs, os.Getenv("USER"))
+
+		if err := checkForAccidentalSettingReverts(upFlagSet, curPrefs, prefs, upCheckEnv{
+			goos:          runtime.GOOS,
+			curExitNodeIP: exitNodeIP(prefs, st),
+		}); err != nil {
 			fatalf("%s", err)
 		}
 	}
@@ -272,7 +314,7 @@ func runUp(ctx context.Context, args []string) error {
 
 	// If we're already running and none of the flags require a
 	// restart, we can just do an EditPrefs call and change the
-	// prefs at runtime (e.g. changing hostname, changinged
+	// prefs at runtime (e.g. changing hostname, changing
 	// advertised tags, routes, etc)
 	justEdit := st.BackendState == ipn.Running.String() &&
 		!upArgs.forceReauth &&
@@ -280,6 +322,13 @@ func runUp(ctx context.Context, args []string) error {
 		upArgs.authKey == "" &&
 		!controlURLChanged
 	if justEdit {
+		mp := new(ipn.MaskedPrefs)
+		mp.WantRunningSet = true
+		mp.Prefs = *prefs
+		upFlagSet.Visit(func(f *flag.Flag) {
+			updateMaskedPrefsFromUpFlag(mp, f.Name)
+		})
+
 		_, err := tailscale.EditPrefs(ctx, mp)
 		return err
 	}
@@ -287,7 +336,10 @@ func runUp(ctx context.Context, args []string) error {
 	// simpleUp is whether we're running a simple "tailscale up"
 	// to transition to running from a previously-logged-in but
 	// down state, without changing any settings.
-	simpleUp := len(flagSet) == 0 && curPrefs.Persist != nil && curPrefs.Persist.LoginName != ""
+	simpleUp := upFlagSet.NFlag() == 0 &&
+		curPrefs.Persist != nil &&
+		curPrefs.Persist.LoginName != "" &&
+		st.BackendState != ipn.NeedsLogin.String()
 
 	// At this point we need to subscribe to the IPN bus to watch
 	// for state transitions and possible need to authenticate.
@@ -296,7 +348,8 @@ func runUp(ctx context.Context, args []string) error {
 
 	startingOrRunning := make(chan bool, 1) // gets value once starting or running
 	gotEngineUpdate := make(chan bool, 1)   // gets value upon an engine update
-	go pump(pumpCtx, bc, c)
+	pumpErr := make(chan error, 1)
+	go func() { pumpErr <- pump(pumpCtx, bc, c) }()
 
 	printed := !simpleUp
 	var loginOnce sync.Once
@@ -342,7 +395,7 @@ func runUp(ctx context.Context, args []string) error {
 				cancel()
 			}
 		}
-		if url := n.BrowseToURL; url != nil {
+		if url := n.BrowseToURL; url != nil && printAuthURL(*url) {
 			printed = true
 			fmt.Fprintf(os.Stderr, "\nTo authenticate, visit:\n\n\t%s\n\n", *url)
 		}
@@ -356,6 +409,8 @@ func runUp(ctx context.Context, args []string) error {
 	case <-gotEngineUpdate:
 	case <-pumpCtx.Done():
 		return pumpCtx.Err()
+	case err := <-pumpErr:
+		return err
 	}
 
 	// Special case: bare "tailscale up" means to just start
@@ -371,11 +426,10 @@ func runUp(ctx context.Context, args []string) error {
 			return err
 		}
 	} else {
-		bc.SetPrefs(prefs)
-
 		opts := ipn.Options{
-			StateKey: ipn.GlobalDaemonStateKey,
-			AuthKey:  upArgs.authKey,
+			StateKey:    ipn.GlobalDaemonStateKey,
+			AuthKey:     upArgs.authKey,
+			UpdatePrefs: prefs,
 		}
 		// On Windows, we still run in mostly the "legacy" way that
 		// predated the server's StateStore. That is, we send an empty
@@ -393,7 +447,9 @@ func runUp(ctx context.Context, args []string) error {
 		}
 
 		bc.Start(opts)
-		startLoginInteractive()
+		if upArgs.forceReauth {
+			startLoginInteractive()
+		}
 	}
 
 	select {
@@ -406,18 +462,26 @@ func runUp(ctx context.Context, args []string) error {
 		default:
 		}
 		return pumpCtx.Err()
+	case err := <-pumpErr:
+		return err
 	}
 }
 
 var (
-	flagForPref = map[string]string{} // "ExitNodeIP" => "exit-node"
-	prefsOfFlag = map[string][]string{}
+	prefsOfFlag = map[string][]string{} // "exit-node" => ExitNodeIP, ExitNodeID
 )
 
 func init() {
+	// Both these have the same ipn.Pref:
+	addPrefFlagMapping("advertise-exit-node", "AdvertiseRoutes")
+	addPrefFlagMapping("advertise-routes", "AdvertiseRoutes")
+
+	// And this flag has two ipn.Prefs:
+	addPrefFlagMapping("exit-node", "ExitNodeIP", "ExitNodeID")
+
+	// The rest are 1:1:
 	addPrefFlagMapping("accept-dns", "CorpDNS")
 	addPrefFlagMapping("accept-routes", "RouteAll")
-	addPrefFlagMapping("advertise-routes", "AdvertiseRoutes")
 	addPrefFlagMapping("advertise-tags", "AdvertiseTags")
 	addPrefFlagMapping("host-routes", "AllowSingleHosts")
 	addPrefFlagMapping("hostname", "Hostname")
@@ -425,17 +489,15 @@ func init() {
 	addPrefFlagMapping("netfilter-mode", "NetfilterMode")
 	addPrefFlagMapping("shields-up", "ShieldsUp")
 	addPrefFlagMapping("snat-subnet-routes", "NoSNAT")
-	addPrefFlagMapping("exit-node", "ExitNodeIP", "ExitNodeIP")
 	addPrefFlagMapping("exit-node-allow-lan-access", "ExitNodeAllowLANAccess")
 	addPrefFlagMapping("unattended", "ForceDaemon")
+	addPrefFlagMapping("operator", "OperatorUser")
 }
 
 func addPrefFlagMapping(flagName string, prefNames ...string) {
 	prefsOfFlag[flagName] = prefNames
 	prefType := reflect.TypeOf(ipn.Prefs{})
 	for _, pref := range prefNames {
-		flagForPref[pref] = flagName
-
 		// Crash at runtime if there's a typo in the prefName.
 		if _, ok := prefType.FieldByName(pref); !ok {
 			panic(fmt.Sprintf("invalid ipn.Prefs field %q", pref))
@@ -443,26 +505,45 @@ func addPrefFlagMapping(flagName string, prefNames ...string) {
 	}
 }
 
+// preflessFlag reports whether flagName is a flag that doesn't
+// correspond to an ipn.Pref.
+func preflessFlag(flagName string) bool {
+	switch flagName {
+	case "authkey", "force-reauth", "reset":
+		return true
+	}
+	return false
+}
+
 func updateMaskedPrefsFromUpFlag(mp *ipn.MaskedPrefs, flagName string) {
+	if preflessFlag(flagName) {
+		return
+	}
 	if prefs, ok := prefsOfFlag[flagName]; ok {
 		for _, pref := range prefs {
 			reflect.ValueOf(mp).Elem().FieldByName(pref + "Set").SetBool(true)
 		}
 		return
 	}
-	switch flagName {
-	case "authkey", "force-reauth", "reset":
-		// Not pref-related flags.
-	case "advertise-exit-node":
-		// This pref is a shorthand for advertise-routes.
-	default:
-		panic("internal error: unhandled flag " + flagName)
-	}
+	panic(fmt.Sprintf("internal error: unhandled flag %q", flagName))
 }
 
-// checkForAccidentalSettingReverts checks for people running
-// "tailscale up" with a subset of the flags they originally ran it
-// with.
+const accidentalUpPrefix = "Error: changing settings via 'tailscale up' requires mentioning all\n" +
+	"non-default flags. To proceed, either re-run your command with --reset or\n" +
+	"use the command below to explicitly mention the current value of\n" +
+	"all non-default settings:\n\n" +
+	"\ttailscale up"
+
+// upCheckEnv are extra parameters describing the environment as
+// needed by checkForAccidentalSettingReverts and friends.
+type upCheckEnv struct {
+	goos          string
+	curExitNodeIP netaddr.IP
+}
+
+// checkForAccidentalSettingReverts (the "up checker") checks for
+// people running "tailscale up" with a subset of the flags they
+// originally ran it with.
 //
 // For example, in Tailscale 1.6 and prior, a user might've advertised
 // a tag, but later tried to change just one other setting and forgot
@@ -474,61 +555,223 @@ func updateMaskedPrefsFromUpFlag(mp *ipn.MaskedPrefs, flagName string) {
 //
 // mp is the mask of settings actually set, where mp.Prefs is the new
 // preferences to set, including any values set from implicit flags.
-func checkForAccidentalSettingReverts(flagSet map[string]bool, curPrefs *ipn.Prefs, mp *ipn.MaskedPrefs) error {
-	if len(flagSet) == 0 {
+func checkForAccidentalSettingReverts(flagSet *flag.FlagSet, curPrefs, newPrefs *ipn.Prefs, env upCheckEnv) error {
+	if curPrefs.ControlURL == "" {
+		// Don't validate things on initial "up" before a control URL has been set.
+		return nil
+	}
+
+	flagIsSet := map[string]bool{}
+	flagSet.Visit(func(f *flag.Flag) {
+		flagIsSet[f.Name] = true
+	})
+
+	if len(flagIsSet) == 0 {
 		// A bare "tailscale up" is a special case to just
 		// mean bringing the network up without any changes.
 		return nil
 	}
-	curWithExplicitEdits := curPrefs.Clone()
-	curWithExplicitEdits.ApplyEdits(mp)
 
-	prefType := reflect.TypeOf(ipn.Prefs{})
+	// flagsCur is what flags we'd need to use to keep the exact
+	// settings as-is.
+	flagsCur := prefsToFlags(env, curPrefs)
+	flagsNew := prefsToFlags(env, newPrefs)
 
-	// Explicit values (current + explicit edit):
-	ev := reflect.ValueOf(curWithExplicitEdits).Elem()
-	// Implicit values (what we'd get if we replaced everything with flag defaults):
-	iv := reflect.ValueOf(&mp.Prefs).Elem()
-	var errs []error
-	var didExitNodeErr bool
-	for i := 0; i < prefType.NumField(); i++ {
-		prefName := prefType.Field(i).Name
-		if prefName == "Persist" {
+	var missing []string
+	for flagName := range flagsCur {
+		valCur, valNew := flagsCur[flagName], flagsNew[flagName]
+		if flagIsSet[flagName] {
 			continue
 		}
-		flagName, hasFlag := flagForPref[prefName]
-		if hasFlag && flagSet[flagName] {
+		if reflect.DeepEqual(valCur, valNew) {
 			continue
 		}
-		// Get explicit value and implicit value
-		evi, ivi := ev.Field(i).Interface(), iv.Field(i).Interface()
-		if reflect.DeepEqual(evi, ivi) {
-			continue
+		missing = append(missing, fmtFlagValueArg(flagName, valCur))
+	}
+	if len(missing) == 0 {
+		return nil
+	}
+	sort.Strings(missing)
+
+	// Compute the stringification of the explicitly provided args in flagSet
+	// to prepend to the command to run.
+	var explicit []string
+	flagSet.Visit(func(f *flag.Flag) {
+		type isBool interface {
+			IsBoolFlag() bool
 		}
-		switch flagName {
-		case "":
-			errs = append(errs, fmt.Errorf("'tailscale up' without --reset requires all preferences with changing values to be explicitly mentioned; this command would change the value of flagless pref %q", prefName))
-		case "exit-node":
-			if !didExitNodeErr {
-				didExitNodeErr = true
-				errs = append(errs, errors.New("'tailscale up' without --reset requires all preferences with changing values to be explicitly mentioned; --exit-node is not specified but an exit node is currently configured"))
+		if ib, ok := f.Value.(isBool); ok && ib.IsBoolFlag() {
+			if f.Value.String() == "false" {
+				explicit = append(explicit, "--"+f.Name+"=false")
+			} else {
+				explicit = append(explicit, "--"+f.Name)
 			}
-		default:
-			errs = append(errs, fmt.Errorf("'tailscale up' without --reset requires all preferences with changing values to be explicitly mentioned; --%s is not specified but its default value of %v differs from current value %v",
-				flagName, fmtSettingVal(ivi), fmtSettingVal(evi)))
+		} else {
+			explicit = append(explicit, fmtFlagValueArg(f.Name, f.Value.String()))
 		}
+	})
+
+	var sb strings.Builder
+	sb.WriteString(accidentalUpPrefix)
+
+	for _, a := range append(explicit, missing...) {
+		fmt.Fprintf(&sb, " %s", a)
 	}
-	return multierror.New(errs)
+	sb.WriteString("\n\n")
+	return errors.New(sb.String())
 }
 
-func fmtSettingVal(v interface{}) string {
-	switch v := v.(type) {
-	case bool:
-		return strconv.FormatBool(v)
-	case string, preftype.NetfilterMode:
-		return fmt.Sprintf("%q", v)
-	case []string:
-		return strings.Join(v, ",")
+// applyImplicitPrefs mutates prefs to add implicit preferences. Currently
+// this is just the operator user, which only needs to be set if it doesn't
+// match the current user.
+//
+// curUser is os.Getenv("USER"). It's pulled out for testability.
+func applyImplicitPrefs(prefs, oldPrefs *ipn.Prefs, curUser string) {
+	if prefs.OperatorUser == "" && oldPrefs.OperatorUser == curUser {
+		prefs.OperatorUser = oldPrefs.OperatorUser
 	}
-	return fmt.Sprint(v)
+}
+
+func flagAppliesToOS(flag, goos string) bool {
+	switch flag {
+	case "netfilter-mode", "snat-subnet-routes":
+		return goos == "linux"
+	case "unattended":
+		return goos == "windows"
+	}
+	return true
+}
+
+func prefsToFlags(env upCheckEnv, prefs *ipn.Prefs) (flagVal map[string]interface{}) {
+	ret := make(map[string]interface{})
+
+	exitNodeIPStr := func() string {
+		if !prefs.ExitNodeIP.IsZero() {
+			return prefs.ExitNodeIP.String()
+		}
+		if prefs.ExitNodeID.IsZero() || env.curExitNodeIP.IsZero() {
+			return ""
+		}
+		return env.curExitNodeIP.String()
+	}
+
+	fs := newUpFlagSet(env.goos, new(upArgsT) /* dummy */)
+	fs.VisitAll(func(f *flag.Flag) {
+		if preflessFlag(f.Name) {
+			return
+		}
+		set := func(v interface{}) {
+			if flagAppliesToOS(f.Name, env.goos) {
+				ret[f.Name] = v
+			} else {
+				ret[f.Name] = nil
+			}
+		}
+		switch f.Name {
+		default:
+			panic(fmt.Sprintf("unhandled flag %q", f.Name))
+		case "login-server":
+			set(prefs.ControlURL)
+		case "accept-routes":
+			set(prefs.RouteAll)
+		case "host-routes":
+			set(prefs.AllowSingleHosts)
+		case "accept-dns":
+			set(prefs.CorpDNS)
+		case "shields-up":
+			set(prefs.ShieldsUp)
+		case "exit-node":
+			set(exitNodeIPStr())
+		case "exit-node-allow-lan-access":
+			set(prefs.ExitNodeAllowLANAccess)
+		case "advertise-tags":
+			set(strings.Join(prefs.AdvertiseTags, ","))
+		case "hostname":
+			set(prefs.Hostname)
+		case "operator":
+			set(prefs.OperatorUser)
+		case "advertise-routes":
+			var sb strings.Builder
+			for i, r := range withoutExitNodes(prefs.AdvertiseRoutes) {
+				if i > 0 {
+					sb.WriteByte(',')
+				}
+				sb.WriteString(r.String())
+			}
+			set(sb.String())
+		case "advertise-exit-node":
+			set(hasExitNodeRoutes(prefs.AdvertiseRoutes))
+		case "snat-subnet-routes":
+			set(!prefs.NoSNAT)
+		case "netfilter-mode":
+			set(prefs.NetfilterMode.String())
+		case "unattended":
+			set(prefs.ForceDaemon)
+		}
+	})
+	return ret
+}
+
+func fmtFlagValueArg(flagName string, val interface{}) string {
+	if val == true {
+		return "--" + flagName
+	}
+	if val == "" {
+		return "--" + flagName + "="
+	}
+	return fmt.Sprintf("--%s=%v", flagName, shellquote.Join(fmt.Sprint(val)))
+}
+
+func hasExitNodeRoutes(rr []netaddr.IPPrefix) bool {
+	var v4, v6 bool
+	for _, r := range rr {
+		if r.Bits() == 0 {
+			if r.IP().Is4() {
+				v4 = true
+			} else if r.IP().Is6() {
+				v6 = true
+			}
+		}
+	}
+	return v4 && v6
+}
+
+// withoutExitNodes returns rr unchanged if it has only 1 or 0 /0
+// routes. If it has both IPv4 and IPv6 /0 routes, then it returns
+// a copy with all /0 routes removed.
+func withoutExitNodes(rr []netaddr.IPPrefix) []netaddr.IPPrefix {
+	if !hasExitNodeRoutes(rr) {
+		return rr
+	}
+	var out []netaddr.IPPrefix
+	for _, r := range rr {
+		if r.Bits() > 0 {
+			out = append(out, r)
+		}
+	}
+	return out
+}
+
+// exitNodeIP returns the exit node IP from p, using st to map
+// it from its ID form to an IP address if needed.
+func exitNodeIP(p *ipn.Prefs, st *ipnstate.Status) (ip netaddr.IP) {
+	if p == nil {
+		return
+	}
+	if !p.ExitNodeIP.IsZero() {
+		return p.ExitNodeIP
+	}
+	id := p.ExitNodeID
+	if id.IsZero() {
+		return
+	}
+	for _, p := range st.Peer {
+		if p.ID == id {
+			if len(p.TailscaleIPs) > 0 {
+				return p.TailscaleIPs[0]
+			}
+			break
+		}
+	}
+	return
 }

cmd/tailscale/cli/web.go

@@ -73,7 +73,11 @@ func runWeb(ctx context.Context, args []string) error {
 	}
 
 	if webArgs.cgi {
-		return cgi.Serve(http.HandlerFunc(webHandler))
+		if err := cgi.Serve(http.HandlerFunc(webHandler)); err != nil {
+			log.Printf("tailscale.cgi: %v", err)
+			return err
+		}
+		return nil
 	}
 	return http.ListenAndServe(webArgs.listen, http.HandlerFunc(webHandler))
 }
@@ -208,9 +212,10 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 	if r.Method == "POST" {
 		type mi map[string]interface{}
 		w.Header().Set("Content-Type", "application/json")
-		url, err := tailscaleUp(r.Context())
+		url, err := tailscaleUpForceReauth(r.Context())
 		if err != nil {
-			json.NewEncoder(w).Encode(mi{"error": err})
+			w.WriteHeader(500)
+			json.NewEncoder(w).Encode(mi{"error": err.Error()})
 			return
 		}
 		json.NewEncoder(w).Encode(mi{"url": url})
@@ -244,9 +249,9 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 }
 
 // TODO(crawshaw): some of this is very similar to the code in 'tailscale up', can we share anything?
-func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
+func tailscaleUpForceReauth(ctx context.Context) (authURL string, retErr error) {
 	prefs := ipn.NewPrefs()
-	prefs.ControlURL = "https://login.tailscale.com"
+	prefs.ControlURL = ipn.DefaultControlURL
 	prefs.WantRunning = true
 	prefs.CorpDNS = true
 	prefs.AllowSingleHosts = true
@@ -256,10 +261,31 @@ func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
 		prefs.NetfilterMode = preftype.NetfilterOff
 	}
 
-	c, bc, ctx, cancel := connect(ctx)
+	st, err := tailscale.Status(ctx)
+	if err != nil {
+		return "", fmt.Errorf("can't fetch status: %v", err)
+	}
+	origAuthURL := st.AuthURL
+
+	// printAuthURL reports whether we should print out the
+	// provided auth URL from an IPN notify.
+	printAuthURL := func(url string) bool {
+		return url != origAuthURL
+	}
+
+	c, bc, pumpCtx, cancel := connect(ctx)
 	defer cancel()
 
+	gotEngineUpdate := make(chan bool, 1) // gets value upon an engine update
+	go pump(pumpCtx, bc, c)
+
 	bc.SetNotifyCallback(func(n ipn.Notify) {
+		if n.Engine != nil {
+			select {
+			case gotEngineUpdate <- true:
+			default:
+			}
+		}
 		if n.ErrMessage != nil {
 			msg := *n.ErrMessage
 			if msg == ipn.ErrMsgPermissionDenied {
@@ -272,11 +298,21 @@ func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
 			}
 			retErr = fmt.Errorf("backend error: %v", msg)
 			cancel()
-		} else if url := n.BrowseToURL; url != nil {
+		} else if url := n.BrowseToURL; url != nil && printAuthURL(*url) {
 			authURL = *url
 			cancel()
 		}
 	})
+	// Wait for backend client to be connected so we know
+	// we're subscribed to updates. Otherwise we can miss
+	// an update upon its transition to running. Do so by causing some traffic
+	// back to the bus that we then wait on.
+	bc.RequestEngineStatus()
+	select {
+	case <-gotEngineUpdate:
+	case <-pumpCtx.Done():
+		return authURL, pumpCtx.Err()
+	}
 
 	bc.SetPrefs(prefs)
 
@@ -284,8 +320,11 @@ func tailscaleUp(ctx context.Context) (authURL string, retErr error) {
 		StateKey: ipn.GlobalDaemonStateKey,
 	})
 	bc.StartLoginInteractive()
-	pump(ctx, bc, c)
 
+	<-pumpCtx.Done() // wait for authURL or complete failure
+	if authURL == "" && retErr == nil {
+		retErr = pumpCtx.Err()
+	}
 	if authURL == "" && retErr == nil {
 		return "", fmt.Errorf("login failed with no backend error message")
 	}

cmd/tailscale/cli/web.html

@@ -11,140 +11,133 @@
 </head>
 
 <body class="py-14">
-	<main class="container max-w-lg mx-auto py-6 px-8 bg-white rounded-md shadow-2xl" style="width: 95%">
-		<header class="flex justify-between items-center min-width-0 py-2 mb-8">
-			<svg width="26" height="26" viewBox="0 0 23 23" title="Tailscale" fill="none" xmlns="http://www.w3.org/2000/svg"
-				class="flex-shrink-0 mr-4">
-				<circle opacity="0.2" cx="3.4" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle cx="3.4" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="3.4" cy="19.5" r="2.7" fill="currentColor"></circle>
-				<circle cx="11.5" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle cx="11.5" cy="19.5" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="11.5" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="19.5" cy="3.25" r="2.7" fill="currentColor"></circle>
-				<circle cx="19.5" cy="11.3" r="2.7" fill="currentColor"></circle>
-				<circle opacity="0.2" cx="19.5" cy="19.5" r="2.7" fill="currentColor"></circle>
-			</svg>
-			<div class="flex items-center justify-end space-x-2 w-2/3">
-				{{ with .Profile.LoginName }}
-				<div class="text-right truncate leading-4">
-					<h4 class="truncate">{{.}}</h4>
-					<a href="#" class="text-xs text-gray-500 hover:text-gray-700 js-loginButton">Switch account</a>
-				</div>
+<main class="container max-w-lg mx-auto py-6 px-8 bg-white rounded-md shadow-2xl" style="width: 95%">
+	<header class="flex justify-between items-center min-width-0 py-2 mb-8">
+		<svg width="26" height="26" viewBox="0 0 23 23" title="Tailscale" fill="none" xmlns="http://www.w3.org/2000/svg"
+			class="flex-shrink-0 mr-4">
+			<circle opacity="0.2" cx="3.4" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle cx="3.4" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="3.4" cy="19.5" r="2.7" fill="currentColor"></circle>
+			<circle cx="11.5" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle cx="11.5" cy="19.5" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="11.5" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="19.5" cy="3.25" r="2.7" fill="currentColor"></circle>
+			<circle cx="19.5" cy="11.3" r="2.7" fill="currentColor"></circle>
+			<circle opacity="0.2" cx="19.5" cy="19.5" r="2.7" fill="currentColor"></circle>
+		</svg>
+		<div class="flex items-center justify-end space-x-2 w-2/3">
+			{{ with .Profile.LoginName }}
+			<div class="text-right truncate leading-4">
+				<h4 class="truncate">{{.}}</h4>
+				<a href="#" class="text-xs text-gray-500 hover:text-gray-700 js-loginButton">Switch account</a>
+			</div>
+			{{ end }}
+			<div class="relative flex-shrink-0 w-8 h-8 rounded-full overflow-hidden">
+				{{ with .Profile.ProfilePicURL }}
+				<div class="w-8 h-8 flex pointer-events-none rounded-full bg-gray-200"
+					style="background-image: url('{{.}}'); background-size: cover;"></div>
+				{{ else }}
+				<div class="w-8 h-8 flex pointer-events-none rounded-full border border-gray-400 border-dashed"></div>
 				{{ end }}
-				<div class="relative flex-shrink-0 w-8 h-8 rounded-full overflow-hidden">
-					{{ with .Profile.ProfilePicURL }}
-					<div class="w-8 h-8 flex pointer-events-none rounded-full bg-gray-200"
-						style="background-image: url('{{.}}'); background-size: cover;"></div>
-					{{ else }}
-					<div class="w-8 h-8 flex pointer-events-none rounded-full border border-gray-400 border-dashed"></div>
-					{{ end }}
-				</div>
 			</div>
-		</header>
-		{{ if .IP }}
-		<div
-			class="border border-gray-200 bg-gray-0 rounded-lg p-2 pl-3 pr-3 mb-8 width-full flex items-center justify-between">
-			<div class="flex items-center min-width-0">
-				<svg class="flex-shrink-0 text-gray-600 mr-3 ml-1" xmlns="http://www.w3.org/2000/svg" width="20" height="20"
-					viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
-					stroke-linejoin="round">
-					<rect x="2" y="2" width="20" height="8" rx="2" ry="2"></rect>
-					<rect x="2" y="14" width="20" height="8" rx="2" ry="2"></rect>
-					<line x1="6" y1="6" x2="6.01" y2="6"></line>
-					<line x1="6" y1="18" x2="6.01" y2="18"></line>
-				</svg>
-				<h4 class="font-semibold truncate mr-2">{{.DeviceName}}</h4>
-			</div>
-			<h5>{{.IP}}</h5>
 		</div>
-		{{ end }}
-		{{ if or (eq .Status "NeedsLogin") (eq .Status "NoState") }}
-		{{ if .IP }}
-		<div class="mb-6">
-			<p class="text-gray-700">Your device's key has expired. Reauthenticate this device by logging in again, or <a
-					href="https://tailscale.com/kb/1028/key-expiry" class="link" target="_blank">learn more</a>.</p>
+	</header>
+	{{ if .IP }}
+	<div
+		class="border border-gray-200 bg-gray-0 rounded-lg p-2 pl-3 pr-3 mb-8 width-full flex items-center justify-between">
+		<div class="flex items-center min-width-0">
+			<svg class="flex-shrink-0 text-gray-600 mr-3 ml-1" xmlns="http://www.w3.org/2000/svg" width="20" height="20"
+				viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
+				stroke-linejoin="round">
+				<rect x="2" y="2" width="20" height="8" rx="2" ry="2"></rect>
+				<rect x="2" y="14" width="20" height="8" rx="2" ry="2"></rect>
+				<line x1="6" y1="6" x2="6.01" y2="6"></line>
+				<line x1="6" y1="18" x2="6.01" y2="18"></line>
+			</svg>
+			<h4 class="font-semibold truncate mr-2">{{.DeviceName}}</h4>
 		</div>
-		<a href="#" class="mb-4 js-loginButton" target="_blank">
-			<button class="button button-blue w-full">Reauthenticate</button>
-		</a>
-		{{ else }}
-		<div class="mb-6">
-			<h3 class="text-3xl font-semibold mb-3">Log in</h3>
-			<p class="text-gray-700">Get started by logging in to your Tailscale network. Or,&nbsp;learn&nbsp;more at <a
-					href="https://tailscale.com/" class="link" target="_blank">tailscale.com</a>.</p>
-		</div>
-		<a href="#" class="mb-4 js-loginButton" target="_blank">
-			<button class="button button-blue w-full">Log In</button>
-		</a>
-		{{ end }}
-		{{ else if eq .Status "NeedsMachineAuth" }}
-		<div class="mb-4">
-			This device is authorized, but needs approval from a network admin before it can connect to the network.
-		</div>
-		{{ else }}
-		<div class="mb-4">
-			<p>You are connected! Access this device over Tailscale using the device name or IP address above.</p>
-		</div>
-		<a href="#" class="mb-4 link font-medium js-loginButton" target="_blank">Reauthenticate</a>
-		{{ end }}
-	</main>
-	<script>
-		(function () {
-			let loginButtons = document.querySelectorAll(".js-loginButton");
-			let fetchingUrl = false;
+		<h5>{{.IP}}</h5>
+	</div>
+	{{ end }}
+	{{ if or (eq .Status "NeedsLogin") (eq .Status "NoState") }}
+	{{ if .IP }}
+	<div class="mb-6">
+		<p class="text-gray-700">Your device's key has expired. Reauthenticate this device by logging in again, or <a
+				href="https://tailscale.com/kb/1028/key-expiry" class="link" target="_blank">learn more</a>.</p>
+	</div>
+	<a href="#" class="mb-4 js-loginButton" target="_blank">
+		<button class="button button-blue w-full">Reauthenticate</button>
+	</a>
+	{{ else }}
+	<div class="mb-6">
+		<h3 class="text-3xl font-semibold mb-3">Log in</h3>
+		<p class="text-gray-700">Get started by logging in to your Tailscale network. Or,&nbsp;learn&nbsp;more at <a
+				href="https://tailscale.com/" class="link" target="_blank">tailscale.com</a>.</p>
+	</div>
+	<a href="#" class="mb-4 js-loginButton" target="_blank">
+		<button class="button button-blue w-full">Log In</button>
+	</a>
+	{{ end }}
+	{{ else if eq .Status "NeedsMachineAuth" }}
+	<div class="mb-4">
+		This device is authorized, but needs approval from a network admin before it can connect to the network.
+	</div>
+	{{ else }}
+	<div class="mb-4">
+		<p>You are connected! Access this device over Tailscale using the device name or IP address above.</p>
+	</div>
+	<a href="#" class="mb-4 link font-medium js-loginButton" target="_blank">Reauthenticate</a>
+	{{ end }}
+</main>
+<script>(function () {
+let loginButtons = document.querySelectorAll(".js-loginButton");
+let fetchingUrl = false;
 
-			function handleClick(e) {
-				e.preventDefault();
+function handleClick(e) {
+	e.preventDefault();
 
-				if (fetchingUrl) {
-					return;
-				}
+	if (fetchingUrl) {
+		return;
+	}
 
-				fetchingUrl = true;
-				const urlParams = new URLSearchParams(window.location.search);
-				const token = urlParams.get("SynoToken");
-				const nextParams = new URLSearchParams({ up: true });
-				if (token) {
-					nextParams.set("SynoToken", token)
-				}
-				const nextUrl = new URL(window.location);
-				nextUrl.search = nextParams.toString()
-				const url = nextUrl.toString();
+	fetchingUrl = true;
+	const urlParams = new URLSearchParams(window.location.search);
+	const token = urlParams.get("SynoToken");
+	const nextParams = new URLSearchParams({ up: true });
+	if (token) {
+		nextParams.set("SynoToken", token)
+	}
+	const nextUrl = new URL(window.location);
+	nextUrl.search = nextParams.toString()
+	const url = nextUrl.toString();
 
-				const tab = window.open("/redirect", "_blank");
+	fetch(url, {
+		method: "POST",
+		headers: {
+			"Accept": "application/json",
+			"Content-Type": "application/json",
+		}
+	}).then(res => res.json()).then(res => {
+		fetchingUrl = false;
+		const err = res["error"];
+		if (err) {
+			throw new Error(err);
+		}
+		const url = res["url"];
+		if (url) {
+			document.location.href = url;
+		} else {
+			location.reload();
+		}
+	}).catch(err => {
+		alert("Failed to log in: " + err.message);
+	});
+}
 
-				fetch(url, {
-					method: "POST",
-					headers: {
-						"Accept": "application/json",
-						"Content-Type": "application/json",
-					}
-				}).then(res => res.json()).then(res => {
-					fetchingUrl = false;
-					const err = res["error"];
-					if (err) {
-						throw new Error(err);
-					}
-					const url = res["url"];
-					if (url) {
-						authUrl = url;
-						tab.location = url;
-						tab.focus();
-					} else {
-						location.reload();
-					}
-				}).catch(err => {
-					tab.close();
-					alert("Failed to log in: " + err.message);
-				});
-			}
-
-			Array.from(loginButtons).forEach(el => {
-				el.addEventListener("click", handleClick);
-			})
-		})();
-	</script>
+Array.from(loginButtons).forEach(el => {
+	el.addEventListener("click", handleClick);
+})
+})();</script>
 </body>
 
 </html>

cmd/tailscale/depaware.txt

@@ -1,8 +1,9 @@
 tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/depaware)
 
-   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate
+   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate+
+   W    github.com/alexbrainman/sspi/internal/common                 from github.com/alexbrainman/sspi/negotiate
    W 💣 github.com/alexbrainman/sspi/negotiate                       from tailscale.com/net/tshttpproxy
-        github.com/go-multierror/multierror                          from tailscale.com/cmd/tailscale/cli
+        github.com/kballard/go-shellquote                            from tailscale.com/cmd/tailscale/cli
         github.com/peterbourgon/ff/v2                                from github.com/peterbourgon/ff/v2/ffcli
         github.com/peterbourgon/ff/v2/ffcli                          from tailscale.com/cmd/tailscale/cli
         github.com/tcnksm/go-httpstat                                from tailscale.com/net/netcheck
@@ -15,7 +16,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         rsc.io/goversion/version                                     from tailscale.com/version
         tailscale.com/atomicfile                                     from tailscale.com/ipn
         tailscale.com/client/tailscale                               from tailscale.com/cmd/tailscale/cli
-        tailscale.com/client/tailscale/apitype                       from tailscale.com/client/tailscale
+        tailscale.com/client/tailscale/apitype                       from tailscale.com/client/tailscale+
         tailscale.com/cmd/tailscale/cli                              from tailscale.com/cmd/tailscale
         tailscale.com/derp                                           from tailscale.com/derp/derphttp
         tailscale.com/derp/derphttp                                  from tailscale.com/net/netcheck
@@ -33,7 +34,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/net/portmapper                                 from tailscale.com/net/netcheck+
         tailscale.com/net/stun                                       from tailscale.com/net/netcheck
         tailscale.com/net/tlsdial                                    from tailscale.com/derp/derphttp
-        tailscale.com/net/tsaddr                                     from tailscale.com/net/interfaces
+        tailscale.com/net/tsaddr                                     from tailscale.com/net/interfaces+
      💣 tailscale.com/net/tshttpproxy                                from tailscale.com/derp/derphttp+
         tailscale.com/paths                                          from tailscale.com/cmd/tailscale/cli+
         tailscale.com/safesocket                                     from tailscale.com/cmd/tailscale/cli+
@@ -48,7 +49,6 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/types/opt                                      from tailscale.com/net/netcheck+
         tailscale.com/types/persist                                  from tailscale.com/ipn
         tailscale.com/types/preftype                                 from tailscale.com/cmd/tailscale/cli+
-        tailscale.com/types/strbuilder                               from tailscale.com/net/packet
         tailscale.com/types/structs                                  from tailscale.com/ipn+
         tailscale.com/types/wgkey                                    from tailscale.com/types/netmap+
         tailscale.com/util/dnsname                                   from tailscale.com/cmd/tailscale/cli+
@@ -85,7 +85,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         golang.org/x/text/transform                                  from golang.org/x/text/secure/bidirule+
         golang.org/x/text/unicode/bidi                               from golang.org/x/net/idna+
         golang.org/x/text/unicode/norm                               from golang.org/x/net/idna
-        golang.org/x/time/rate                                       from tailscale.com/types/logger+
+        golang.org/x/time/rate                                       from tailscale.com/cmd/tailscale/cli+
         bufio                                                        from compress/flate+
         bytes                                                        from bufio+
         compress/flate                                               from compress/gzip+

cmd/tailscaled/depaware.txt

@@ -1,41 +1,42 @@
 tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/depaware)
 
-   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate
+   W 💣 github.com/alexbrainman/sspi                                 from github.com/alexbrainman/sspi/negotiate+
+   W    github.com/alexbrainman/sspi/internal/common                 from github.com/alexbrainman/sspi/negotiate
    W 💣 github.com/alexbrainman/sspi/negotiate                       from tailscale.com/net/tshttpproxy
    L    github.com/coreos/go-iptables/iptables                       from tailscale.com/wgengine/router
-   W 💣 github.com/github/certstore                                  from tailscale.com/control/controlclient
         github.com/go-multierror/multierror                          from tailscale.com/wgengine/router+
    W 💣 github.com/go-ole/go-ole                                     from github.com/go-ole/go-ole/oleutil+
    W 💣 github.com/go-ole/go-ole/oleutil                             from tailscale.com/wgengine/winnet
    L 💣 github.com/godbus/dbus/v5                                    from tailscale.com/net/dns
+        github.com/golang/snappy                                     from github.com/klauspost/compress/zstd
         github.com/google/btree                                      from inet.af/netstack/tcpip/header+
    L    github.com/josharian/native                                  from github.com/mdlayher/netlink+
    L 💣 github.com/jsimonetti/rtnetlink                              from tailscale.com/wgengine/monitor
    L    github.com/jsimonetti/rtnetlink/internal/unix                from github.com/jsimonetti/rtnetlink
         github.com/klauspost/compress/fse                            from github.com/klauspost/compress/huff0
         github.com/klauspost/compress/huff0                          from github.com/klauspost/compress/zstd
-        github.com/klauspost/compress/snappy                         from github.com/klauspost/compress/zstd
         github.com/klauspost/compress/zstd                           from tailscale.com/smallzstd
         github.com/klauspost/compress/zstd/internal/xxhash           from github.com/klauspost/compress/zstd
    L 💣 github.com/mdlayher/netlink                                  from github.com/jsimonetti/rtnetlink+
    L 💣 github.com/mdlayher/netlink/nlenc                            from github.com/jsimonetti/rtnetlink+
    L    github.com/mdlayher/sdnotify                                 from tailscale.com/util/systemd
-   W    github.com/pkg/errors                                        from github.com/github/certstore
-     💣 github.com/tailscale/wireguard-go/conn                       from github.com/tailscale/wireguard-go/device+
-   W 💣 github.com/tailscale/wireguard-go/conn/winrio                from github.com/tailscale/wireguard-go/conn
-     💣 github.com/tailscale/wireguard-go/device                     from tailscale.com/wgengine+
-     💣 github.com/tailscale/wireguard-go/ipc                        from github.com/tailscale/wireguard-go/device
-   W 💣 github.com/tailscale/wireguard-go/ipc/winpipe                from github.com/tailscale/wireguard-go/ipc
-        github.com/tailscale/wireguard-go/ratelimiter                from github.com/tailscale/wireguard-go/device
-        github.com/tailscale/wireguard-go/replay                     from github.com/tailscale/wireguard-go/device
-        github.com/tailscale/wireguard-go/rwcancel                   from github.com/tailscale/wireguard-go/device+
-        github.com/tailscale/wireguard-go/tai64n                     from github.com/tailscale/wireguard-go/device+
-     💣 github.com/tailscale/wireguard-go/tun                        from github.com/tailscale/wireguard-go/device+
-   W 💣 github.com/tailscale/wireguard-go/tun/wintun                 from github.com/tailscale/wireguard-go/tun+
+   W    github.com/pkg/errors                                        from github.com/tailscale/certstore
+   W 💣 github.com/tailscale/certstore                               from tailscale.com/control/controlclient
         github.com/tcnksm/go-httpstat                                from tailscale.com/net/netcheck
      💣 go4.org/intern                                               from inet.af/netaddr
      💣 go4.org/mem                                                  from tailscale.com/control/controlclient+
         go4.org/unsafe/assume-no-moving-gc                           from go4.org/intern
+     💣 golang.zx2c4.com/wireguard/conn                              from golang.zx2c4.com/wireguard/device+
+   W 💣 golang.zx2c4.com/wireguard/conn/winrio                       from golang.zx2c4.com/wireguard/conn
+     💣 golang.zx2c4.com/wireguard/device                            from tailscale.com/net/tstun+
+     💣 golang.zx2c4.com/wireguard/ipc                               from golang.zx2c4.com/wireguard/device
+   W 💣 golang.zx2c4.com/wireguard/ipc/winpipe                       from golang.zx2c4.com/wireguard/ipc
+        golang.zx2c4.com/wireguard/ratelimiter                       from golang.zx2c4.com/wireguard/device
+        golang.zx2c4.com/wireguard/replay                            from golang.zx2c4.com/wireguard/device
+        golang.zx2c4.com/wireguard/rwcancel                          from golang.zx2c4.com/wireguard/device+
+        golang.zx2c4.com/wireguard/tai64n                            from golang.zx2c4.com/wireguard/device+
+     💣 golang.zx2c4.com/wireguard/tun                               from golang.zx2c4.com/wireguard/device+
+   W 💣 golang.zx2c4.com/wireguard/tun/wintun                        from golang.zx2c4.com/wireguard/tun+
    W 💣 golang.zx2c4.com/wireguard/windows/tunnel/winipcfg           from tailscale.com/net/interfaces+
         inet.af/netaddr                                              from tailscale.com/control/controlclient+
      💣 inet.af/netstack/gohacks                                     from inet.af/netstack/state/wire+
@@ -69,6 +70,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         inet.af/netstack/tcpip/transport/udp                         from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/waiter                                      from inet.af/netstack/tcpip+
         inet.af/peercred                                             from tailscale.com/ipn/ipnserver
+   W 💣 inet.af/wf                                                   from tailscale.com/wf
         rsc.io/goversion/version                                     from tailscale.com/version
         tailscale.com/atomicfile                                     from tailscale.com/ipn+
         tailscale.com/client/tailscale/apitype                       from tailscale.com/ipn/ipnlocal+
@@ -78,7 +80,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/derp/derpmap                                   from tailscale.com/cmd/tailscaled+
         tailscale.com/disco                                          from tailscale.com/derp+
         tailscale.com/health                                         from tailscale.com/control/controlclient+
-        tailscale.com/internal/deepprint                             from tailscale.com/ipn/ipnlocal+
+        tailscale.com/internal/deephash                              from tailscale.com/ipn/ipnlocal+
         tailscale.com/ipn                                            from tailscale.com/ipn/ipnserver+
         tailscale.com/ipn/ipnlocal                                   from tailscale.com/ipn/ipnserver+
         tailscale.com/ipn/ipnserver                                  from tailscale.com/cmd/tailscaled
@@ -115,7 +117,6 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/smallzstd                                      from tailscale.com/ipn/ipnserver+
         tailscale.com/syncs                                          from tailscale.com/net/interfaces+
         tailscale.com/tailcfg                                        from tailscale.com/control/controlclient+
-   W 💣 tailscale.com/tempfork/wireguard-windows/firewall            from tailscale.com/cmd/tailscaled
    W    tailscale.com/tsconst                                        from tailscale.com/net/interfaces
         tailscale.com/tstime                                         from tailscale.com/wgengine/magicsock
         tailscale.com/types/empty                                    from tailscale.com/control/controlclient+
@@ -128,21 +129,24 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/types/opt                                      from tailscale.com/control/controlclient+
         tailscale.com/types/persist                                  from tailscale.com/control/controlclient+
         tailscale.com/types/preftype                                 from tailscale.com/ipn+
-        tailscale.com/types/strbuilder                               from tailscale.com/net/packet
         tailscale.com/types/structs                                  from tailscale.com/control/controlclient+
         tailscale.com/types/wgkey                                    from tailscale.com/control/controlclient+
+   L    tailscale.com/util/cmpver                                    from tailscale.com/net/dns
         tailscale.com/util/dnsname                                   from tailscale.com/ipn/ipnstate+
   LW    tailscale.com/util/endian                                    from tailscale.com/net/netns+
    L    tailscale.com/util/lineread                                  from tailscale.com/control/controlclient+
+        tailscale.com/util/osshare                                   from tailscale.com/cmd/tailscaled+
         tailscale.com/util/pidowner                                  from tailscale.com/ipn/ipnserver
         tailscale.com/util/racebuild                                 from tailscale.com/logpolicy
         tailscale.com/util/systemd                                   from tailscale.com/control/controlclient+
+        tailscale.com/util/uniq                                      from tailscale.com/wgengine/magicsock
         tailscale.com/util/winutil                                   from tailscale.com/logpolicy+
         tailscale.com/version                                        from tailscale.com/cmd/tailscaled+
         tailscale.com/version/distro                                 from tailscale.com/control/controlclient+
+   W    tailscale.com/wf                                             from tailscale.com/cmd/tailscaled
         tailscale.com/wgengine                                       from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/filter                                from tailscale.com/control/controlclient+
-        tailscale.com/wgengine/magicsock                             from tailscale.com/cmd/tailscaled+
+        tailscale.com/wgengine/magicsock                             from tailscale.com/wgengine+
         tailscale.com/wgengine/monitor                               from tailscale.com/wgengine+
         tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled
         tailscale.com/wgengine/router                                from tailscale.com/cmd/tailscaled+
@@ -151,7 +155,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/wgengine/wglog                                 from tailscale.com/wgengine
    W 💣 tailscale.com/wgengine/winnet                                from tailscale.com/wgengine/router
         golang.org/x/crypto/blake2b                                  from golang.org/x/crypto/nacl/box
-        golang.org/x/crypto/blake2s                                  from github.com/tailscale/wireguard-go/device+
+        golang.org/x/crypto/blake2s                                  from golang.zx2c4.com/wireguard/device+
         golang.org/x/crypto/chacha20                                 from golang.org/x/crypto/chacha20poly1305
         golang.org/x/crypto/chacha20poly1305                         from crypto/tls+
         golang.org/x/crypto/cryptobyte                               from crypto/ecdsa+
@@ -160,7 +164,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         golang.org/x/crypto/hkdf                                     from crypto/tls
         golang.org/x/crypto/nacl/box                                 from tailscale.com/control/controlclient+
         golang.org/x/crypto/nacl/secretbox                           from golang.org/x/crypto/nacl/box
-        golang.org/x/crypto/poly1305                                 from github.com/tailscale/wireguard-go/device+
+        golang.org/x/crypto/poly1305                                 from golang.org/x/crypto/chacha20poly1305+
         golang.org/x/crypto/salsa20/salsa                            from golang.org/x/crypto/nacl/box+
         golang.org/x/net/bpf                                         from github.com/mdlayher/netlink+
         golang.org/x/net/dns/dnsmessage                              from net+
@@ -168,23 +172,24 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         golang.org/x/net/http/httpproxy                              from net/http
         golang.org/x/net/http2/hpack                                 from net/http
         golang.org/x/net/idna                                        from golang.org/x/net/http/httpguts+
-        golang.org/x/net/ipv4                                        from github.com/tailscale/wireguard-go/device
-        golang.org/x/net/ipv6                                        from github.com/tailscale/wireguard-go/device+
+        golang.org/x/net/ipv4                                        from golang.zx2c4.com/wireguard/device
+        golang.org/x/net/ipv6                                        from golang.zx2c4.com/wireguard/device+
         golang.org/x/net/proxy                                       from tailscale.com/net/netns
    D    golang.org/x/net/route                                       from net+
         golang.org/x/sync/errgroup                                   from tailscale.com/derp
         golang.org/x/sync/singleflight                               from tailscale.com/net/dnscache
         golang.org/x/sys/cpu                                         from golang.org/x/crypto/blake2b+
   LD    golang.org/x/sys/unix                                        from github.com/jsimonetti/rtnetlink/internal/unix+
-   W    golang.org/x/sys/windows                                     from github.com/tailscale/wireguard-go/conn+
+   W    golang.org/x/sys/windows                                     from github.com/go-ole/go-ole+
    W    golang.org/x/sys/windows/registry                            from golang.zx2c4.com/wireguard/windows/tunnel/winipcfg+
-   W    golang.org/x/sys/windows/svc                                 from tailscale.com/cmd/tailscaled
+   W    golang.org/x/sys/windows/svc                                 from tailscale.com/cmd/tailscaled+
+   W    golang.org/x/sys/windows/svc/mgr                             from tailscale.com/cmd/tailscaled
         golang.org/x/term                                            from tailscale.com/logpolicy
         golang.org/x/text/secure/bidirule                            from golang.org/x/net/idna
         golang.org/x/text/transform                                  from golang.org/x/text/secure/bidirule+
         golang.org/x/text/unicode/bidi                               from golang.org/x/net/idna+
         golang.org/x/text/unicode/norm                               from golang.org/x/net/idna
-        golang.org/x/time/rate                                       from tailscale.com/types/logger+
+        golang.org/x/time/rate                                       from inet.af/netstack/tcpip/stack+
         bufio                                                        from compress/flate+
         bytes                                                        from bufio+
         compress/flate                                               from compress/gzip+
@@ -217,7 +222,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         debug/elf                                                    from rsc.io/goversion/version
         debug/macho                                                  from rsc.io/goversion/version
         debug/pe                                                     from rsc.io/goversion/version
-        embed                                                        from tailscale.com/net/dns
+   L    embed                                                        from tailscale.com/net/dns
         encoding                                                     from encoding/json+
         encoding/asn1                                                from crypto/x509+
         encoding/base64                                              from encoding/json+

cmd/tailscaled/install_windows.go

@@ -0,0 +1,123 @@
+// Copyright (c) 2021 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"
+	"errors"
+	"fmt"
+	"os"
+	"time"
+
+	"golang.org/x/sys/windows"
+	"golang.org/x/sys/windows/svc"
+	"golang.org/x/sys/windows/svc/mgr"
+	"tailscale.com/logtail/backoff"
+	"tailscale.com/types/logger"
+	"tailscale.com/util/osshare"
+)
+
+func init() {
+	installSystemDaemon = installSystemDaemonWindows
+	uninstallSystemDaemon = uninstallSystemDaemonWindows
+}
+
+func installSystemDaemonWindows(args []string) (err error) {
+	m, err := mgr.Connect()
+	if err != nil {
+		return fmt.Errorf("failed to connect to Windows service manager: %v", err)
+	}
+
+	service, err := m.OpenService(serviceName)
+	if err == nil {
+		service.Close()
+		return fmt.Errorf("service %q is already installed", serviceName)
+	}
+
+	// no such service; proceed to install the service.
+
+	exe, err := os.Executable()
+	if err != nil {
+		return err
+	}
+
+	c := mgr.Config{
+		ServiceType:  windows.SERVICE_WIN32_OWN_PROCESS,
+		StartType:    mgr.StartAutomatic,
+		ErrorControl: mgr.ErrorNormal,
+		DisplayName:  serviceName,
+		Description:  "Connects this computer to others on the Tailscale network.",
+	}
+
+	service, err = m.CreateService(serviceName, exe, c)
+	if err != nil {
+		return fmt.Errorf("failed to create %q service: %v", serviceName, err)
+	}
+	defer service.Close()
+
+	// Exponential backoff is often too aggressive, so use (mostly)
+	// squares instead.
+	ra := []mgr.RecoveryAction{
+		{mgr.ServiceRestart, 1 * time.Second},
+		{mgr.ServiceRestart, 2 * time.Second},
+		{mgr.ServiceRestart, 4 * time.Second},
+		{mgr.ServiceRestart, 9 * time.Second},
+		{mgr.ServiceRestart, 16 * time.Second},
+		{mgr.ServiceRestart, 25 * time.Second},
+		{mgr.ServiceRestart, 36 * time.Second},
+		{mgr.ServiceRestart, 49 * time.Second},
+		{mgr.ServiceRestart, 64 * time.Second},
+	}
+	const resetPeriodSecs = 60
+	err = service.SetRecoveryActions(ra, resetPeriodSecs)
+	if err != nil {
+		return fmt.Errorf("failed to set service recovery actions: %v", err)
+	}
+
+	return nil
+}
+
+func uninstallSystemDaemonWindows(args []string) (ret error) {
+	// Remove file sharing from Windows shell (noop in non-windows)
+	osshare.SetFileSharingEnabled(false, logger.Discard)
+
+	m, err := mgr.Connect()
+	if err != nil {
+		return fmt.Errorf("failed to connect to Windows service manager: %v", err)
+	}
+	defer m.Disconnect()
+
+	service, err := m.OpenService(serviceName)
+	if err != nil {
+		return fmt.Errorf("failed to open %q service: %v", serviceName, err)
+	}
+
+	st, err := service.Query()
+	if err != nil {
+		service.Close()
+		return fmt.Errorf("failed to query service state: %v", err)
+	}
+	if st.State != svc.Stopped {
+		service.Control(svc.Stop)
+	}
+	err = service.Delete()
+	service.Close()
+	if err != nil {
+		return fmt.Errorf("failed to delete service: %v", err)
+	}
+
+	bo := backoff.NewBackoff("uninstall", logger.Discard, 30*time.Second)
+	end := time.Now().Add(15 * time.Second)
+	for time.Until(end) > 0 {
+		service, err = m.OpenService(serviceName)
+		if err != nil {
+			// service is no longer openable; success!
+			break
+		}
+		service.Close()
+		bo.BackOff(context.Background(), errors.New("service not deleted"))
+	}
+	return nil
+}

cmd/tailscaled/tailscaled.go

@@ -29,19 +29,21 @@ import (
 	"time"
 
 	"github.com/go-multierror/multierror"
+	"inet.af/netaddr"
 	"tailscale.com/ipn/ipnserver"
 	"tailscale.com/logpolicy"
 	"tailscale.com/net/dns"
 	"tailscale.com/net/socks5"
+	"tailscale.com/net/tsaddr"
 	"tailscale.com/net/tstun"
 	"tailscale.com/paths"
 	"tailscale.com/types/flagtype"
 	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
+	"tailscale.com/util/osshare"
 	"tailscale.com/version"
 	"tailscale.com/version/distro"
 	"tailscale.com/wgengine"
-	"tailscale.com/wgengine/magicsock"
 	"tailscale.com/wgengine/monitor"
 	"tailscale.com/wgengine/netstack"
 	"tailscale.com/wgengine/router"
@@ -114,7 +116,7 @@ func main() {
 	flag.StringVar(&args.debug, "debug", "", "listen address ([ip]:port) of optional debug server")
 	flag.StringVar(&args.socksAddr, "socks5-server", "", `optional [ip]:port to run a SOCK5 server (e.g. "localhost:1080")`)
 	flag.StringVar(&args.tunname, "tun", defaultTunName(), `tunnel interface name; use "userspace-networking" (beta) to not use TUN`)
-	flag.Var(flagtype.PortValue(&args.port, magicsock.DefaultPort), "port", "UDP port to listen on for WireGuard and peer-to-peer traffic; 0 means automatically select")
+	flag.Var(flagtype.PortValue(&args.port, 0), "port", "UDP port to listen on for WireGuard and peer-to-peer traffic; 0 means automatically select")
 	flag.StringVar(&args.statepath, "state", paths.DefaultTailscaledStateFile(), "path of state file")
 	flag.StringVar(&args.socketpath, "socket", paths.DefaultTailscaledSocket(), "path of the service unix socket")
 	flag.BoolVar(&printVersion, "version", false, "print version information and exit")
@@ -158,7 +160,12 @@ func main() {
 		log.Fatalf("--socket is required")
 	}
 
-	if err := run(); err != nil {
+	err := run()
+
+	// Remove file sharing from Windows shell (noop in non-windows)
+	osshare.SetFileSharingEnabled(false, logger.Discard)
+
+	if err != nil {
 		// No need to log; the func already did
 		os.Exit(1)
 	}
@@ -239,30 +246,31 @@ func run() error {
 		srv := &socks5.Server{
 			Logf: logger.WithPrefix(logf, "socks5: "),
 		}
-		// TODO: also consider wrapNetstack, where dials can go to either Tailscale
-		// or non-Tailscale targets. But that's also basically what
-		// https://github.com/tailscale/tailscale/issues/1617 is about, so do them
-		// both at the same time.
-		if useNetstack {
-			srv.Dialer = func(ctx context.Context, network, addr string) (net.Conn, error) {
+		var (
+			mu  sync.Mutex // guards the following field
+			dns netstack.DNSMap
+		)
+		e.AddNetworkMapCallback(func(nm *netmap.NetworkMap) {
+			mu.Lock()
+			defer mu.Unlock()
+			dns = netstack.DNSMapFromNetworkMap(nm)
+		})
+		useNetstackForIP := func(ip netaddr.IP) bool {
+			// TODO(bradfitz): this isn't exactly right.
+			// We should also support subnets when the
+			// prefs are configured as such.
+			return tsaddr.IsTailscaleIP(ip)
+		}
+		srv.Dialer = func(ctx context.Context, network, addr string) (net.Conn, error) {
+			ipp, err := dns.Resolve(ctx, addr)
+			if err != nil {
+				return nil, err
+			}
+			if ns != nil && useNetstackForIP(ipp.IP()) {
 				return ns.DialContextTCP(ctx, addr)
 			}
-		} else {
-			var mu sync.Mutex
-			var dns netstack.DNSMap
-			e.AddNetworkMapCallback(func(nm *netmap.NetworkMap) {
-				mu.Lock()
-				defer mu.Unlock()
-				dns = netstack.DNSMapFromNetworkMap(nm)
-			})
-			srv.Dialer = func(ctx context.Context, network, addr string) (net.Conn, error) {
-				ipp, err := dns.Resolve(ctx, addr)
-				if err != nil {
-					return nil, err
-				}
-				var d net.Dialer
-				return d.DialContext(ctx, network, ipp.String())
-			}
+			var d net.Dialer
+			return d.DialContext(ctx, network, ipp.String())
 		}
 		go func() {
 			log.Fatalf("SOCKS5 server exited: %v", srv.Serve(socksListener))
@@ -295,7 +303,7 @@ func run() error {
 		Port:               41112,
 		StatePath:          args.statepath,
 		AutostartStateKey:  globalStateKey,
-		SurviveDisconnects: true,
+		SurviveDisconnects: runtime.GOOS != "windows",
 		DebugMux:           debugMux,
 	}
 	err = ipnserver.Run(ctx, logf, pol.PublicID.String(), ipnserver.FixedEngine(e), opts)

cmd/tailscaled/tailscaled_windows.go

@@ -21,7 +21,6 @@ import (
 	"context"
 	"fmt"
 	"log"
-	"net"
 	"os"
 	"time"
 
@@ -32,9 +31,9 @@ import (
 	"tailscale.com/logpolicy"
 	"tailscale.com/net/dns"
 	"tailscale.com/net/tstun"
-	"tailscale.com/tempfork/wireguard-windows/firewall"
 	"tailscale.com/types/logger"
 	"tailscale.com/version"
+	"tailscale.com/wf"
 	"tailscale.com/wgengine"
 	"tailscale.com/wgengine/netstack"
 	"tailscale.com/wgengine/router"
@@ -144,13 +143,13 @@ func beFirewallKillswitch() bool {
 
 	luid, err := winipcfg.LUIDFromGUID(&guid)
 	if err != nil {
-		log.Fatalf("no interface with GUID %q", guid)
+		log.Fatalf("no interface with GUID %q: %v", guid, err)
 	}
 
-	noProtection := false
-	var dnsIPs []net.IP // unused in called code.
 	start := time.Now()
-	firewall.EnableFirewall(uint64(luid), noProtection, dnsIPs)
+	if _, err := wf.New(uint64(luid)); err != nil {
+		log.Fatalf("filewall creation failed: %v", err)
+	}
 	log.Printf("killswitch enabled, took %s", time.Since(start))
 
 	// Block until the monitor goroutine shuts us down.
@@ -159,18 +158,16 @@ func beFirewallKillswitch() bool {
 
 func startIPNServer(ctx context.Context, logid string) error {
 	var logf logger.Logf = log.Printf
-	var eng wgengine.Engine
-	var err error
 
-	getEngine := func() (wgengine.Engine, error) {
+	getEngineRaw := func() (wgengine.Engine, error) {
 		dev, devName, err := tstun.New(logf, "Tailscale")
 		if err != nil {
-			return nil, err
+			return nil, fmt.Errorf("TUN: %w", err)
 		}
 		r, err := router.New(logf, dev)
 		if err != nil {
 			dev.Close()
-			return nil, err
+			return nil, fmt.Errorf("router: %w", err)
 		}
 		if wrapNetstack {
 			r = netstack.NewSubnetRouterWrapper(r)
@@ -179,7 +176,7 @@ func startIPNServer(ctx context.Context, logid string) error {
 		if err != nil {
 			r.Close()
 			dev.Close()
-			return nil, err
+			return nil, fmt.Errorf("DNS: %w", err)
 		}
 		eng, err := wgengine.NewUserspaceEngine(logf, wgengine.Config{
 			Tun:        dev,
@@ -190,7 +187,7 @@ func startIPNServer(ctx context.Context, logid string) error {
 		if err != nil {
 			r.Close()
 			dev.Close()
-			return nil, err
+			return nil, fmt.Errorf("engine: %w", err)
 		}
 		onlySubnets := true
 		if wrapNetstack {
@@ -199,60 +196,82 @@ func startIPNServer(ctx context.Context, logid string) error {
 		return wgengine.NewWatchdog(eng), nil
 	}
 
-	if msg := os.Getenv("TS_DEBUG_WIN_FAIL"); msg != "" {
-		err = fmt.Errorf("pretending to be a service failure: %v", msg)
-	} else {
-		// We have a bunch of bug reports of wgengine.NewUserspaceEngine returning a few different errors,
-		// all intermittently. A few times I (Brad) have also seen sporadic failures that simply
-		// restarting fixed. So try a few times.
-		for try := 1; try <= 5; try++ {
-			if try > 1 {
-				// Only sleep a bit. Don't do some massive backoff because
-				// the frontend GUI has a 30 second timeout on connecting to us,
-				// but even 5 seconds is too long for them to get any results.
-				// 5 tries * 1 second each seems fine.
-				time.Sleep(time.Second)
-			}
-			eng, err = getEngine()
+	type engineOrError struct {
+		Engine wgengine.Engine
+		Err    error
+	}
+	engErrc := make(chan engineOrError)
+	t0 := time.Now()
+	go func() {
+		const ms = time.Millisecond
+		for try := 1; ; try++ {
+			logf("tailscaled: getting engine... (try %v)", try)
+			t1 := time.Now()
+			eng, err := getEngineRaw()
+			d, dt := time.Since(t1).Round(ms), time.Since(t1).Round(ms)
 			if err != nil {
-				logf("wgengine.NewUserspaceEngine: (try %v) %v", try, err)
-				continue
+				logf("tailscaled: engine fetch error (try %v) in %v (total %v, sysUptime %v): %v",
+					try, d, dt, windowsUptime().Round(time.Second), err)
+			} else {
+				if try > 1 {
+					logf("tailscaled: got engine on try %v in %v (total %v)", try, d, dt)
+				} else {
+					logf("tailscaled: got engine in %v", d)
+				}
 			}
-			if try > 1 {
-				logf("wgengine.NewUserspaceEngine: ended up working on try %v", try)
+			timer := time.NewTimer(5 * time.Second)
+			engErrc <- engineOrError{eng, err}
+			if err == nil {
+				timer.Stop()
+				return
 			}
-			break
+			<-timer.C
 		}
-	}
-	if err != nil {
-		// Log the error, but don't fatalf. We want to
-		// propagate the error message to the UI frontend. So
-		// we continue and tell the ipnserver to return that
-		// in a Notify message.
-		logf("wgengine.NewUserspaceEngine: %v", err)
-	}
+	}()
+
 	opts := ipnserver.Options{
 		Port:               41112,
 		SurviveDisconnects: false,
 		StatePath:          args.statepath,
 	}
-	if err != nil {
-		// Return nicer errors to users, annotated with logids, which helps
-		// when they file bugs.
-		rawGetEngine := getEngine // raw == without verbose logid-containing error
-		getEngine = func() (wgengine.Engine, error) {
-			eng, err := rawGetEngine()
-			if err != nil {
-				return nil, fmt.Errorf("wgengine.NewUserspaceEngine: %v\n\nlogid: %v", err, logid)
-			}
-			return eng, nil
+
+	// getEngine is called by ipnserver to get the engine. It's
+	// not called concurrently and is not called again once it
+	// successfully returns an engine.
+	getEngine := func() (wgengine.Engine, error) {
+		if msg := os.Getenv("TS_DEBUG_WIN_FAIL"); msg != "" {
+			return nil, fmt.Errorf("pretending to be a service failure: %v", msg)
+		}
+		for {
+			res := <-engErrc
+			if res.Engine != nil {
+				return res.Engine, nil
+			}
+			if time.Since(t0) < time.Minute || windowsUptime() < 10*time.Minute {
+				// Ignore errors during early boot. Windows 10 auto logs in the GUI
+				// way sooner than the networking stack components start up.
+				// So the network will fail for a bit (and require a few tries) while
+				// the GUI is still fine.
+				continue
+			}
+			// Return nicer errors to users, annotated with logids, which helps
+			// when they file bugs.
+			return nil, fmt.Errorf("%w\n\nlogid: %v", res.Err, logid)
 		}
-	} else {
-		getEngine = ipnserver.FixedEngine(eng)
 	}
-	err = ipnserver.Run(ctx, logf, logid, getEngine, opts)
+	err := ipnserver.Run(ctx, logf, logid, getEngine, opts)
 	if err != nil {
 		logf("ipnserver.Run: %v", err)
 	}
 	return err
 }
+
+var (
+	kernel32           = windows.NewLazySystemDLL("kernel32.dll")
+	getTickCount64Proc = kernel32.NewProc("GetTickCount64")
+)
+
+func windowsUptime() time.Duration {
+	r, _, _ := getTickCount64Proc.Call()
+	return time.Duration(int64(r)) * time.Millisecond
+}

cmd/tsshd/tsshd.go

@@ -59,11 +59,11 @@ func main() {
 
 	warned := false
 	for {
-		addr, iface, err := interfaces.Tailscale()
+		addrs, iface, err := interfaces.Tailscale()
 		if err != nil {
 			log.Fatalf("listing interfaces: %v", err)
 		}
-		if addr == nil {
+		if len(addrs) == 0 {
 			if !warned {
 				log.Printf("no tailscale interface found; polling until one is available")
 				warned = true
@@ -75,6 +75,13 @@ func main() {
 			continue
 		}
 		warned = false
+		var addr netaddr.IP
+		for _, a := range addrs {
+			if a.Is4() {
+				addr = a
+				break
+			}
+		}
 		listen := net.JoinHostPort(addr.String(), fmt.Sprint(*port))
 		log.Printf("tailscale ssh server listening on %v, %v", iface.Name, listen)
 		s := &ssh.Server{

control/controlclient/auto.go

@@ -2,18 +2,12 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// Package controlclient implements the client for the Tailscale
-// control plane.
-//
-// It handles authentication, port picking, and collects the local
-// network configuration.
 package controlclient
 
 import (
 	"context"
-	"encoding/json"
 	"fmt"
-	"reflect"
+	"log"
 	"sync"
 	"time"
 
@@ -28,77 +22,6 @@ import (
 	"tailscale.com/types/wgkey"
 )
 
-// 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
-)
-
-func (s State) MarshalText() ([]byte, error) {
-	return []byte(s.String()), nil
-}
-
-func (s State) String() string {
-	switch s {
-	case StateNew:
-		return "state:new"
-	case StateNotAuthenticated:
-		return "state:not-authenticated"
-	case StateAuthenticating:
-		return "state:authenticating"
-	case StateURLVisitRequired:
-		return "state:url-visit-required"
-	case StateAuthenticated:
-		return "state:authenticated"
-	case StateSynchronized:
-		return "state:synchronized"
-	default:
-		return fmt.Sprintf("state:unknown:%d", int(s))
-	}
-}
-
-type Status struct {
-	_             structs.Incomparable
-	LoginFinished *empty.Message
-	Err           string
-	URL           string
-	Persist       *persist.Persist   // locally persisted configuration
-	NetMap        *netmap.NetworkMap // server-pushed configuration
-	Hostinfo      *tailcfg.Hostinfo  // current Hostinfo data
-	State         State
-}
-
-// Equal reports whether s and s2 are equal.
-func (s *Status) Equal(s2 *Status) bool {
-	if s == nil && s2 == nil {
-		return true
-	}
-	return s != nil && s2 != nil &&
-		(s.LoginFinished == nil) == (s2.LoginFinished == nil) &&
-		s.Err == s2.Err &&
-		s.URL == s2.URL &&
-		reflect.DeepEqual(s.Persist, s2.Persist) &&
-		reflect.DeepEqual(s.NetMap, s2.NetMap) &&
-		reflect.DeepEqual(s.Hostinfo, s2.Hostinfo) &&
-		s.State == s2.State
-}
-
-func (s Status) String() string {
-	b, err := json.MarshalIndent(s, "", "\t")
-	if err != nil {
-		panic(err)
-	}
-	return s.State.String() + " " + string(b)
-}
-
 type LoginGoal struct {
 	_               structs.Incomparable
 	wantLoggedIn    bool                 // true if we *want* to be logged in
@@ -118,8 +41,9 @@ func (g *LoginGoal) sendLogoutError(err error) {
 	}
 }
 
-// Client connects to a tailcontrol server for a node.
-type Client struct {
+// Auto connects to a tailcontrol server for a node.
+// It's a concrete implementation of the Client interface.
+type Auto struct {
 	direct   *Direct // our interface to the server APIs
 	timeNow  func() time.Time
 	logf     logger.Logf
@@ -152,8 +76,8 @@ type Client struct {
 	mapDone    chan struct{}   // when closed, map goroutine is done
 }
 
-// New creates and starts a new Client.
-func New(opts Options) (*Client, error) {
+// New creates and starts a new Auto.
+func New(opts Options) (*Auto, error) {
 	c, err := NewNoStart(opts)
 	if c != nil {
 		c.Start()
@@ -161,8 +85,8 @@ func New(opts Options) (*Client, error) {
 	return c, err
 }
 
-// NewNoStart creates a new Client, but without calling Start on it.
-func NewNoStart(opts Options) (*Client, error) {
+// NewNoStart creates a new Auto, but without calling Start on it.
+func NewNoStart(opts Options) (*Auto, error) {
 	direct, err := NewDirect(opts)
 	if err != nil {
 		return nil, err
@@ -173,7 +97,7 @@ func NewNoStart(opts Options) (*Client, error) {
 	if opts.TimeNow == nil {
 		opts.TimeNow = time.Now
 	}
-	c := &Client{
+	c := &Auto{
 		direct:   direct,
 		timeNow:  opts.TimeNow,
 		logf:     opts.Logf,
@@ -189,7 +113,7 @@ func NewNoStart(opts Options) (*Client, error) {
 
 }
 
-func (c *Client) onHealthChange(sys health.Subsystem, err error) {
+func (c *Auto) onHealthChange(sys health.Subsystem, err error) {
 	if sys == health.SysOverall {
 		return
 	}
@@ -200,12 +124,13 @@ func (c *Client) onHealthChange(sys health.Subsystem, err error) {
 // SetPaused controls whether HTTP activity should be paused.
 //
 // The client can be paused and unpaused repeatedly, unlike Start and Shutdown, which can only be used once.
-func (c *Client) SetPaused(paused bool) {
+func (c *Auto) SetPaused(paused bool) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	if paused == c.paused {
 		return
 	}
+	c.logf("setPaused(%v)", paused)
 	c.paused = paused
 	if paused {
 		// Only cancel the map routine. (The auth routine isn't expensive
@@ -222,7 +147,7 @@ func (c *Client) SetPaused(paused bool) {
 // Start starts the client's goroutines.
 //
 // It should only be called for clients created by NewNoStart.
-func (c *Client) Start() {
+func (c *Auto) Start() {
 	go c.authRoutine()
 	go c.mapRoutine()
 }
@@ -232,7 +157,8 @@ func (c *Client) Start() {
 // streaming response open), or start a new streaming one if necessary.
 //
 // It should be called whenever there's something new to tell the server.
-func (c *Client) sendNewMapRequest() {
+func (c *Auto) sendNewMapRequest() {
+	log.Println("sendNewMapRequest breakpoint")
 	c.mu.Lock()
 
 	// If we're not already streaming a netmap, or if we're already stuck
@@ -271,7 +197,7 @@ func (c *Client) sendNewMapRequest() {
 	}()
 }
 
-func (c *Client) cancelAuth() {
+func (c *Auto) cancelAuth() {
 	c.mu.Lock()
 	if c.authCancel != nil {
 		c.authCancel()
@@ -282,7 +208,7 @@ func (c *Client) cancelAuth() {
 	c.mu.Unlock()
 }
 
-func (c *Client) cancelMapLocked() {
+func (c *Auto) cancelMapLocked() {
 	if c.mapCancel != nil {
 		c.mapCancel()
 	}
@@ -291,13 +217,13 @@ func (c *Client) cancelMapLocked() {
 	}
 }
 
-func (c *Client) cancelMapUnsafely() {
+func (c *Auto) cancelMapUnsafely() {
 	c.mu.Lock()
 	c.cancelMapLocked()
 	c.mu.Unlock()
 }
 
-func (c *Client) cancelMapSafely() {
+func (c *Auto) cancelMapSafely() {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
@@ -333,7 +259,7 @@ func (c *Client) cancelMapSafely() {
 	}
 }
 
-func (c *Client) authRoutine() {
+func (c *Auto) authRoutine() {
 	defer close(c.authDone)
 	bo := backoff.NewBackoff("authRoutine", c.logf, 30*time.Second)
 
@@ -344,7 +270,7 @@ func (c *Client) authRoutine() {
 		if goal != nil {
 			c.logf("authRoutine: %s; wantLoggedIn=%v", c.state, goal.wantLoggedIn)
 		} else {
-			c.logf("authRoutine: %s; goal=nil", c.state)
+			c.logf("authRoutine: %s; goal=nil paused=%v", c.state, c.paused)
 		}
 		c.mu.Unlock()
 
@@ -452,7 +378,7 @@ 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 {
+func (c *Auto) Expiry() *time.Time {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	return c.expiry
@@ -460,21 +386,21 @@ func (c *Client) Expiry() *time.Time {
 
 // Direct returns the underlying direct client object. Used in tests
 // only.
-func (c *Client) Direct() *Direct {
+func (c *Auto) Direct() *Direct {
 	return c.direct
 }
 
 // unpausedChanLocked returns a new channel that is closed when the
-// current Client pause is unpaused.
+// current Auto pause is unpaused.
 //
 // c.mu must be held
-func (c *Client) unpausedChanLocked() <-chan struct{} {
+func (c *Auto) unpausedChanLocked() <-chan struct{} {
 	unpaused := make(chan struct{})
 	c.unpauseWaiters = append(c.unpauseWaiters, unpaused)
 	return unpaused
 }
 
-func (c *Client) mapRoutine() {
+func (c *Auto) mapRoutine() {
 	defer close(c.mapDone)
 	bo := backoff.NewBackoff("mapRoutine", c.logf, 30*time.Second)
 
@@ -596,7 +522,10 @@ func (c *Client) mapRoutine() {
 	}
 }
 
-func (c *Client) AuthCantContinue() bool {
+func (c *Auto) AuthCantContinue() bool {
+	if c == nil {
+		return true
+	}
 	c.mu.Lock()
 	defer c.mu.Unlock()
 
@@ -604,13 +533,13 @@ func (c *Client) AuthCantContinue() bool {
 }
 
 // SetStatusFunc sets fn as the callback to run on any status change.
-func (c *Client) SetStatusFunc(fn func(Status)) {
+func (c *Auto) SetStatusFunc(fn func(Status)) {
 	c.mu.Lock()
 	c.statusFunc = fn
 	c.mu.Unlock()
 }
 
-func (c *Client) SetHostinfo(hi *tailcfg.Hostinfo) {
+func (c *Auto) SetHostinfo(hi *tailcfg.Hostinfo) {
 	if hi == nil {
 		panic("nil Hostinfo")
 	}
@@ -623,7 +552,7 @@ func (c *Client) SetHostinfo(hi *tailcfg.Hostinfo) {
 	c.sendNewMapRequest()
 }
 
-func (c *Client) SetNetInfo(ni *tailcfg.NetInfo) {
+func (c *Auto) SetNetInfo(ni *tailcfg.NetInfo) {
 	if ni == nil {
 		panic("nil NetInfo")
 	}
@@ -636,7 +565,7 @@ func (c *Client) SetNetInfo(ni *tailcfg.NetInfo) {
 	c.sendNewMapRequest()
 }
 
-func (c *Client) sendStatus(who string, err error, url string, nm *netmap.NetworkMap) {
+func (c *Auto) sendStatus(who string, err error, url string, nm *netmap.NetworkMap) {
 	c.mu.Lock()
 	state := c.state
 	loggedIn := c.loggedIn
@@ -681,7 +610,7 @@ func (c *Client) sendStatus(who string, err error, url string, nm *netmap.Networ
 	c.mu.Unlock()
 }
 
-func (c *Client) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
+func (c *Auto) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
 	c.logf("client.Login(%v, %v)", t != nil, flags)
 
 	c.mu.Lock()
@@ -695,7 +624,7 @@ func (c *Client) Login(t *tailcfg.Oauth2Token, flags LoginFlags) {
 	c.cancelAuth()
 }
 
-func (c *Client) StartLogout() {
+func (c *Auto) StartLogout() {
 	c.logf("client.StartLogout()")
 
 	c.mu.Lock()
@@ -706,7 +635,7 @@ func (c *Client) StartLogout() {
 	c.cancelAuth()
 }
 
-func (c *Client) Logout(ctx context.Context) error {
+func (c *Auto) Logout(ctx context.Context) error {
 	c.logf("client.Logout()")
 
 	errc := make(chan error, 1)
@@ -738,14 +667,14 @@ func (c *Client) Logout(ctx context.Context) error {
 //
 // The localPort field is unused except for integration tests in
 // another repo.
-func (c *Client) UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) {
+func (c *Auto) UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) {
 	changed := c.direct.SetEndpoints(localPort, endpoints)
 	if changed {
 		c.sendNewMapRequest()
 	}
 }
 
-func (c *Client) Shutdown() {
+func (c *Auto) Shutdown() {
 	c.logf("client.Shutdown()")
 
 	c.mu.Lock()
@@ -771,17 +700,17 @@ func (c *Client) Shutdown() {
 
 // NodePublicKey returns the node public key currently in use. This is
 // used exclusively in tests.
-func (c *Client) TestOnlyNodePublicKey() wgkey.Key {
+func (c *Auto) TestOnlyNodePublicKey() wgkey.Key {
 	priv := c.direct.GetPersist()
 	return priv.PrivateNodeKey.Public()
 }
 
-func (c *Client) TestOnlySetAuthKey(authkey string) {
+func (c *Auto) TestOnlySetAuthKey(authkey string) {
 	c.direct.mu.Lock()
 	defer c.direct.mu.Unlock()
 	c.direct.authKey = authkey
 }
 
-func (c *Client) TestOnlyTimeNow() time.Time {
+func (c *Auto) TestOnlyTimeNow() time.Time {
 	return c.timeNow()
 }

control/controlclient/client.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 controlclient implements the client for the Tailscale
+// control plane.
+//
+// It handles authentication, port picking, and collects the local
+// network configuration.
+package controlclient
+
+import (
+	"context"
+
+	"tailscale.com/tailcfg"
+)
+
+type LoginFlags int
+
+const (
+	LoginDefault     = LoginFlags(0)
+	LoginInteractive = LoginFlags(1 << iota) // force user login and key refresh
+)
+
+// Client represents a client connection to the control server.
+// Currently this is done through a pair of polling https requests in
+// the Auto client, but that might change eventually.
+type Client interface {
+	// SetStatusFunc provides a callback to call when control sends us
+	// a message.
+	SetStatusFunc(func(Status))
+	// Shutdown closes this session, which should not be used any further
+	// afterwards.
+	Shutdown()
+	// Login begins an interactive or non-interactive login process.
+	// Client will eventually call the Status callback with either a
+	// LoginFinished flag (on success) or an auth URL (if further
+	// interaction is needed).
+	Login(*tailcfg.Oauth2Token, LoginFlags)
+	// StartLogout starts an asynchronous logout process.
+	// When it finishes, the Status callback will be called while
+	// AuthCantContinue()==true.
+	StartLogout()
+	// Logout starts a synchronous logout process. It doesn't return
+	// until the logout operation has been completed.
+	Logout(context.Context) error
+	// SetPaused pauses or unpauses the controlclient activity as much
+	// as possible, without losing its internal state, to minimize
+	// unnecessary network activity.
+	// TODO: It might be better to simply shutdown the controlclient and
+	// make a new one when it's time to unpause.
+	SetPaused(bool)
+	// AuthCantContinue returns whether authentication is blocked. If it
+	// is, you either need to visit the auth URL (previously sent in a
+	// Status callback) or call the Login function appropriately.
+	// TODO: this probably belongs in the Status itself instead.
+	AuthCantContinue() bool
+	// SetHostinfo changes the Hostinfo structure that will be sent in
+	// subsequent node registration requests.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	SetHostinfo(*tailcfg.Hostinfo)
+	// SetNetinfo changes the NetIinfo structure that will be sent in
+	// subsequent node registration requests.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	SetNetInfo(*tailcfg.NetInfo)
+	// UpdateEndpoints changes the Endpoint structure that will be sent
+	// in subsequent node registration requests.
+	// TODO: localPort seems to be obsolete, remove it.
+	// TODO: a server-side change would let us simply upload this
+	// in a separate http request. It has nothing to do with the rest of
+	// the state machine.
+	UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint)
+}

control/controlclient/controlclient_test.go

@@ -22,7 +22,7 @@ func fieldsOf(t reflect.Type) (fields []string) {
 
 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", "NetMap", "State", "Persist", "Hostinfo"}
 	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)

control/controlclient/direct.go

@@ -23,7 +23,6 @@ import (
 	"path/filepath"
 	"reflect"
 	"runtime"
-	"sort"
 	"strconv"
 	"strings"
 	"sync"
@@ -33,6 +32,7 @@ import (
 	"golang.org/x/crypto/nacl/box"
 	"inet.af/netaddr"
 	"tailscale.com/health"
+	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/log/logheap"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/dnsfallback"
@@ -49,7 +49,6 @@ import (
 	"tailscale.com/util/dnsname"
 	"tailscale.com/util/systemd"
 	"tailscale.com/version"
-	"tailscale.com/wgengine/filter"
 	"tailscale.com/wgengine/monitor"
 )
 
@@ -68,6 +67,7 @@ type Direct struct {
 	debugFlags             []string
 	keepSharerAndUserSplit bool
 	skipIPForwardingCheck  bool
+	pinger                 Pinger
 
 	mu           sync.Mutex // mutex guards the following fields
 	serverKey    wgkey.Key
@@ -81,6 +81,11 @@ type Direct struct {
 	everEndpoints bool   // whether we've ever had non-empty endpoints
 	localPort     uint16 // or zero to mean auto
 }
+type Pinger interface {
+	// Ping is a request to start a discovery ping with the peer handling
+	// the given IP and then call cb with its ping latency & method.
+	Ping(ip netaddr.IP, useTSMP bool, cb func(*ipnstate.PingResult))
+}
 
 type Options struct {
 	Persist              persist.Persist               // initial persistent data
@@ -96,6 +101,7 @@ type Options struct {
 	HTTPTestClient       *http.Client // optional HTTP client to use (for tests only)
 	DebugFlags           []string     // debug settings to send to control
 	LinkMonitor          *monitor.Mon // optional link monitor
+	Pinger               Pinger
 
 	// KeepSharerAndUserSplit controls whether the client
 	// understands Node.Sharer. If false, the Sharer is mapped to the User.
@@ -167,6 +173,7 @@ func NewDirect(opts Options) (*Direct, error) {
 		keepSharerAndUserSplit: opts.KeepSharerAndUserSplit,
 		linkMon:                opts.LinkMonitor,
 		skipIPForwardingCheck:  opts.SkipIPForwardingCheck,
+		pinger:                 opts.Pinger,
 	}
 	if opts.Hostinfo == nil {
 		c.SetHostinfo(NewHostinfo())
@@ -253,13 +260,6 @@ func (c *Direct) GetPersist() persist.Persist {
 	return c.persist
 }
 
-type LoginFlags int
-
-const (
-	LoginDefault     = LoginFlags(0)
-	LoginInteractive = LoginFlags(1 << iota) // force user login and key refresh
-)
-
 func (c *Direct) TryLogout(ctx context.Context) error {
 	c.logf("direct.TryLogout()")
 
@@ -415,7 +415,7 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 
 		// Don't log the common error types. Signatures are not usually enabled,
 		// so these are expected.
-		if err != errCertificateNotConfigured && err != errNoCertStore {
+		if !errors.Is(err, errCertificateNotConfigured) && !errors.Is(err, errNoCertStore) {
 			c.logf("RegisterReq sign error: %v", err)
 		}
 	}
@@ -469,10 +469,10 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 			request.NodeKey.ShortString())
 		return true, "", nil
 	}
-	if persist.Provider == "" {
+	if resp.Login.Provider != "" {
 		persist.Provider = resp.Login.Provider
 	}
-	if persist.LoginName == "" {
+	if resp.Login.LoginName != "" {
 		persist.LoginName = resp.Login.LoginName
 	}
 
@@ -562,6 +562,7 @@ func inTest() bool { return flag.Lookup("test.v") != nil }
 // maxPolls is how many network maps to download; common values are 1
 // or -1 (to keep a long-poll query open to the server).
 func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*netmap.NetworkMap)) error {
+	log.Println("POLLNETMAP BREAKPOINT")
 	return c.sendMapRequest(ctx, maxPolls, cb)
 }
 
@@ -569,12 +570,19 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*netmap.N
 // but does not fetch anything. It returns an error if the server did not return a
 // successful 200 OK response.
 func (c *Direct) SendLiteMapUpdate(ctx context.Context) error {
+	log.Println("SendLiteMapUpdate BREAKPOINT")
 	return c.sendMapRequest(ctx, 1, nil)
 }
 
+// If we go more than pollTimeout without hearing from the server,
+// end the long poll. We should be receiving a keep alive ping
+// every minute.
+const pollTimeout = 120 * time.Second
+
 // cb nil means to omit peers.
 func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netmap.NetworkMap)) error {
 	c.mu.Lock()
+	log.Println("sendMapRequest ENDPOINT")
 	persist := c.persist
 	serverURL := c.serverURL
 	serverKey := c.serverKey
@@ -696,10 +704,6 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		return nil
 	}
 
-	// If we go more than pollTimeout without hearing from the server,
-	// end the long poll. We should be receiving a keep alive ping
-	// every minute.
-	const pollTimeout = 120 * time.Second
 	timeout := time.NewTimer(pollTimeout)
 	timeoutReset := make(chan struct{})
 	pollDone := make(chan struct{})
@@ -729,11 +733,11 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		}
 	}()
 
-	var lastDNSConfig = new(tailcfg.DNSConfig)
-	var lastDERPMap *tailcfg.DERPMap
-	var lastUserProfile = map[tailcfg.UserID]tailcfg.UserProfile{}
-	var lastParsedPacketFilter []filter.Match
-	var collectServices bool
+	sess := newMapSession(persist.PrivateNodeKey)
+	sess.logf = c.logf
+	sess.vlogf = vlogf
+	sess.machinePubKey = machinePubKey
+	sess.keepSharerAndUserSplit = c.keepSharerAndUserSplit
 
 	// If allowStream, then the server will use an HTTP long poll to
 	// return incremental results. There is always one response right
@@ -742,7 +746,6 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 	// the same format before just closing the connection.
 	// We can use this same read loop either way.
 	var msg []byte
-	var previousPeers []*tailcfg.Node // for delta-purposes
 	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
@@ -770,6 +773,11 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		}
 
 		if pr := resp.PingRequest; pr != nil {
+			// return err
+			log.Println("Ping Triggered")
+			for i := 0; i < 10; i++ {
+				c.CustomPing(&resp)
+			}
 			go answerPing(c.logf, c.httpc, pr)
 		}
 
@@ -789,16 +797,6 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			continue
 		}
 
-		undeltaPeers(&resp, previousPeers)
-		previousPeers = cloneNodes(resp.Peers) // defensive/lazy clone, since this escapes to who knows where
-		for _, up := range resp.UserProfiles {
-			lastUserProfile[up.ID] = up
-		}
-
-		if resp.DERPMap != nil {
-			vlogf("netmap: new map contains DERP map")
-			lastDERPMap = resp.DERPMap
-		}
 		if resp.Debug != nil {
 			if resp.Debug.LogHeapPprof {
 				go logheap.LogHeap(resp.Debug.LogHeapURL)
@@ -808,18 +806,40 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			}
 			setControlAtomic(&controlUseDERPRoute, resp.Debug.DERPRoute)
 			setControlAtomic(&controlTrimWGConfig, resp.Debug.TrimWGConfig)
+			if sleep := time.Duration(resp.Debug.SleepSeconds * float64(time.Second)); sleep > 0 {
+				if err := sleepAsRequested(ctx, c.logf, timeoutReset, sleep); err != nil {
+					return err
+				}
+			}
 		}
+
+		nm := sess.netmapForResponse(&resp)
+		if nm.SelfNode == nil {
+			c.logf("MapResponse lacked node")
+			return errors.New("MapResponse lacked node")
+		}
+
 		// 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)
+			anyOld, numDisco := false, 0
+			for _, p := range nm.Peers {
+				if p.DiscoKey.IsZero() {
+					anyOld = true
+				} else {
+					numDisco++
 				}
 			}
-			resp.Peers = filtered
+			if anyOld {
+				filtered := make([]*tailcfg.Node, 0, numDisco)
+				for _, p := range nm.Peers {
+					if !p.DiscoKey.IsZero() {
+						filtered = append(filtered, p)
+					}
+				}
+				nm.Peers = filtered
+			}
 		}
 		if Debug.StripEndpoints {
 			for _, p := range resp.Peers {
@@ -830,18 +850,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			}
 		}
 		if Debug.StripCaps {
-			resp.Node.Capabilities = nil
-		}
-
-		if pf := resp.PacketFilter; pf != nil {
-			lastParsedPacketFilter = c.parsePacketFilter(pf)
-		}
-		if c := resp.DNSConfig; c != nil {
-			lastDNSConfig = c
-		}
-
-		if v, ok := resp.CollectServices.Get(); ok {
-			collectServices = v
+			nm.SelfNode.Capabilities = nil
 		}
 
 		// Get latest localPort. This might've changed if
@@ -849,67 +858,9 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		// the end-to-end test.
 		// TODO(bradfitz): remove the NetworkMap.LocalPort field entirely.
 		c.mu.Lock()
-		localPort = c.localPort
+		nm.LocalPort = c.localPort
 		c.mu.Unlock()
 
-		nm := &netmap.NetworkMap{
-			SelfNode:        resp.Node,
-			NodeKey:         tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
-			PrivateKey:      persist.PrivateNodeKey,
-			MachineKey:      machinePubKey,
-			Expiry:          resp.Node.KeyExpiry,
-			Name:            resp.Node.Name,
-			Addresses:       resp.Node.Addresses,
-			Peers:           resp.Peers,
-			LocalPort:       localPort,
-			User:            resp.Node.User,
-			UserProfiles:    make(map[tailcfg.UserID]tailcfg.UserProfile),
-			Domain:          resp.Domain,
-			DNS:             *lastDNSConfig,
-			Hostinfo:        resp.Node.Hostinfo,
-			PacketFilter:    lastParsedPacketFilter,
-			CollectServices: collectServices,
-			DERPMap:         lastDERPMap,
-			Debug:           resp.Debug,
-		}
-		addUserProfile := func(userID tailcfg.UserID) {
-			if _, dup := nm.UserProfiles[userID]; dup {
-				// Already populated it from a previous peer.
-				return
-			}
-			if up, ok := lastUserProfile[userID]; ok {
-				nm.UserProfiles[userID] = up
-			}
-		}
-		addUserProfile(nm.User)
-		magicDNSSuffix := nm.MagicDNSSuffix()
-		nm.SelfNode.InitDisplayNames(magicDNSSuffix)
-		for _, peer := range resp.Peers {
-			peer.InitDisplayNames(magicDNSSuffix)
-			if !peer.Sharer.IsZero() {
-				if c.keepSharerAndUserSplit {
-					addUserProfile(peer.Sharer)
-				} else {
-					peer.User = peer.Sharer
-				}
-			}
-			addUserProfile(peer.User)
-		}
-		if resp.Node.MachineAuthorized {
-			nm.MachineStatus = tailcfg.MachineAuthorized
-		} else {
-			nm.MachineStatus = tailcfg.MachineUnauthorized
-		}
-		if len(resp.DNS) > 0 {
-			nm.DNS.Nameservers = resp.DNS
-		}
-		if len(resp.SearchPaths) > 0 {
-			nm.DNS.Domains = resp.SearchPaths
-		}
-		if Debug.ProxyDNS {
-			nm.DNS.Proxied = true
-		}
-
 		// 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
@@ -925,6 +876,8 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 		c.expiry = &nm.Expiry
 		c.mu.Unlock()
 
+		// log.Println("MAPRESPONSE: ", resp.Node)
+		// c.logf("MAPRESPONSE: %v", resp.Node)
 		cb(nm)
 	}
 	if ctx.Err() != nil {
@@ -1104,124 +1057,6 @@ func envBool(k string) bool {
 
 var clockNow = time.Now
 
-// undeltaPeers updates mapRes.Peers to be complete based on the
-// provided previous peer list and the PeersRemoved and PeersChanged
-// fields in mapRes, as well as the PeerSeenChange and OnlineChange
-// maps.
-//
-// It then also nils out the delta fields.
-func undeltaPeers(mapRes *tailcfg.MapResponse, prev []*tailcfg.Node) {
-	if len(mapRes.Peers) > 0 {
-		// Not delta encoded.
-		if !nodesSorted(mapRes.Peers) {
-			log.Printf("netmap: undeltaPeers: MapResponse.Peers not sorted; sorting")
-			sortNodes(mapRes.Peers)
-		}
-		return
-	}
-
-	var removed map[tailcfg.NodeID]bool
-	if pr := mapRes.PeersRemoved; len(pr) > 0 {
-		removed = make(map[tailcfg.NodeID]bool, len(pr))
-		for _, id := range pr {
-			removed[id] = true
-		}
-	}
-	changed := mapRes.PeersChanged
-
-	if !nodesSorted(changed) {
-		log.Printf("netmap: undeltaPeers: MapResponse.PeersChanged not sorted; sorting")
-		sortNodes(changed)
-	}
-	if !nodesSorted(prev) {
-		// Internal error (unrelated to the network) if we get here.
-		log.Printf("netmap: undeltaPeers: [unexpected] prev not sorted; sorting")
-		sortNodes(prev)
-	}
-
-	newFull := prev
-	if len(removed) > 0 || len(changed) > 0 {
-		newFull = make([]*tailcfg.Node, 0, len(prev)-len(removed))
-		for len(prev) > 0 && len(changed) > 0 {
-			pID := prev[0].ID
-			cID := changed[0].ID
-			if removed[pID] {
-				prev = prev[1:]
-				continue
-			}
-			switch {
-			case pID < cID:
-				newFull = append(newFull, prev[0])
-				prev = prev[1:]
-			case pID == cID:
-				newFull = append(newFull, changed[0])
-				prev, changed = prev[1:], changed[1:]
-			case cID < pID:
-				newFull = append(newFull, changed[0])
-				changed = changed[1:]
-			}
-		}
-		newFull = append(newFull, changed...)
-		for _, n := range prev {
-			if !removed[n.ID] {
-				newFull = append(newFull, n)
-			}
-		}
-		sortNodes(newFull)
-	}
-
-	if len(mapRes.PeerSeenChange) != 0 || len(mapRes.OnlineChange) != 0 {
-		peerByID := make(map[tailcfg.NodeID]*tailcfg.Node, len(newFull))
-		for _, n := range newFull {
-			peerByID[n.ID] = n
-		}
-		now := clockNow()
-		for nodeID, seen := range mapRes.PeerSeenChange {
-			if n, ok := peerByID[nodeID]; ok {
-				if seen {
-					n.LastSeen = &now
-				} else {
-					n.LastSeen = nil
-				}
-			}
-		}
-		for nodeID, online := range mapRes.OnlineChange {
-			if n, ok := peerByID[nodeID]; ok {
-				online := online
-				n.Online = &online
-			}
-		}
-	}
-
-	mapRes.Peers = newFull
-	mapRes.PeersChanged = nil
-	mapRes.PeersRemoved = nil
-}
-
-func nodesSorted(v []*tailcfg.Node) bool {
-	for i, n := range v {
-		if i > 0 && n.ID <= v[i-1].ID {
-			return false
-		}
-	}
-	return true
-}
-
-func sortNodes(v []*tailcfg.Node) {
-	sort.Slice(v, func(i, j int) bool { return v[i].ID < v[j].ID })
-}
-
-func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
-	if v1 == nil {
-		return nil
-	}
-	v2 := make([]*tailcfg.Node, len(v1))
-	for i, n := range v1 {
-		v2[i] = n.Clone()
-	}
-	return v2
-}
-
 // opt.Bool configs from control.
 var (
 	controlUseDERPRoute atomic.Value
@@ -1275,7 +1110,7 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 	localIPs := map[netaddr.IP]bool{}
 	for _, addrs := range state.InterfaceIPs {
 		for _, pfx := range addrs {
-			localIPs[pfx.IP] = true
+			localIPs[pfx.IP()] = true
 		}
 	}
 
@@ -1284,10 +1119,10 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 		// It's possible to advertise a route to one of the local
 		// machine's local IPs. IP forwarding isn't required for this
 		// to work, so we shouldn't warn for such exports.
-		if r.IsSingleIP() && localIPs[r.IP] {
+		if r.IsSingleIP() && localIPs[r.IP()] {
 			continue
 		}
-		if r.IP.Is4() {
+		if r.IP().Is4() {
 			v4Routes = true
 		} else {
 			v6Routes = true
@@ -1364,3 +1199,66 @@ func answerPing(logf logger.Logf, c *http.Client, pr *tailcfg.PingRequest) {
 		logf("answerPing complete to %v (after %v)", pr.URL, d)
 	}
 }
+
+func sleepAsRequested(ctx context.Context, logf logger.Logf, timeoutReset chan<- struct{}, d time.Duration) error {
+	const maxSleep = 5 * time.Minute
+	if d > maxSleep {
+		logf("sleeping for %v, capped from server-requested %v ...", maxSleep, d)
+		d = maxSleep
+	} else {
+		logf("sleeping for server-requested %v ...", d)
+	}
+
+	ticker := time.NewTicker(pollTimeout / 2)
+	defer ticker.Stop()
+	timer := time.NewTimer(d)
+	defer timer.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case <-timer.C:
+			return nil
+		case <-ticker.C:
+			select {
+			case timeoutReset <- struct{}{}:
+			case <-timer.C:
+				return nil
+			case <-ctx.Done():
+				return ctx.Err()
+			}
+		}
+	}
+}
+
+// Run the ping suite from this client to another one
+// Send the ping results via http to the adminhttp handlers.
+// This is where we hopefully will run the ping suite similar to CLI
+func (c *Direct) CustomPing(mr *tailcfg.MapResponse) bool {
+	log.Printf("Custom Ping Triggered with %d number of peers\n", len(mr.Peers))
+	log.Println("Ping Request: ", mr.PingRequest)
+	log.Println("ALOHA")
+	log.Println("CP PEERLIST : ", mr.Peers, mr.PeersChanged, mr.PeersRemoved, mr.PeerSeenChange)
+	if len(mr.Peers) > 0 {
+		log.Println("Peer data: ", mr.Peers[0].ID)
+	}
+	ip := mr.PingRequest.TestIP
+	log.Println("TestIP : ", ip)
+	start := time.Now()
+	// Run the ping
+	var pingRes *ipnstate.PingResult
+	for i := 1; i <= 10; i++ {
+		log.Println("Ping attempt ", i)
+		go c.pinger.Ping(ip, true, func(res *ipnstate.PingResult) {
+			log.Println("Callback", res, (res.NodeIP))
+			pingRes = res
+		})
+	}
+
+	log.Println("PINGRES", pingRes)
+	duration := time.Since(start)
+	// Send the data to the handler in api.go admin/api/ping
+	log.Printf("Ping operation took %f seconds\n", duration.Seconds())
+
+	return len(mr.Peers) > 0
+}

control/controlclient/direct_test.go

@@ -6,169 +6,13 @@ package controlclient
 
 import (
 	"encoding/json"
-	"fmt"
-	"reflect"
-	"strings"
 	"testing"
-	"time"
 
 	"inet.af/netaddr"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/wgkey"
 )
 
-func TestUndeltaPeers(t *testing.T) {
-	defer func(old func() time.Time) { clockNow = old }(clockNow)
-
-	var curTime time.Time
-	clockNow = func() time.Time {
-		return curTime
-	}
-	online := func(v bool) func(*tailcfg.Node) {
-		return func(n *tailcfg.Node) {
-			n.Online = &v
-		}
-	}
-	seenAt := func(t time.Time) func(*tailcfg.Node) {
-		return func(n *tailcfg.Node) {
-			n.LastSeen = &t
-		}
-	}
-	n := func(id tailcfg.NodeID, name string, mod ...func(*tailcfg.Node)) *tailcfg.Node {
-		n := &tailcfg.Node{ID: id, Name: name}
-		for _, f := range mod {
-			f(n)
-		}
-		return n
-	}
-	peers := func(nv ...*tailcfg.Node) []*tailcfg.Node { return nv }
-	tests := []struct {
-		name    string
-		mapRes  *tailcfg.MapResponse
-		curTime time.Time
-		prev    []*tailcfg.Node
-		want    []*tailcfg.Node
-	}{
-		{
-			name: "full_peers",
-			mapRes: &tailcfg.MapResponse{
-				Peers: peers(n(1, "foo"), n(2, "bar")),
-			},
-			want: peers(n(1, "foo"), n(2, "bar")),
-		},
-		{
-			name: "full_peers_ignores_deltas",
-			mapRes: &tailcfg.MapResponse{
-				Peers:        peers(n(1, "foo"), n(2, "bar")),
-				PeersRemoved: []tailcfg.NodeID{2},
-			},
-			want: peers(n(1, "foo"), n(2, "bar")),
-		},
-		{
-			name: "add_and_update",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersChanged: peers(n(0, "zero"), n(2, "bar2"), n(3, "three")),
-			},
-			want: peers(n(0, "zero"), n(1, "foo"), n(2, "bar2"), n(3, "three")),
-		},
-		{
-			name: "remove",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersRemoved: []tailcfg.NodeID{1},
-			},
-			want: peers(n(2, "bar")),
-		},
-		{
-			name: "add_and_remove",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				PeersChanged: peers(n(1, "foo2")),
-				PeersRemoved: []tailcfg.NodeID{2},
-			},
-			want: peers(n(1, "foo2")),
-		},
-		{
-			name:   "unchanged",
-			prev:   peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{},
-			want:   peers(n(1, "foo"), n(2, "bar")),
-		},
-		{
-			name: "online_change",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				OnlineChange: map[tailcfg.NodeID]bool{
-					1: true,
-				},
-			},
-			want: peers(
-				n(1, "foo", online(true)),
-				n(2, "bar"),
-			),
-		},
-		{
-			name: "online_change_offline",
-			prev: peers(n(1, "foo"), n(2, "bar")),
-			mapRes: &tailcfg.MapResponse{
-				OnlineChange: map[tailcfg.NodeID]bool{
-					1: false,
-					2: true,
-				},
-			},
-			want: peers(
-				n(1, "foo", online(false)),
-				n(2, "bar", online(true)),
-			),
-		},
-		{
-			name:    "peer_seen_at",
-			prev:    peers(n(1, "foo", seenAt(time.Unix(111, 0))), n(2, "bar")),
-			curTime: time.Unix(123, 0),
-			mapRes: &tailcfg.MapResponse{
-				PeerSeenChange: map[tailcfg.NodeID]bool{
-					1: false,
-					2: true,
-				},
-			},
-			want: peers(
-				n(1, "foo"),
-				n(2, "bar", seenAt(time.Unix(123, 0))),
-			),
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			if !tt.curTime.IsZero() {
-				curTime = tt.curTime
-			}
-			undeltaPeers(tt.mapRes, tt.prev)
-			if !reflect.DeepEqual(tt.mapRes.Peers, tt.want) {
-				t.Errorf("wrong results\n got: %s\nwant: %s", formatNodes(tt.mapRes.Peers), formatNodes(tt.want))
-			}
-		})
-	}
-}
-
-func formatNodes(nodes []*tailcfg.Node) string {
-	var sb strings.Builder
-	for i, n := range nodes {
-		if i > 0 {
-			sb.WriteString(", ")
-		}
-		var extra string
-		if n.Online != nil {
-			extra += fmt.Sprintf(", online=%v", *n.Online)
-		}
-		if n.LastSeen != nil {
-			extra += fmt.Sprintf(", lastSeen=%v", n.LastSeen.Unix())
-		}
-		fmt.Fprintf(&sb, "(%d, %q%s)", n.ID, n.Name, extra)
-	}
-	return sb.String()
-}
-
 func TestNewDirect(t *testing.T) {
 	hi := NewHostinfo()
 	ni := tailcfg.NetInfo{LinkType: "wired"}
@@ -242,7 +86,7 @@ func TestNewDirect(t *testing.T) {
 func fakeEndpoints(ports ...uint16) (ret []tailcfg.Endpoint) {
 	for _, port := range ports {
 		ret = append(ret, tailcfg.Endpoint{
-			Addr: netaddr.IPPort{Port: port},
+			Addr: netaddr.IPPortFrom(netaddr.IP{}, port),
 		})
 	}
 	return
@@ -259,3 +103,37 @@ func TestNewHostinfo(t *testing.T) {
 	}
 	t.Logf("Got: %s", j)
 }
+
+// Currently not working properly
+func TestPingFromMapResponse(t *testing.T) {
+	hi := NewHostinfo()
+	ni := tailcfg.NetInfo{LinkType: "wired"}
+	hi.NetInfo = &ni
+
+	key, err := wgkey.NewPrivate()
+	if err != nil {
+		t.Error(err)
+	}
+	opts := Options{
+		ServerURL: "https://example.com",
+		Hostinfo:  hi,
+		GetMachinePrivateKey: func() (wgkey.Private, error) {
+			return key, nil
+		},
+	}
+	c, err := NewDirect(opts)
+	if c == nil || err != nil {
+		t.Errorf("Direct not created %w", err)
+	}
+	peers := []*tailcfg.Node{
+		{ID: 1},
+		{ID: 2},
+		{ID: 3},
+	}
+	pingRequest := tailcfg.PingRequest{URL: "localhost:3040", Log: true, PayloadSize: 10}
+	mr := &tailcfg.MapResponse{Peers: peers, Domain: "DumbTest", PingRequest: &pingRequest}
+	if !c.CustomPing(mr) {
+		t.Errorf("Custom ping failed!\n")
+	}
+	t.Log("Successful ping")
+}

control/controlclient/filter.go

@@ -1,20 +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 controlclient
-
-import (
-	"tailscale.com/tailcfg"
-	"tailscale.com/wgengine/filter"
-)
-
-// Parse a backward-compatible FilterRule used by control's wire
-// format, producing the most current filter format.
-func (c *Direct) parsePacketFilter(pf []tailcfg.FilterRule) []filter.Match {
-	mm, err := filter.MatchesFromFilterRules(pf)
-	if err != nil {
-		c.logf("parsePacketFilter: %s\n", err)
-	}
-	return mm
-}

control/controlclient/map.go

@@ -0,0 +1,282 @@
+// Copyright (c) 2021 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 (
+	"log"
+	"sort"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/wgkey"
+	"tailscale.com/wgengine/filter"
+)
+
+// mapSession holds the state over a long-polled "map" request to the
+// control plane.
+//
+// It accepts incremental tailcfg.MapResponse values to
+// netMapForResponse and returns fully inflated NetworkMaps, filling
+// in the omitted data implicit from prior MapResponse values from
+// within the same session (the same long-poll HTTP response to the
+// one MapRequest).
+type mapSession struct {
+	// Immutable fields.
+	privateNodeKey         wgkey.Private
+	logf                   logger.Logf
+	vlogf                  logger.Logf
+	machinePubKey          tailcfg.MachineKey
+	keepSharerAndUserSplit bool // see Options.KeepSharerAndUserSplit
+
+	// Fields storing state over the the coards of multiple MapResponses.
+	lastNode               *tailcfg.Node
+	lastDNSConfig          *tailcfg.DNSConfig
+	lastDERPMap            *tailcfg.DERPMap
+	lastUserProfile        map[tailcfg.UserID]tailcfg.UserProfile
+	lastParsedPacketFilter []filter.Match
+	collectServices        bool
+	previousPeers          []*tailcfg.Node // for delta-purposes
+	lastDomain             string
+
+	// netMapBuilding is non-nil during a netmapForResponse call,
+	// containing the value to be returned, once fully populated.
+	netMapBuilding *netmap.NetworkMap
+}
+
+func newMapSession(privateNodeKey wgkey.Private) *mapSession {
+	ms := &mapSession{
+		privateNodeKey:  privateNodeKey,
+		logf:            logger.Discard,
+		vlogf:           logger.Discard,
+		lastDNSConfig:   new(tailcfg.DNSConfig),
+		lastUserProfile: map[tailcfg.UserID]tailcfg.UserProfile{},
+	}
+	return ms
+}
+
+func (ms *mapSession) addUserProfile(userID tailcfg.UserID) {
+	nm := ms.netMapBuilding
+	if _, dup := nm.UserProfiles[userID]; dup {
+		// Already populated it from a previous peer.
+		return
+	}
+	if up, ok := ms.lastUserProfile[userID]; ok {
+		nm.UserProfiles[userID] = up
+	}
+}
+
+// netmapForResponse returns a fully populated NetworkMap from a full
+// or incremental MapResponse within the session, filling in omitted
+// information from prior MapResponse values.
+func (ms *mapSession) netmapForResponse(resp *tailcfg.MapResponse) *netmap.NetworkMap {
+	undeltaPeers(resp, ms.previousPeers)
+
+	ms.previousPeers = cloneNodes(resp.Peers) // defensive/lazy clone, since this escapes to who knows where
+	for _, up := range resp.UserProfiles {
+		ms.lastUserProfile[up.ID] = up
+	}
+
+	if resp.DERPMap != nil {
+		ms.vlogf("netmap: new map contains DERP map")
+		ms.lastDERPMap = resp.DERPMap
+	}
+
+	if pf := resp.PacketFilter; pf != nil {
+		var err error
+		ms.lastParsedPacketFilter, err = filter.MatchesFromFilterRules(pf)
+		if err != nil {
+			ms.logf("parsePacketFilter: %v", err)
+		}
+	}
+	if c := resp.DNSConfig; c != nil {
+		ms.lastDNSConfig = c
+	}
+
+	if v, ok := resp.CollectServices.Get(); ok {
+		ms.collectServices = v
+	}
+	if resp.Domain != "" {
+		ms.lastDomain = resp.Domain
+	}
+
+	nm := &netmap.NetworkMap{
+		NodeKey:         tailcfg.NodeKey(ms.privateNodeKey.Public()),
+		PrivateKey:      ms.privateNodeKey,
+		MachineKey:      ms.machinePubKey,
+		Peers:           resp.Peers,
+		UserProfiles:    make(map[tailcfg.UserID]tailcfg.UserProfile),
+		Domain:          ms.lastDomain,
+		DNS:             *ms.lastDNSConfig,
+		PacketFilter:    ms.lastParsedPacketFilter,
+		CollectServices: ms.collectServices,
+		DERPMap:         ms.lastDERPMap,
+		Debug:           resp.Debug,
+	}
+	ms.netMapBuilding = nm
+
+	if resp.Node != nil {
+		ms.lastNode = resp.Node
+	}
+	if node := ms.lastNode.Clone(); node != nil {
+		nm.SelfNode = node
+		nm.Expiry = node.KeyExpiry
+		nm.Name = node.Name
+		nm.Addresses = node.Addresses
+		nm.User = node.User
+		nm.Hostinfo = node.Hostinfo
+		if node.MachineAuthorized {
+			nm.MachineStatus = tailcfg.MachineAuthorized
+		} else {
+			nm.MachineStatus = tailcfg.MachineUnauthorized
+		}
+	}
+
+	ms.addUserProfile(nm.User)
+	magicDNSSuffix := nm.MagicDNSSuffix()
+	if nm.SelfNode != nil {
+		nm.SelfNode.InitDisplayNames(magicDNSSuffix)
+	}
+	for _, peer := range resp.Peers {
+		peer.InitDisplayNames(magicDNSSuffix)
+		if !peer.Sharer.IsZero() {
+			if ms.keepSharerAndUserSplit {
+				ms.addUserProfile(peer.Sharer)
+			} else {
+				peer.User = peer.Sharer
+			}
+		}
+		ms.addUserProfile(peer.User)
+	}
+	if len(resp.DNS) > 0 {
+		nm.DNS.Nameservers = resp.DNS
+	}
+	if len(resp.SearchPaths) > 0 {
+		nm.DNS.Domains = resp.SearchPaths
+	}
+	if Debug.ProxyDNS {
+		nm.DNS.Proxied = true
+	}
+	ms.netMapBuilding = nil
+	return nm
+}
+
+// undeltaPeers updates mapRes.Peers to be complete based on the
+// provided previous peer list and the PeersRemoved and PeersChanged
+// fields in mapRes, as well as the PeerSeenChange and OnlineChange
+// maps.
+//
+// It then also nils out the delta fields.
+func undeltaPeers(mapRes *tailcfg.MapResponse, prev []*tailcfg.Node) {
+	if len(mapRes.Peers) > 0 {
+		// Not delta encoded.
+		if !nodesSorted(mapRes.Peers) {
+			log.Printf("netmap: undeltaPeers: MapResponse.Peers not sorted; sorting")
+			sortNodes(mapRes.Peers)
+		}
+		return
+	}
+
+	var removed map[tailcfg.NodeID]bool
+	if pr := mapRes.PeersRemoved; len(pr) > 0 {
+		removed = make(map[tailcfg.NodeID]bool, len(pr))
+		for _, id := range pr {
+			removed[id] = true
+		}
+	}
+	changed := mapRes.PeersChanged
+
+	if !nodesSorted(changed) {
+		log.Printf("netmap: undeltaPeers: MapResponse.PeersChanged not sorted; sorting")
+		sortNodes(changed)
+	}
+	if !nodesSorted(prev) {
+		// Internal error (unrelated to the network) if we get here.
+		log.Printf("netmap: undeltaPeers: [unexpected] prev not sorted; sorting")
+		sortNodes(prev)
+	}
+
+	newFull := prev
+	if len(removed) > 0 || len(changed) > 0 {
+		newFull = make([]*tailcfg.Node, 0, len(prev)-len(removed))
+		for len(prev) > 0 && len(changed) > 0 {
+			pID := prev[0].ID
+			cID := changed[0].ID
+			if removed[pID] {
+				prev = prev[1:]
+				continue
+			}
+			switch {
+			case pID < cID:
+				newFull = append(newFull, prev[0])
+				prev = prev[1:]
+			case pID == cID:
+				newFull = append(newFull, changed[0])
+				prev, changed = prev[1:], changed[1:]
+			case cID < pID:
+				newFull = append(newFull, changed[0])
+				changed = changed[1:]
+			}
+		}
+		newFull = append(newFull, changed...)
+		for _, n := range prev {
+			if !removed[n.ID] {
+				newFull = append(newFull, n)
+			}
+		}
+		sortNodes(newFull)
+	}
+
+	if len(mapRes.PeerSeenChange) != 0 || len(mapRes.OnlineChange) != 0 {
+		peerByID := make(map[tailcfg.NodeID]*tailcfg.Node, len(newFull))
+		for _, n := range newFull {
+			peerByID[n.ID] = n
+		}
+		now := clockNow()
+		for nodeID, seen := range mapRes.PeerSeenChange {
+			if n, ok := peerByID[nodeID]; ok {
+				if seen {
+					n.LastSeen = &now
+				} else {
+					n.LastSeen = nil
+				}
+			}
+		}
+		for nodeID, online := range mapRes.OnlineChange {
+			if n, ok := peerByID[nodeID]; ok {
+				online := online
+				n.Online = &online
+			}
+		}
+	}
+
+	mapRes.Peers = newFull
+	mapRes.PeersChanged = nil
+	mapRes.PeersRemoved = nil
+}
+
+func nodesSorted(v []*tailcfg.Node) bool {
+	for i, n := range v {
+		if i > 0 && n.ID <= v[i-1].ID {
+			return false
+		}
+	}
+	return true
+}
+
+func sortNodes(v []*tailcfg.Node) {
+	sort.Slice(v, func(i, j int) bool { return v[i].ID < v[j].ID })
+}
+
+func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
+	if v1 == nil {
+		return nil
+	}
+	v2 := make([]*tailcfg.Node, len(v1))
+	for i, n := range v1 {
+		v2[i] = n.Clone()
+	}
+	return v2
+}

control/controlclient/map_test.go

@@ -0,0 +1,311 @@
+// Copyright (c) 2021 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 (
+	"encoding/json"
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/wgkey"
+)
+
+func TestUndeltaPeers(t *testing.T) {
+	defer func(old func() time.Time) { clockNow = old }(clockNow)
+
+	var curTime time.Time
+	clockNow = func() time.Time {
+		return curTime
+	}
+	online := func(v bool) func(*tailcfg.Node) {
+		return func(n *tailcfg.Node) {
+			n.Online = &v
+		}
+	}
+	seenAt := func(t time.Time) func(*tailcfg.Node) {
+		return func(n *tailcfg.Node) {
+			n.LastSeen = &t
+		}
+	}
+	n := func(id tailcfg.NodeID, name string, mod ...func(*tailcfg.Node)) *tailcfg.Node {
+		n := &tailcfg.Node{ID: id, Name: name}
+		for _, f := range mod {
+			f(n)
+		}
+		return n
+	}
+	peers := func(nv ...*tailcfg.Node) []*tailcfg.Node { return nv }
+	tests := []struct {
+		name    string
+		mapRes  *tailcfg.MapResponse
+		curTime time.Time
+		prev    []*tailcfg.Node
+		want    []*tailcfg.Node
+	}{
+		{
+			name: "full_peers",
+			mapRes: &tailcfg.MapResponse{
+				Peers: peers(n(1, "foo"), n(2, "bar")),
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "full_peers_ignores_deltas",
+			mapRes: &tailcfg.MapResponse{
+				Peers:        peers(n(1, "foo"), n(2, "bar")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "add_and_update",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(0, "zero"), n(2, "bar2"), n(3, "three")),
+			},
+			want: peers(n(0, "zero"), n(1, "foo"), n(2, "bar2"), n(3, "three")),
+		},
+		{
+			name: "remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersRemoved: []tailcfg.NodeID{1},
+			},
+			want: peers(n(2, "bar")),
+		},
+		{
+			name: "add_and_remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(1, "foo2")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo2")),
+		},
+		{
+			name:   "unchanged",
+			prev:   peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{},
+			want:   peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "online_change",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				OnlineChange: map[tailcfg.NodeID]bool{
+					1: true,
+				},
+			},
+			want: peers(
+				n(1, "foo", online(true)),
+				n(2, "bar"),
+			),
+		},
+		{
+			name: "online_change_offline",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				OnlineChange: map[tailcfg.NodeID]bool{
+					1: false,
+					2: true,
+				},
+			},
+			want: peers(
+				n(1, "foo", online(false)),
+				n(2, "bar", online(true)),
+			),
+		},
+		{
+			name:    "peer_seen_at",
+			prev:    peers(n(1, "foo", seenAt(time.Unix(111, 0))), n(2, "bar")),
+			curTime: time.Unix(123, 0),
+			mapRes: &tailcfg.MapResponse{
+				PeerSeenChange: map[tailcfg.NodeID]bool{
+					1: false,
+					2: true,
+				},
+			},
+			want: peers(
+				n(1, "foo"),
+				n(2, "bar", seenAt(time.Unix(123, 0))),
+			),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if !tt.curTime.IsZero() {
+				curTime = tt.curTime
+			}
+			undeltaPeers(tt.mapRes, tt.prev)
+			if !reflect.DeepEqual(tt.mapRes.Peers, tt.want) {
+				t.Errorf("wrong results\n got: %s\nwant: %s", formatNodes(tt.mapRes.Peers), formatNodes(tt.want))
+			}
+		})
+	}
+}
+
+func formatNodes(nodes []*tailcfg.Node) string {
+	var sb strings.Builder
+	for i, n := range nodes {
+		if i > 0 {
+			sb.WriteString(", ")
+		}
+		var extra string
+		if n.Online != nil {
+			extra += fmt.Sprintf(", online=%v", *n.Online)
+		}
+		if n.LastSeen != nil {
+			extra += fmt.Sprintf(", lastSeen=%v", n.LastSeen.Unix())
+		}
+		fmt.Fprintf(&sb, "(%d, %q%s)", n.ID, n.Name, extra)
+	}
+	return sb.String()
+}
+
+func newTestMapSession(t *testing.T) *mapSession {
+	k, err := wgkey.NewPrivate()
+	if err != nil {
+		t.Fatal(err)
+	}
+	return newMapSession(k)
+}
+
+func TestNetmapForResponse(t *testing.T) {
+	t.Run("implicit_packetfilter", func(t *testing.T) {
+		somePacketFilter := []tailcfg.FilterRule{
+			{
+				SrcIPs: []string{"*"},
+				DstPorts: []tailcfg.NetPortRange{
+					{IP: "10.2.3.4", Ports: tailcfg.PortRange{First: 22, Last: 22}},
+				},
+			},
+		}
+		ms := newTestMapSession(t)
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:         new(tailcfg.Node),
+			PacketFilter: somePacketFilter,
+		})
+		if len(nm1.PacketFilter) == 0 {
+			t.Fatalf("zero length PacketFilter")
+		}
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:         new(tailcfg.Node),
+			PacketFilter: nil, // testing that the server can omit this.
+		})
+		if len(nm1.PacketFilter) == 0 {
+			t.Fatalf("zero length PacketFilter in 2nd netmap")
+		}
+		if !reflect.DeepEqual(nm1.PacketFilter, nm2.PacketFilter) {
+			t.Error("packet filters differ")
+		}
+	})
+	t.Run("implicit_dnsconfig", func(t *testing.T) {
+		someDNSConfig := &tailcfg.DNSConfig{Domains: []string{"foo", "bar"}}
+		ms := newTestMapSession(t)
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:      new(tailcfg.Node),
+			DNSConfig: someDNSConfig,
+		})
+		if !reflect.DeepEqual(nm1.DNS, *someDNSConfig) {
+			t.Fatalf("1st DNS wrong")
+		}
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:      new(tailcfg.Node),
+			DNSConfig: nil, // implict
+		})
+		if !reflect.DeepEqual(nm2.DNS, *someDNSConfig) {
+			t.Fatalf("2nd DNS wrong")
+		}
+	})
+	t.Run("collect_services", func(t *testing.T) {
+		ms := newTestMapSession(t)
+		var nm *netmap.NetworkMap
+		wantCollect := func(v bool) {
+			t.Helper()
+			if nm.CollectServices != v {
+				t.Errorf("netmap.CollectServices = %v; want %v", nm.CollectServices, v)
+			}
+		}
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: new(tailcfg.Node),
+		})
+		wantCollect(false)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "false",
+		})
+		wantCollect(false)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "true",
+		})
+		wantCollect(true)
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:            new(tailcfg.Node),
+			CollectServices: "",
+		})
+		wantCollect(true)
+	})
+	t.Run("implicit_domain", func(t *testing.T) {
+		ms := newTestMapSession(t)
+		var nm *netmap.NetworkMap
+		want := func(v string) {
+			t.Helper()
+			if nm.Domain != v {
+				t.Errorf("netmap.Domain = %q; want %q", nm.Domain, v)
+			}
+		}
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node:   new(tailcfg.Node),
+			Domain: "foo.com",
+		})
+		want("foo.com")
+
+		nm = ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: new(tailcfg.Node),
+		})
+		want("foo.com")
+	})
+	t.Run("implicit_node", func(t *testing.T) {
+		someNode := &tailcfg.Node{
+			Name: "foo",
+		}
+		wantNode := &tailcfg.Node{
+			Name:                 "foo",
+			ComputedName:         "foo",
+			ComputedNameWithHost: "foo",
+		}
+		ms := newTestMapSession(t)
+
+		nm1 := ms.netmapForResponse(&tailcfg.MapResponse{
+			Node: someNode,
+		})
+		if nm1.SelfNode == nil {
+			t.Fatal("nil Node in 1st netmap")
+		}
+		if !reflect.DeepEqual(nm1.SelfNode, wantNode) {
+			j, _ := json.Marshal(nm1.SelfNode)
+			t.Errorf("Node mismatch in 1st netmap; got: %s", j)
+		}
+
+		nm2 := ms.netmapForResponse(&tailcfg.MapResponse{})
+		if nm2.SelfNode == nil {
+			t.Fatal("nil Node in 1st netmap")
+		}
+		if !reflect.DeepEqual(nm2.SelfNode, wantNode) {
+			j, _ := json.Marshal(nm2.SelfNode)
+			t.Errorf("Node mismatch in 2nd netmap; got: %s", j)
+		}
+	})
+}

control/controlclient/sign_supported.go

@@ -18,7 +18,7 @@ import (
 	"fmt"
 	"sync"
 
-	"github.com/github/certstore"
+	"github.com/tailscale/certstore"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/wgkey"
 	"tailscale.com/util/winutil"

control/controlclient/status.go

@@ -0,0 +1,103 @@
+// 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 (
+	"encoding/json"
+	"fmt"
+	"reflect"
+
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/empty"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/persist"
+	"tailscale.com/types/structs"
+)
+
+// 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 the state, as it's now obsolete.
+//
+// apenwarr: Historical note: controlclient.Auto was originally
+// intended to be the state machine for the whole tailscale client, but that
+// turned out to not be the right abstraction layer, and it moved to
+// ipn.Backend. Since ipn.Backend now has a state machine, it would be
+// much better if controlclient could be a simple stateless API. But the
+// current server-side API (two interlocking polling https calls) makes that
+// very hard to implement. A server side API change could untangle this and
+// remove all the statefulness.
+type State int
+
+const (
+	StateNew = State(iota)
+	StateNotAuthenticated
+	StateAuthenticating
+	StateURLVisitRequired
+	StateAuthenticated
+	StateSynchronized // connected and received map update
+)
+
+func (s State) MarshalText() ([]byte, error) {
+	return []byte(s.String()), nil
+}
+
+func (s State) String() string {
+	switch s {
+	case StateNew:
+		return "state:new"
+	case StateNotAuthenticated:
+		return "state:not-authenticated"
+	case StateAuthenticating:
+		return "state:authenticating"
+	case StateURLVisitRequired:
+		return "state:url-visit-required"
+	case StateAuthenticated:
+		return "state:authenticated"
+	case StateSynchronized:
+		return "state:synchronized"
+	default:
+		return fmt.Sprintf("state:unknown:%d", int(s))
+	}
+}
+
+type Status struct {
+	_             structs.Incomparable
+	LoginFinished *empty.Message // nonempty when login finishes
+	Err           string
+	URL           string             // interactive URL to visit to finish logging in
+	NetMap        *netmap.NetworkMap // server-pushed configuration
+
+	// The internal state should not be exposed outside this
+	// package, but we have some automated tests elsewhere that need to
+	// use them. Please don't use these fields.
+	// TODO(apenwarr): Unexport or remove these.
+	State    State
+	Persist  *persist.Persist  // locally persisted configuration
+	Hostinfo *tailcfg.Hostinfo // current Hostinfo data
+}
+
+// Equal reports whether s and s2 are equal.
+func (s *Status) Equal(s2 *Status) bool {
+	if s == nil && s2 == nil {
+		return true
+	}
+	return s != nil && s2 != nil &&
+		(s.LoginFinished == nil) == (s2.LoginFinished == nil) &&
+		s.Err == s2.Err &&
+		s.URL == s2.URL &&
+		reflect.DeepEqual(s.Persist, s2.Persist) &&
+		reflect.DeepEqual(s.NetMap, s2.NetMap) &&
+		reflect.DeepEqual(s.Hostinfo, s2.Hostinfo) &&
+		s.State == s2.State
+}
+
+func (s Status) String() string {
+	b, err := json.MarshalIndent(s, "", "\t")
+	if err != nil {
+		panic(err)
+	}
+	return s.State.String() + " " + string(b)
+}

derp/derpmap/derpmap.go

@@ -83,6 +83,9 @@ func Prod() *tailcfg.DERPMap {
 			10: derpRegion(10, "sea", "Seattle",
 				derpNode("a", "137.220.36.168", "2001:19f0:8001:2d9:5400:2ff:feef:bbb1"),
 			),
+			11: derpRegion(11, "sao", "São Paulo",
+				derpNode("a", "18.230.97.74", "2600:1f1e:ee4:5611:ec5c:1736:d43b:a454"),
+			),
 		},
 	}
 }

disco/disco.go

@@ -147,9 +147,9 @@ const epLength = 16 + 2 // 16 byte IP address + 2 byte port
 func (m *CallMeMaybe) AppendMarshal(b []byte) []byte {
 	ret, p := appendMsgHeader(b, TypeCallMeMaybe, v0, epLength*len(m.MyNumber))
 	for _, ipp := range m.MyNumber {
-		a := ipp.IP.As16()
+		a := ipp.IP().As16()
 		copy(p[:], a[:])
-		binary.BigEndian.PutUint16(p[16:], ipp.Port)
+		binary.BigEndian.PutUint16(p[16:], ipp.Port())
 		p = p[epLength:]
 	}
 	return ret
@@ -164,10 +164,9 @@ func parseCallMeMaybe(ver uint8, p []byte) (m *CallMeMaybe, err error) {
 	for len(p) > 0 {
 		var a [16]byte
 		copy(a[:], p)
-		m.MyNumber = append(m.MyNumber, netaddr.IPPort{
-			IP:   netaddr.IPFrom16(a),
-			Port: binary.BigEndian.Uint16(p[16:18]),
-		})
+		m.MyNumber = append(m.MyNumber, netaddr.IPPortFrom(
+			netaddr.IPFrom16(a),
+			binary.BigEndian.Uint16(p[16:18])))
 		p = p[epLength:]
 	}
 	return m, nil
@@ -187,9 +186,9 @@ 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()
+	ip16 := m.Src.IP().As16()
 	d = d[copy(d, ip16[:]):]
-	binary.BigEndian.PutUint16(d, m.Src.Port)
+	binary.BigEndian.PutUint16(d, m.Src.Port())
 	return ret
 }
 
@@ -201,10 +200,10 @@ func parsePong(ver uint8, p []byte) (m *Pong, err error) {
 	copy(m.TxID[:], p)
 	p = p[12:]
 
-	m.Src.IP, _ = netaddr.FromStdIP(net.IP(p[:16]))
+	srcIP, _ := netaddr.FromStdIP(net.IP(p[:16]))
 	p = p[16:]
-
-	m.Src.Port = binary.BigEndian.Uint16(p)
+	port := binary.BigEndian.Uint16(p)
+	m.Src = netaddr.IPPortFrom(srcIP, port)
 	return m, nil
 }
 

go.mod

@@ -3,45 +3,44 @@ module tailscale.com
 go 1.16
 
 require (
-	github.com/alexbrainman/sspi v0.0.0-20180613141037-e580b900e9f5
-	github.com/anmitsu/go-shlex v0.0.0-20161002113705-648efa622239 // indirect
-	github.com/coreos/go-iptables v0.4.5
-	github.com/flynn/go-shlex v0.0.0-20150515145356-3f9db97f8568 // indirect
-	github.com/github/certstore v0.1.0
-	github.com/gliderlabs/ssh v0.2.2
+	github.com/alexbrainman/sspi v0.0.0-20210105120005-909beea2cc74
+	github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be // indirect
+	github.com/coreos/go-iptables v0.6.0
+	github.com/frankban/quicktest v1.13.0
+	github.com/tailscale/certstore v0.0.0-20210528134328-066c94b793d3
+	github.com/gliderlabs/ssh v0.3.2
 	github.com/go-multierror/multierror v1.0.2
-	github.com/go-ole/go-ole v1.2.4
-	github.com/godbus/dbus/v5 v5.0.3
-	github.com/google/go-cmp v0.5.4
-	github.com/goreleaser/nfpm v1.1.10
-	github.com/jsimonetti/rtnetlink v0.0.0-20210212075122-66c871082f2b
-	github.com/klauspost/compress v1.10.10
+	github.com/go-ole/go-ole v1.2.5
+	github.com/godbus/dbus/v5 v5.0.4
+	github.com/google/go-cmp v0.5.5
+	github.com/google/goexpect v0.0.0-20210430020637-ab937bf7fd6f
+	github.com/goreleaser/nfpm v1.10.3
+	github.com/jsimonetti/rtnetlink v0.0.0-20210409061457-9561dc9288a7
+	github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51
+	github.com/klauspost/compress v1.12.2
 	github.com/kr/pty v1.1.8
-	github.com/mdlayher/netlink v1.3.2
-	github.com/mdlayher/sdnotify v0.0.0-20200625151349-e4a4f32afc4a
-	github.com/miekg/dns v1.1.30
-	github.com/pborman/getopt v0.0.0-20190409184431-ee0cd42419d3
+	github.com/mdlayher/netlink v1.4.0
+	github.com/mdlayher/sdnotify v0.0.0-20210228150836-ea3ec207d697
+	github.com/miekg/dns v1.1.42
+	github.com/pborman/getopt v1.1.0
 	github.com/peterbourgon/ff/v2 v2.0.0
-	github.com/pkg/errors v0.9.1 // indirect
 	github.com/tailscale/depaware v0.0.0-20201214215404-77d1e9757027
-	github.com/tailscale/wireguard-go v0.0.0-20210403171604-17614717a9b5
 	github.com/tcnksm/go-httpstat v0.2.0
 	github.com/toqueteos/webbrowser v1.2.0
 	go4.org/mem v0.0.0-20201119185036-c04c5a6ff174
-	golang.org/x/crypto v0.0.0-20210317152858-513c2a44f670
-	golang.org/x/net v0.0.0-20210226172049-e18ecbb05110
+	golang.org/x/crypto v0.0.0-20210513164829-c07d793c2f9a
+	golang.org/x/net v0.0.0-20210525063256-abc453219eb5
 	golang.org/x/sync v0.0.0-20210220032951-036812b2e83c
-	golang.org/x/sys v0.0.0-20210403161142-5e06dd20ab57
-	golang.org/x/term v0.0.0-20210317153231-de623e64d2a6
+	golang.org/x/sys v0.0.0-20210525143221-35b2ab0089ea
+	golang.org/x/term v0.0.0-20210503060354-a79de5458b56
 	golang.org/x/time v0.0.0-20210220033141-f8bda1e9f3ba
-	golang.org/x/tools v0.0.0-20201211185031-d93e913c1a58
-	golang.zx2c4.com/wireguard/windows v0.1.2-0.20201113162609-9b85be97fdf8
-	gopkg.in/yaml.v2 v2.2.8 // indirect
-	honnef.co/go/tools v0.1.0
-	inet.af/netaddr v0.0.0-20210222205655-a1ec2b7b8c44
+	golang.org/x/tools v0.1.2
+	golang.zx2c4.com/wireguard v0.0.0-20210525143454-64cb82f2b3f5
+	golang.zx2c4.com/wireguard/windows v0.3.15-0.20210525143335-94c0476d63e3
+	honnef.co/go/tools v0.1.4
+	inet.af/netaddr v0.0.0-20210523191804-d57edf19c517
 	inet.af/netstack v0.0.0-20210317161235-a1bf4e56ef22
-	inet.af/peercred v0.0.0-20210302202138-56e694897155
+	inet.af/peercred v0.0.0-20210318190834-4259e17bb763
+	inet.af/wf v0.0.0-20210516214145-a5343001b756
 	rsc.io/goversion v1.2.0
 )
-
-replace github.com/github/certstore => github.com/cyolosecurity/certstore v0.0.0-20200922073901-ece7f1d353c2

health/health.go

@@ -11,6 +11,7 @@ import (
 	"fmt"
 	"sort"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/go-multierror/multierror"
@@ -36,6 +37,7 @@ var (
 	ipnState                string
 	ipnWantRunning          bool
 	anyInterfaceUp          = true // until told otherwise
+	udp4Unbound             bool
 )
 
 // Subsystem is the name of a subsystem whose health can be monitored.
@@ -46,7 +48,7 @@ const (
 	// the system, rather than one particular subsystem.
 	SysOverall = Subsystem("overall")
 
-	// SysRouter is the name the wgengine/router subsystem.
+	// SysRouter is the name of the wgengine/router subsystem.
 	SysRouter = Subsystem("router")
 
 	// SysDNS is the name of the net/dns subsystem.
@@ -213,9 +215,18 @@ func SetAnyInterfaceUp(up bool) {
 	selfCheckLocked()
 }
 
+// SetUDP4Unbound sets whether the udp4 bind failed completely.
+func SetUDP4Unbound(unbound bool) {
+	mu.Lock()
+	defer mu.Unlock()
+	udp4Unbound = unbound
+	selfCheckLocked()
+}
+
 func timerSelfCheck() {
 	mu.Lock()
 	defer mu.Unlock()
+	checkReceiveFuncs()
 	selfCheckLocked()
 	if timer != nil {
 		timer.Reset(time.Minute)
@@ -255,6 +266,9 @@ func overallErrorLocked() error {
 	if d := now.Sub(derpRegionLastFrame[rid]).Round(time.Second); d > tooIdle {
 		return fmt.Errorf("haven't heard from home DERP region %v in %v", rid, d)
 	}
+	if udp4Unbound {
+		return errors.New("no udp4 bind")
+	}
 
 	// TODO: use
 	_ = inMapPollSince
@@ -263,6 +277,11 @@ func overallErrorLocked() error {
 	_ = lastMapRequestHeard
 
 	var errs []error
+	for _, recv := range receiveFuncs {
+		if recv.missing {
+			errs = append(errs, fmt.Errorf("%s is not running", recv.name))
+		}
+	}
 	for sys, err := range sysErr {
 		if err == nil || sys == SysOverall {
 			continue
@@ -275,3 +294,58 @@ func overallErrorLocked() error {
 	})
 	return multierror.New(errs)
 }
+
+var (
+	ReceiveIPv4 = ReceiveFuncStats{name: "ReceiveIPv4"}
+	ReceiveIPv6 = ReceiveFuncStats{name: "ReceiveIPv6"}
+	ReceiveDERP = ReceiveFuncStats{name: "ReceiveDERP"}
+
+	receiveFuncs = []*ReceiveFuncStats{&ReceiveIPv4, &ReceiveIPv6, &ReceiveDERP}
+)
+
+// ReceiveFuncStats tracks the calls made to a wireguard-go receive func.
+type ReceiveFuncStats struct {
+	// name is the name of the receive func.
+	name string
+	// numCalls is the number of times the receive func has ever been called.
+	// It is required because it is possible for a receive func's wireguard-go goroutine
+	// to be active even though the receive func isn't.
+	// The wireguard-go goroutine alternates between calling the receive func and
+	// processing what the func returned.
+	numCalls uint64 // accessed atomically
+	// prevNumCalls is the value of numCalls last time the health check examined it.
+	prevNumCalls uint64
+	// inCall indicates whether the receive func is currently running.
+	inCall uint32 // bool, accessed atomically
+	// missing indicates whether the receive func is not running.
+	missing bool
+}
+
+func (s *ReceiveFuncStats) Enter() {
+	atomic.AddUint64(&s.numCalls, 1)
+	atomic.StoreUint32(&s.inCall, 1)
+}
+
+func (s *ReceiveFuncStats) Exit() {
+	atomic.StoreUint32(&s.inCall, 0)
+}
+
+func checkReceiveFuncs() {
+	for _, recv := range receiveFuncs {
+		recv.missing = false
+		prev := recv.prevNumCalls
+		numCalls := atomic.LoadUint64(&recv.numCalls)
+		recv.prevNumCalls = numCalls
+		if numCalls > prev {
+			// OK: the function has gotten called since last we checked
+			continue
+		}
+		if atomic.LoadUint32(&recv.inCall) == 1 {
+			// OK: the function is active, probably blocked due to inactivity
+			continue
+		}
+		// Not OK: The function is not active, and not accumulating new calls.
+		// It is probably MIA.
+		recv.missing = true
+	}
+}

internal/deephash/deephash.go

@@ -0,0 +1,226 @@
+// 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 deephash hashes a Go value recursively, in a predictable
+// order, without looping.
+package deephash
+
+import (
+	"bufio"
+	"crypto/sha256"
+	"encoding/hex"
+	"fmt"
+	"hash"
+	"reflect"
+	"strconv"
+	"sync"
+)
+
+func calcHash(v interface{}) string {
+	h := sha256.New()
+	b := bufio.NewWriterSize(h, h.BlockSize())
+	scratch := make([]byte, 0, 128)
+	printTo(b, v, scratch)
+	b.Flush()
+	scratch = h.Sum(scratch[:0])
+	hex.Encode(scratch[:cap(scratch)], scratch[:sha256.Size])
+	return string(scratch[:sha256.Size*2])
+}
+
+// UpdateHash sets last to the hash of v and reports whether its value changed.
+func UpdateHash(last *string, v ...interface{}) (changed bool) {
+	sig := calcHash(v)
+	if *last != sig {
+		*last = sig
+		return true
+	}
+	return false
+}
+
+func printTo(w *bufio.Writer, v interface{}, scratch []byte) {
+	print(w, reflect.ValueOf(v), make(map[uintptr]bool), scratch)
+}
+
+var appenderToType = reflect.TypeOf((*appenderTo)(nil)).Elem()
+
+type appenderTo interface {
+	AppendTo([]byte) []byte
+}
+
+// print hashes v into w.
+// It reports whether it was able to do so without hitting a cycle.
+func print(w *bufio.Writer, v reflect.Value, visited map[uintptr]bool, scratch []byte) (acyclic bool) {
+	if !v.IsValid() {
+		return true
+	}
+
+	if v.CanInterface() {
+		// Use AppendTo methods, if available and cheap.
+		if v.CanAddr() && v.Type().Implements(appenderToType) {
+			a := v.Addr().Interface().(appenderTo)
+			scratch = a.AppendTo(scratch[:0])
+			w.Write(scratch)
+			return true
+		}
+	}
+
+	// Generic handling.
+	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 false
+		}
+		visited[ptr] = true
+		return print(w, v.Elem(), visited, scratch)
+	case reflect.Struct:
+		acyclic = true
+		w.WriteString("struct{\n")
+		for i, n := 0, v.NumField(); i < n; i++ {
+			fmt.Fprintf(w, " [%d]: ", i)
+			if !print(w, v.Field(i), visited, scratch) {
+				acyclic = false
+			}
+			w.WriteString("\n")
+		}
+		w.WriteString("}\n")
+		return acyclic
+	case reflect.Slice, reflect.Array:
+		if v.Type().Elem().Kind() == reflect.Uint8 && v.CanInterface() {
+			fmt.Fprintf(w, "%q", v.Interface())
+			return true
+		}
+		fmt.Fprintf(w, "[%d]{\n", v.Len())
+		acyclic = true
+		for i, ln := 0, v.Len(); i < ln; i++ {
+			fmt.Fprintf(w, " [%d]: ", i)
+			if !print(w, v.Index(i), visited, scratch) {
+				acyclic = false
+			}
+			w.WriteString("\n")
+		}
+		w.WriteString("}\n")
+		return acyclic
+	case reflect.Interface:
+		return print(w, v.Elem(), visited, scratch)
+	case reflect.Map:
+		if hashMapAcyclic(w, v, visited, scratch) {
+			return true
+		}
+		return hashMapFallback(w, v, visited, scratch)
+	case reflect.String:
+		w.WriteString(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:
+		scratch = strconv.AppendUint(scratch[:0], v.Uint(), 10)
+		w.Write(scratch)
+	case reflect.Float32, reflect.Float64:
+		fmt.Fprintf(w, "%v", v.Float())
+	case reflect.Complex64, reflect.Complex128:
+		fmt.Fprintf(w, "%v", v.Complex())
+	}
+	return true
+}
+
+type mapHasher struct {
+	xbuf [sha256.Size]byte // XOR'ed accumulated buffer
+	ebuf [sha256.Size]byte // scratch buffer
+	s256 hash.Hash         // sha256 hash.Hash
+	bw   *bufio.Writer     // to hasher into ebuf
+	val  valueCache        // re-usable values for map iteration
+	iter *reflect.MapIter  // re-usable map iterator
+}
+
+func (mh *mapHasher) Reset() {
+	for i := range mh.xbuf {
+		mh.xbuf[i] = 0
+	}
+}
+
+func (mh *mapHasher) startEntry() {
+	for i := range mh.ebuf {
+		mh.ebuf[i] = 0
+	}
+	mh.bw.Flush()
+	mh.s256.Reset()
+}
+
+func (mh *mapHasher) endEntry() {
+	mh.bw.Flush()
+	for i, b := range mh.s256.Sum(mh.ebuf[:0]) {
+		mh.xbuf[i] ^= b
+	}
+}
+
+var mapHasherPool = &sync.Pool{
+	New: func() interface{} {
+		mh := new(mapHasher)
+		mh.s256 = sha256.New()
+		mh.bw = bufio.NewWriter(mh.s256)
+		mh.val = make(valueCache)
+		mh.iter = new(reflect.MapIter)
+		return mh
+	},
+}
+
+type valueCache map[reflect.Type]reflect.Value
+
+func (c valueCache) get(t reflect.Type) reflect.Value {
+	v, ok := c[t]
+	if !ok {
+		v = reflect.New(t).Elem()
+		c[t] = v
+	}
+	return v
+}
+
+// hashMapAcyclic is the faster sort-free version of map hashing. If
+// it detects a cycle it returns false and guarantees that nothing was
+// written to w.
+func hashMapAcyclic(w *bufio.Writer, v reflect.Value, visited map[uintptr]bool, scratch []byte) (acyclic bool) {
+	mh := mapHasherPool.Get().(*mapHasher)
+	defer mapHasherPool.Put(mh)
+	mh.Reset()
+	iter := mapIter(mh.iter, v)
+	defer mapIter(mh.iter, reflect.Value{}) // avoid pinning v from mh.iter when we return
+	k := mh.val.get(v.Type().Key())
+	e := mh.val.get(v.Type().Elem())
+	for iter.Next() {
+		key := iterKey(iter, k)
+		val := iterVal(iter, e)
+		mh.startEntry()
+		if !print(mh.bw, key, visited, scratch) {
+			return false
+		}
+		if !print(mh.bw, val, visited, scratch) {
+			return false
+		}
+		mh.endEntry()
+	}
+	w.Write(mh.xbuf[:])
+	return true
+}
+
+func hashMapFallback(w *bufio.Writer, v reflect.Value, visited map[uintptr]bool, scratch []byte) (acyclic bool) {
+	acyclic = true
+	sm := newSortedMap(v)
+	fmt.Fprintf(w, "map[%d]{\n", len(sm.Key))
+	for i, k := range sm.Key {
+		if !print(w, k, visited, scratch) {
+			acyclic = false
+		}
+		w.WriteString(": ")
+		if !print(w, sm.Value[i], visited, scratch) {
+			acyclic = false
+		}
+		w.WriteString("\n")
+	}
+	w.WriteString("}\n")
+	return acyclic
+}

internal/deephash/deephash_test.go

@@ -0,0 +1,136 @@
+// 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 deephash
+
+import (
+	"bufio"
+	"bytes"
+	"fmt"
+	"reflect"
+	"testing"
+
+	"inet.af/netaddr"
+	"tailscale.com/tailcfg"
+	"tailscale.com/util/dnsname"
+	"tailscale.com/wgengine/router"
+	"tailscale.com/wgengine/wgcfg"
+)
+
+func TestDeepHash(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()
+
+	hash1 := calcHash(v)
+	t.Logf("hash: %v", hash1)
+	for i := 0; i < 20; i++ {
+		hash2 := calcHash(getVal())
+		if hash1 != hash2 {
+			t.Error("second hash didn't match")
+		}
+	}
+}
+
+func getVal() []interface{} {
+	return []interface{}{
+		&wgcfg.Config{
+			Name:      "foo",
+			Addresses: []netaddr.IPPrefix{netaddr.IPPrefixFrom(netaddr.IPFrom16([16]byte{3: 3}), 5)},
+			Peers: []wgcfg.Peer{
+				{
+					Endpoints: wgcfg.Endpoints{
+						IPPorts: wgcfg.NewIPPortSet(netaddr.MustParseIPPort("42.42.42.42:5")),
+					},
+				},
+			},
+		},
+		&router.Config{
+			Routes: []netaddr.IPPrefix{
+				netaddr.MustParseIPPrefix("1.2.3.0/24"),
+				netaddr.MustParseIPPrefix("1234::/64"),
+			},
+		},
+		map[dnsname.FQDN][]netaddr.IP{
+			dnsname.FQDN("a."): {netaddr.MustParseIP("1.2.3.4"), netaddr.MustParseIP("4.3.2.1")},
+			dnsname.FQDN("b."): {netaddr.MustParseIP("8.8.8.8"), netaddr.MustParseIP("9.9.9.9")},
+			dnsname.FQDN("c."): {netaddr.MustParseIP("6.6.6.6"), netaddr.MustParseIP("7.7.7.7")},
+			dnsname.FQDN("d."): {netaddr.MustParseIP("6.7.6.6"), netaddr.MustParseIP("7.7.7.8")},
+			dnsname.FQDN("e."): {netaddr.MustParseIP("6.8.6.6"), netaddr.MustParseIP("7.7.7.9")},
+			dnsname.FQDN("f."): {netaddr.MustParseIP("6.9.6.6"), netaddr.MustParseIP("7.7.7.0")},
+		},
+		map[dnsname.FQDN][]netaddr.IPPort{
+			dnsname.FQDN("a."): {netaddr.MustParseIPPort("1.2.3.4:11"), netaddr.MustParseIPPort("4.3.2.1:22")},
+			dnsname.FQDN("b."): {netaddr.MustParseIPPort("8.8.8.8:11"), netaddr.MustParseIPPort("9.9.9.9:22")},
+			dnsname.FQDN("c."): {netaddr.MustParseIPPort("8.8.8.8:12"), netaddr.MustParseIPPort("9.9.9.9:23")},
+			dnsname.FQDN("d."): {netaddr.MustParseIPPort("8.8.8.8:13"), netaddr.MustParseIPPort("9.9.9.9:24")},
+			dnsname.FQDN("e."): {netaddr.MustParseIPPort("8.8.8.8:14"), netaddr.MustParseIPPort("9.9.9.9:25")},
+		},
+		map[tailcfg.DiscoKey]bool{
+			{1: 1}: true,
+			{1: 2}: false,
+			{2: 3}: true,
+			{3: 4}: false,
+		},
+	}
+}
+
+func BenchmarkHash(b *testing.B) {
+	b.ReportAllocs()
+	v := getVal()
+	for i := 0; i < b.N; i++ {
+		calcHash(v)
+	}
+}
+
+func TestHashMapAcyclic(t *testing.T) {
+	m := map[int]string{}
+	for i := 0; i < 100; i++ {
+		m[i] = fmt.Sprint(i)
+	}
+	got := map[string]bool{}
+
+	var buf bytes.Buffer
+	bw := bufio.NewWriter(&buf)
+
+	for i := 0; i < 20; i++ {
+		visited := map[uintptr]bool{}
+		scratch := make([]byte, 0, 64)
+		v := reflect.ValueOf(m)
+		buf.Reset()
+		bw.Reset(&buf)
+		if !hashMapAcyclic(bw, v, visited, scratch) {
+			t.Fatal("returned false")
+		}
+		if got[string(buf.Bytes())] {
+			continue
+		}
+		got[string(buf.Bytes())] = true
+	}
+	if len(got) != 1 {
+		t.Errorf("got %d results; want 1", len(got))
+	}
+}
+
+func BenchmarkHashMapAcyclic(b *testing.B) {
+	b.ReportAllocs()
+	m := map[int]string{}
+	for i := 0; i < 100; i++ {
+		m[i] = fmt.Sprint(i)
+	}
+
+	var buf bytes.Buffer
+	bw := bufio.NewWriter(&buf)
+	visited := map[uintptr]bool{}
+	scratch := make([]byte, 0, 64)
+	v := reflect.ValueOf(m)
+
+	for i := 0; i < b.N; i++ {
+		buf.Reset()
+		bw.Reset(&buf)
+		if !hashMapAcyclic(bw, v, visited, scratch) {
+			b.Fatal("returned false")
+		}
+	}
+}

internal/deephash/fmtsort.go

@@ -10,7 +10,7 @@
 
 // This is a slightly modified fork of Go's src/internal/fmtsort/sort.go
 
-package deepprint
+package deephash
 
 import (
 	"reflect"

internal/deephash/mapiter.go

@@ -0,0 +1,37 @@
+// Copyright (c) 2021 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 !tailscale_go
+
+package deephash
+
+import "reflect"
+
+// iterKey returns the current iter key.
+// scratch is a re-usable reflect.Value.
+// iterKey may store the iter key in scratch and return scratch,
+// or it may allocate and return a new reflect.Value.
+func iterKey(iter *reflect.MapIter, _ reflect.Value) reflect.Value {
+	return iter.Key()
+}
+
+// iterVal returns the current iter val.
+// scratch is a re-usable reflect.Value.
+// iterVal may store the iter val in scratch and return scratch,
+// or it may allocate and return a new reflect.Value.
+func iterVal(iter *reflect.MapIter, _ reflect.Value) reflect.Value {
+	return iter.Value()
+}
+
+// mapIter returns a map iterator for mapVal.
+// scratch is a re-usable reflect.MapIter.
+// mapIter may re-use scratch and return it,
+// or it may allocate and return a new *reflect.MapIter.
+// If mapVal is the zero reflect.Value, mapIter may return nil.
+func mapIter(_ *reflect.MapIter, mapVal reflect.Value) *reflect.MapIter {
+	if !mapVal.IsValid() {
+		return nil
+	}
+	return mapVal.MapRange()
+}

internal/deephash/mapiter_future.go

@@ -0,0 +1,42 @@
+// Copyright (c) 2021 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 tailscale_go
+
+package deephash
+
+import "reflect"
+
+// iterKey returns the current iter key.
+// scratch is a re-usable reflect.Value.
+// iterKey may store the iter key in scratch and return scratch,
+// or it may allocate and return a new reflect.Value.
+func iterKey(iter *reflect.MapIter, scratch reflect.Value) reflect.Value {
+	iter.SetKey(scratch)
+	return scratch
+}
+
+// iterVal returns the current iter val.
+// scratch is a re-usable reflect.Value.
+// iterVal may store the iter val in scratch and return scratch,
+// or it may allocate and return a new reflect.Value.
+func iterVal(iter *reflect.MapIter, scratch reflect.Value) reflect.Value {
+	iter.SetValue(scratch)
+	return scratch
+}
+
+// mapIter returns a map iterator for mapVal.
+// scratch is a re-usable reflect.MapIter.
+// mapIter may re-use scratch and return it,
+// or it may allocate and return a new *reflect.MapIter.
+// If mapVal is the zero reflect.Value, mapIter may return nil.
+func mapIter(scratch *reflect.MapIter, mapVal reflect.Value) *reflect.MapIter {
+	scratch.Reset(mapVal) // always Reset, to allow the caller to avoid pinning memory
+	if !mapVal.IsValid() {
+		// Returning scratch would also be OK.
+		// Do this for consistency with the non-optimized version.
+		return nil
+	}
+	return scratch
+}

internal/deepprint/deepprint.go

@@ -1,103 +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 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))
-}
-
-// UpdateHash sets last to the hash of v and reports whether its value changed.
-func UpdateHash(last *string, v ...interface{}) (changed bool) {
-	sig := Hash(v)
-	if *last != sig {
-		*last = sig
-		return true
-	}
-	return false
-}
-
-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

@@ -1,64 +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 deepprint
-
-import (
-	"bytes"
-	"testing"
-
-	"inet.af/netaddr"
-	"tailscale.com/wgengine/router"
-	"tailscale.com/wgengine/wgcfg"
-)
-
-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: []netaddr.IPPrefix{{Bits: 5, IP: netaddr.IPFrom16([16]byte{3: 3})}},
-			Peers: []wgcfg.Peer{
-				{
-					Endpoints: "foo:5",
-				},
-			},
-		},
-		&router.Config{
-			Routes: []netaddr.IPPrefix{
-				netaddr.MustParseIPPrefix("1.2.3.0/24"),
-				netaddr.MustParseIPPrefix("1234::/64"),
-			},
-		},
-		map[string]string{
-			"key1": "val1",
-			"key2": "val2",
-			"key3": "val3",
-			"key4": "val4",
-			"key5": "val5",
-			"key6": "val6",
-			"key7": "val7",
-			"key8": "val8",
-			"key9": "val9",
-		},
-	}
-}

ipn/backend.go

@@ -5,6 +5,8 @@
 package ipn
 
 import (
+	"fmt"
+	"strings"
 	"time"
 
 	"tailscale.com/ipn/ipnstate"
@@ -55,17 +57,21 @@ type EngineStatus struct {
 // that they have not changed.
 // They are JSON-encoded on the wire, despite the lack of struct tags.
 type Notify struct {
-	_             structs.Incomparable
-	Version       string             // version number of IPN backend
-	ErrMessage    *string            // critical error message, if any; for InUseOtherUser, the details
-	LoginFinished *empty.Message     // event: non-nil when login process succeeded
-	State         *State             // current IPN state has changed
-	Prefs         *Prefs             // preferences were changed
-	NetMap        *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
-	PingResult    *ipnstate.PingResult
+	_       structs.Incomparable
+	Version string // version number of IPN backend
+
+	// ErrMessage, if non-nil, contains a critical error message.
+	// For State InUseOtherUser, ErrMessage is not critical and just contains the details.
+	ErrMessage *string
+
+	LoginFinished *empty.Message       // non-nil when/if the login process succeeded
+	State         *State               // if non-nil, the new or current IPN state
+	Prefs         *Prefs               // if non-nil, the new or current preferences
+	NetMap        *netmap.NetworkMap   // if non-nil, the new or current netmap
+	Engine        *EngineStatus        // if non-nil, the new or urrent wireguard stats
+	BrowseToURL   *string              // if non-nil, UI should open a browser right now
+	BackendLogID  *string              // if non-nil, the public logtail ID used by backend
+	PingResult    *ipnstate.PingResult // if non-nil, a ping response arrived
 
 	// FilesWaiting if non-nil means that files are buffered in
 	// the Tailscale daemon and ready for local transfer to the
@@ -88,6 +94,49 @@ type Notify struct {
 	// type is mirrored in xcode/Shared/IPN.swift
 }
 
+func (n Notify) String() string {
+	var sb strings.Builder
+	sb.WriteString("Notify{")
+	if n.ErrMessage != nil {
+		fmt.Fprintf(&sb, "err=%q ", *n.ErrMessage)
+	}
+	if n.LoginFinished != nil {
+		sb.WriteString("LoginFinished ")
+	}
+	if n.State != nil {
+		fmt.Fprintf(&sb, "state=%v ", *n.State)
+	}
+	if n.Prefs != nil {
+		fmt.Fprintf(&sb, "%v ", n.Prefs.Pretty())
+	}
+	if n.NetMap != nil {
+		sb.WriteString("NetMap{...} ")
+	}
+	if n.Engine != nil {
+		fmt.Fprintf(&sb, "wg=%v ", *n.Engine)
+	}
+	if n.BrowseToURL != nil {
+		sb.WriteString("URL=<...> ")
+	}
+	if n.BackendLogID != nil {
+		sb.WriteString("BackendLogID ")
+	}
+	if n.PingResult != nil {
+		fmt.Fprintf(&sb, "ping=%v ", *n.PingResult)
+	}
+	if n.FilesWaiting != nil {
+		sb.WriteString("FilesWaiting ")
+	}
+	if len(n.IncomingFiles) != 0 {
+		sb.WriteString("IncomingFiles ")
+	}
+	if n.LocalTCPPort != nil {
+		fmt.Fprintf(&sb, "tcpport=%v ", n.LocalTCPPort)
+	}
+	s := sb.String()
+	return s[0:len(s)-1] + "}"
+}
+
 // PartialFile represents an in-progress file transfer.
 type PartialFile struct {
 	Name         string    // e.g. "foo.jpg"
@@ -95,12 +144,15 @@ type PartialFile struct {
 	DeclaredSize int64     // or -1 if unknown
 	Received     int64     // bytes copied thus far
 
-	// FinalPath is non-empty when the final has been completely
-	// written and renamed into place. This is then the complete
-	// path to the file post-rename. This is only set in
-	// "direct" file mode where the peerapi isn't being used; see
-	// LocalBackend.SetDirectFileRoot.
-	FinalPath string `json:",omitempty"`
+	// PartialPath is set non-empty in "direct" file mode to the
+	// in-progress '*.partial' file's path when the peerapi isn't
+	// being used; see LocalBackend.SetDirectFileRoot.
+	PartialPath string `json:",omitempty"`
+
+	// Done is set in "direct" mode when the partial file has been
+	// closed and is ready for the caller to rename away the
+	// ".partial" suffix.
+	Done bool `json:",omitempty"`
 }
 
 // StateKey is an opaque identifier for a set of LocalBackend state
@@ -133,8 +185,30 @@ type Options struct {
 	//    state and use/update that.
 	//  - StateKey!="" && Prefs!=nil: like the previous case, but do
 	//    an initial overwrite of backend state with Prefs.
+	//
+	// NOTE(apenwarr): The above means that this Prefs field does not do
+	// what you probably think it does. It will overwrite your encryption
+	// keys. Do not use unless you know what you're doing.
 	StateKey StateKey
 	Prefs    *Prefs
+	// UpdatePrefs, if provided, overrides Options.Prefs *and* the Prefs
+	// already stored in the backend state, *except* for the Persist
+	// Persist member. If you just want to provide prefs, this is
+	// probably what you want.
+	//
+	// UpdatePrefs.Persist is always ignored. Prefs.Persist will still
+	// be used even if UpdatePrefs is provided. Other than Persist,
+	// UpdatePrefs takes precedence over Prefs.
+	//
+	// This is intended as a purely temporary workaround for the
+	// currently unexpected behaviour of Options.Prefs.
+	//
+	// TODO(apenwarr): Remove this, or rename Prefs to something else
+	//   and rename this to Prefs. Or, move Prefs.Persist elsewhere
+	//   entirely (as it always should have been), and then we wouldn't
+	//   need two separate fields at all. Or, move the fancy state
+	//   migration stuff out of Start().
+	UpdatePrefs *Prefs
 	// AuthKey is an optional node auth key used to authorize a
 	// new node key without user interaction.
 	AuthKey string

ipn/fake_test.go

@@ -19,7 +19,7 @@ type FakeBackend struct {
 }
 
 func (b *FakeBackend) Start(opts Options) error {
-	b.serverURL = opts.Prefs.ControlURL
+	b.serverURL = opts.Prefs.ControlURLOrDefault()
 	if b.notify == nil {
 		panic("FakeBackend.Start: SetNotifyCallback not called")
 	}

ipn/handle.go

@@ -156,7 +156,7 @@ func (h *Handle) Expiry() time.Time {
 }
 
 func (h *Handle) AdminPageURL() string {
-	return h.prefsCache.ControlURL + "/admin/machines"
+	return h.prefsCache.ControlURLOrDefault() + "/admin/machines"
 }
 
 func (h *Handle) StartLoginInteractive() {

ipn/ipnlocal/local_test.go

@@ -5,14 +5,20 @@
 package ipnlocal
 
 import (
+	"fmt"
+	"net/http"
 	"reflect"
 	"testing"
+	"time"
 
 	"inet.af/netaddr"
+	"tailscale.com/ipn"
 	"tailscale.com/net/interfaces"
 	"tailscale.com/net/tsaddr"
 	"tailscale.com/tailcfg"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
+	"tailscale.com/wgengine"
 	"tailscale.com/wgengine/wgcfg"
 )
 
@@ -165,7 +171,7 @@ func TestShrinkDefaultRoute(t *testing.T) {
 			out: []string{
 				"fe80::1",
 				"ff00::1",
-				tsaddr.TailscaleULARange().IP.String(),
+				tsaddr.TailscaleULARange().IP().String(),
 			},
 			localIPFn: func(ip netaddr.IP) bool { return !inRemove(ip) && ip.Is6() },
 		},
@@ -419,3 +425,71 @@ func TestPeerAPIBase(t *testing.T) {
 		})
 	}
 }
+
+type panicOnUseTransport struct{}
+
+func (panicOnUseTransport) RoundTrip(*http.Request) (*http.Response, error) {
+	panic("unexpected HTTP request")
+}
+
+// Issue 1573: don't generate a machine key if we don't want to be running.
+func TestLazyMachineKeyGeneration(t *testing.T) {
+	defer func(old bool) { panicOnMachineKeyGeneration = old }(panicOnMachineKeyGeneration)
+	panicOnMachineKeyGeneration = true
+
+	var logf logger.Logf = logger.Discard
+	store := new(ipn.MemoryStore)
+	eng, err := wgengine.NewFakeUserspaceEngine(logf, 0)
+	if err != nil {
+		t.Fatalf("NewFakeUserspaceEngine: %v", err)
+	}
+	lb, err := NewLocalBackend(logf, "logid", store, eng)
+	if err != nil {
+		t.Fatalf("NewLocalBackend: %v", err)
+	}
+
+	lb.SetHTTPTestClient(&http.Client{
+		Transport: panicOnUseTransport{}, // validate we don't send HTTP requests
+	})
+
+	if err := lb.Start(ipn.Options{
+		StateKey: ipn.GlobalDaemonStateKey,
+	}); err != nil {
+		t.Fatalf("Start: %v", err)
+	}
+
+	// Give the controlclient package goroutines (if they're
+	// accidentally started) extra time to schedule and run (and thus
+	// hit panicOnUseTransport).
+	time.Sleep(500 * time.Millisecond)
+}
+
+func TestFileTargets(t *testing.T) {
+	b := new(LocalBackend)
+	_, err := b.FileTargets()
+	if got, want := fmt.Sprint(err), "not connected"; got != want {
+		t.Errorf("before connect: got %q; want %q", got, want)
+	}
+
+	b.netMap = new(netmap.NetworkMap)
+	_, err = b.FileTargets()
+	if got, want := fmt.Sprint(err), "not connected"; got != want {
+		t.Errorf("non-running netmap: got %q; want %q", got, want)
+	}
+
+	b.state = ipn.Running
+	_, err = b.FileTargets()
+	if got, want := fmt.Sprint(err), "file sharing not enabled by Tailscale admin"; got != want {
+		t.Errorf("without cap: got %q; want %q", got, want)
+	}
+
+	b.capFileSharing = true
+	got, err := b.FileTargets()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(got) != 0 {
+		t.Fatalf("unexpected %d peers", len(got))
+	}
+	// (other cases handled by TestPeerAPIBase above)
+}

ipn/ipnlocal/loglines_test.go

@@ -15,6 +15,7 @@ import (
 	"tailscale.com/tailcfg"
 	"tailscale.com/tstest"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 	"tailscale.com/types/persist"
 	"tailscale.com/wgengine"
 )
@@ -30,6 +31,12 @@ func TestLocalLogLines(t *testing.T) {
 	})
 	defer logListen.Close()
 
+	// Put a rate-limiter with a burst of 0 between the components below.
+	// This instructs the rate-limiter to eliminate all logging that
+	// isn't explicitly exempt from rate-limiting.
+	// This lets the logListen tracker verify that the rate-limiter allows these key lines.
+	logf := logger.RateLimitedFnWithClock(logListen.Logf, 5*time.Second, 0, 10, time.Now)
+
 	logid := func(hex byte) logtail.PublicID {
 		var ret logtail.PublicID
 		for i := 0; i < len(ret); i++ {
@@ -41,12 +48,12 @@ func TestLocalLogLines(t *testing.T) {
 
 	// set up a LocalBackend, super bare bones. No functional data.
 	store := &ipn.MemoryStore{}
-	e, err := wgengine.NewFakeUserspaceEngine(logListen.Logf, 0)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, 0)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	lb, err := NewLocalBackend(logListen.Logf, idA.String(), store, e)
+	lb, err := NewLocalBackend(logf, idA.String(), store, e)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -61,6 +68,7 @@ func TestLocalLogLines(t *testing.T) {
 	testWantRemain := func(wantRemain ...string) func(t *testing.T) {
 		return func(t *testing.T) {
 			if remain := logListen.Check(); !reflect.DeepEqual(remain, wantRemain) {
+				t.Helper()
 				t.Errorf("remain %q, want %q", remain, wantRemain)
 			}
 		}
@@ -75,17 +83,30 @@ func TestLocalLogLines(t *testing.T) {
 	t.Run("after_prefs", testWantRemain("[v1] peer keys: %s", "[v1] v%v peers: %v"))
 
 	// log peers, peer keys
-	status := &wgengine.Status{
+	lb.mu.Lock()
+	lb.parseWgStatusLocked(&wgengine.Status{
 		Peers: []ipnstate.PeerStatusLite{{
 			TxBytes:       10,
 			RxBytes:       10,
 			LastHandshake: time.Now(),
 			NodeKey:       tailcfg.NodeKey(key.NewPrivate()),
 		}},
-	}
-	lb.mu.Lock()
-	lb.parseWgStatusLocked(status)
+	})
 	lb.mu.Unlock()
 
 	t.Run("after_peers", testWantRemain())
+
+	// Log it again with different stats to ensure it's not dup-suppressed.
+	logListen.Reset()
+	lb.mu.Lock()
+	lb.parseWgStatusLocked(&wgengine.Status{
+		Peers: []ipnstate.PeerStatusLite{{
+			TxBytes:       11,
+			RxBytes:       12,
+			LastHandshake: time.Now(),
+			NodeKey:       tailcfg.NodeKey(key.NewPrivate()),
+		}},
+	})
+	lb.mu.Unlock()
+	t.Run("after_second_peer_status", testWantRemain("SetPrefs: %v"))
 }

ipn/ipnlocal/peerapi.go

@@ -11,6 +11,7 @@ import (
 	"hash/crc32"
 	"html"
 	"io"
+	"io/fs"
 	"net"
 	"net/http"
 	"net/url"
@@ -18,14 +19,18 @@ import (
 	"path"
 	"path/filepath"
 	"runtime"
+	"sort"
 	"strconv"
 	"strings"
 	"sync"
 	"time"
+	"unicode"
+	"unicode/utf8"
 
 	"inet.af/netaddr"
 	"tailscale.com/client/tailscale/apitype"
 	"tailscale.com/ipn"
+	"tailscale.com/logtail/backoff"
 	"tailscale.com/net/interfaces"
 	"tailscale.com/syncs"
 	"tailscale.com/tailcfg"
@@ -45,20 +50,66 @@ type peerAPIServer struct {
 	// download directory (as *.partial files), rather than making
 	// the frontend retrieve it over localapi HTTP and write it
 	// somewhere itself. This is used on GUI macOS version.
+	// In directFileMode, the peerapi doesn't do the final rename
+	// from "foo.jpg.partial" to "foo.jpg".
 	directFileMode bool
 }
 
-const partialSuffix = ".partial"
+const (
+	// partialSuffix is the suffix appened to files while they're
+	// still in the process of being transferred.
+	partialSuffix = ".partial"
+
+	// deletedSuffix is the suffix for a deleted marker file
+	// that's placed next to a file (without the suffix) that we
+	// tried to delete, but Windows wouldn't let us. These are
+	// only written on Windows (and in tests), but they're not
+	// permitted to be uploaded directly on any platform, like
+	// partial files.
+	deletedSuffix = ".deleted"
+)
+
+func validFilenameRune(r rune) bool {
+	switch r {
+	case '/':
+		return false
+	case '\\', ':', '*', '"', '<', '>', '|':
+		// Invalid stuff on Windows, but we reject them everywhere
+		// for now.
+		// TODO(bradfitz): figure out a better plan. We initially just
+		// wrote things to disk URL path-escaped, but that's gross
+		// when debugging, and just moves the problem to callers.
+		// So now we put the UTF-8 filenames on disk directly as
+		// sent.
+		return false
+	}
+	return unicode.IsPrint(r)
+}
 
 func (s *peerAPIServer) diskPath(baseName string) (fullPath string, ok bool) {
+	if !utf8.ValidString(baseName) {
+		return "", false
+	}
+	if strings.TrimSpace(baseName) != baseName {
+		return "", false
+	}
+	if len(baseName) > 255 {
+		return "", false
+	}
+	// TODO: validate unicode normalization form too? Varies by platform.
 	clean := path.Clean(baseName)
 	if clean != baseName ||
-		clean == "." ||
-		strings.ContainsAny(clean, `/\`) ||
+		clean == "." || clean == ".." ||
+		strings.HasSuffix(clean, deletedSuffix) ||
 		strings.HasSuffix(clean, partialSuffix) {
 		return "", false
 	}
-	return filepath.Join(s.rootDir, strings.ReplaceAll(url.PathEscape(baseName), ":", "%3a")), true
+	for _, r := range baseName {
+		if !validFilenameRune(r) {
+			return "", false
+		}
+	}
+	return filepath.Join(s.rootDir, baseName), true
 }
 
 // hasFilesWaiting reports whether any files are buffered in the
@@ -83,11 +134,28 @@ func (s *peerAPIServer) hasFilesWaiting() bool {
 	for {
 		des, err := f.ReadDir(10)
 		for _, de := range des {
-			if strings.HasSuffix(de.Name(), partialSuffix) {
+			name := de.Name()
+			if strings.HasSuffix(name, partialSuffix) {
+				continue
+			}
+			if strings.HasSuffix(name, deletedSuffix) { // for Windows + tests
+				// After we're done looping over files, then try
+				// to delete this file. Don't do it proactively,
+				// as the OS may return "foo.jpg.deleted" before "foo.jpg"
+				// and we don't want to delete the ".deleted" file before
+				// enumerating to the "foo.jpg" file.
+				defer tryDeleteAgain(filepath.Join(s.rootDir, strings.TrimSuffix(name, deletedSuffix)))
 				continue
 			}
 			if de.Type().IsRegular() {
-				return true
+				_, err := os.Stat(filepath.Join(s.rootDir, name+deletedSuffix))
+				if os.IsNotExist(err) {
+					return true
+				}
+				if err == nil {
+					tryDeleteAgain(filepath.Join(s.rootDir, name))
+					continue
+				}
 			}
 		}
 		if err == io.EOF {
@@ -100,6 +168,12 @@ func (s *peerAPIServer) hasFilesWaiting() bool {
 	return false
 }
 
+// WaitingFiles returns the list of files that have been sent by a
+// peer that are waiting in the buffered "pick up" directory owned by
+// the Tailscale daemon.
+//
+// As a side effect, it also does any lazy deletion of files as
+// required by Windows.
 func (s *peerAPIServer) WaitingFiles() (ret []apitype.WaitingFile, err error) {
 	if s.rootDir == "" {
 		return nil, errors.New("peerapi disabled; no storage configured")
@@ -112,6 +186,7 @@ func (s *peerAPIServer) WaitingFiles() (ret []apitype.WaitingFile, err error) {
 		return nil, err
 	}
 	defer f.Close()
+	var deleted map[string]bool // "foo.jpg" => true (if "foo.jpg.deleted" exists)
 	for {
 		des, err := f.ReadDir(10)
 		for _, de := range des {
@@ -119,6 +194,13 @@ func (s *peerAPIServer) WaitingFiles() (ret []apitype.WaitingFile, err error) {
 			if strings.HasSuffix(name, partialSuffix) {
 				continue
 			}
+			if strings.HasSuffix(name, deletedSuffix) { // for Windows + tests
+				if deleted == nil {
+					deleted = map[string]bool{}
+				}
+				deleted[strings.TrimSuffix(name, deletedSuffix)] = true
+				continue
+			}
 			if de.Type().IsRegular() {
 				fi, err := de.Info()
 				if err != nil {
@@ -137,9 +219,41 @@ func (s *peerAPIServer) WaitingFiles() (ret []apitype.WaitingFile, err error) {
 			return nil, err
 		}
 	}
+	if len(deleted) > 0 {
+		// Filter out any return values "foo.jpg" where a
+		// "foo.jpg.deleted" marker file exists on disk.
+		all := ret
+		ret = ret[:0]
+		for _, wf := range all {
+			if !deleted[wf.Name] {
+				ret = append(ret, wf)
+			}
+		}
+		// And do some opportunistic deleting while we're here.
+		// Maybe Windows is done virus scanning the file we tried
+		// to delete a long time ago and will let us delete it now.
+		for name := range deleted {
+			tryDeleteAgain(filepath.Join(s.rootDir, name))
+		}
+	}
+	sort.Slice(ret, func(i, j int) bool { return ret[i].Name < ret[j].Name })
 	return ret, nil
 }
 
+// tryDeleteAgain tries to delete path (and path+deletedSuffix) after
+// it failed earlier.  This happens on Windows when various anti-virus
+// tools hook into filesystem operations and have the file open still
+// while we're trying to delete it. In that case we instead mark it as
+// deleted (writing a "foo.jpg.deleted" marker file), but then we
+// later try to clean them up.
+//
+// fullPath is the full path to the file without the deleted suffix.
+func tryDeleteAgain(fullPath string) {
+	if err := os.Remove(fullPath); err == nil || os.IsNotExist(err) {
+		os.Remove(fullPath + deletedSuffix)
+	}
+}
+
 func (s *peerAPIServer) DeleteFile(baseName string) error {
 	if s.rootDir == "" {
 		return errors.New("peerapi disabled; no storage configured")
@@ -151,11 +265,59 @@ func (s *peerAPIServer) DeleteFile(baseName string) error {
 	if !ok {
 		return errors.New("bad filename")
 	}
-	err := os.Remove(path)
-	if err != nil && !os.IsNotExist(err) {
-		return err
+	var bo *backoff.Backoff
+	logf := s.b.logf
+	t0 := time.Now()
+	for {
+		err := os.Remove(path)
+		if err != nil && !os.IsNotExist(err) {
+			err = redactErr(err)
+			// Put a retry loop around deletes on Windows. Windows
+			// file descriptor closes are effectively asynchronous,
+			// as a bunch of hooks run on/after close, and we can't
+			// necessarily delete the file for a while after close,
+			// as we need to wait for everybody to be done with
+			// it. (on Windows, unlike Unix, a file can't be deleted
+			// if it's open anywhere)
+			// So try a few times but ultimately just leave a
+			// "foo.jpg.deleted" marker file to note that it's
+			// deleted and we clean it up later.
+			if runtime.GOOS == "windows" {
+				if bo == nil {
+					bo = backoff.NewBackoff("delete-retry", logf, 1*time.Second)
+				}
+				if time.Since(t0) < 5*time.Second {
+					bo.BackOff(context.Background(), err)
+					continue
+				}
+				if err := touchFile(path + deletedSuffix); err != nil {
+					logf("peerapi: failed to leave deleted marker: %v", err)
+				}
+			}
+			logf("peerapi: failed to DeleteFile: %v", err)
+			return err
+		}
+		return nil
 	}
-	return nil
+}
+
+// redacted is a fake path name we use in errors, to avoid
+// accidentally logging actual filenames anywhere.
+const redacted = "redacted"
+
+func redactErr(err error) error {
+	if pe, ok := err.(*os.PathError); ok {
+		pe.Path = redacted
+	}
+	return err
+}
+
+func touchFile(path string) error {
+	f, err := os.OpenFile(path, os.O_RDWR|os.O_CREATE, 0666)
+	if err != nil {
+		return redactErr(err)
+	}
+	return f.Close()
 }
 
 func (s *peerAPIServer) OpenFile(baseName string) (rc io.ReadCloser, size int64, err error) {
@@ -169,14 +331,18 @@ func (s *peerAPIServer) OpenFile(baseName string) (rc io.ReadCloser, size int64,
 	if !ok {
 		return nil, 0, errors.New("bad filename")
 	}
+	if fi, err := os.Stat(path + deletedSuffix); err == nil && fi.Mode().IsRegular() {
+		tryDeleteAgain(path)
+		return nil, 0, &fs.PathError{Op: "open", Path: redacted, Err: fs.ErrNotExist}
+	}
 	f, err := os.Open(path)
 	if err != nil {
-		return nil, 0, err
+		return nil, 0, redactErr(err)
 	}
 	fi, err := f.Stat()
 	if err != nil {
 		f.Close()
-		return nil, 0, err
+		return nil, 0, redactErr(err)
 	}
 	return f, fi.Size(), nil
 }
@@ -290,7 +456,6 @@ func (pln *peerAPIListener) serve() {
 			remoteAddr: ipp,
 			peerNode:   peerNode,
 			peerUser:   peerUser,
-			lb:         pln.lb,
 		}
 		httpServer := &http.Server{
 			Handler: h,
@@ -324,7 +489,6 @@ type peerAPIHandler struct {
 	isSelf     bool                // whether peerNode is owned by same user as this node
 	peerNode   *tailcfg.Node       // peerNode is who's making the request
 	peerUser   tailcfg.UserProfile // profile of peerNode
-	lb         *LocalBackend
 }
 
 func (h *peerAPIHandler) logf(format string, a ...interface{}) {
@@ -333,7 +497,11 @@ func (h *peerAPIHandler) logf(format string, a ...interface{}) {
 
 func (h *peerAPIHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	if strings.HasPrefix(r.URL.Path, "/v0/put/") {
-		h.put(w, r)
+		h.handlePeerPut(w, r)
+		return
+	}
+	if r.URL.Path == "/v0/goroutines" {
+		h.handleServeGoroutines(w, r)
 		return
 	}
 	who := h.peerUser.DisplayName
@@ -342,7 +510,7 @@ func (h *peerAPIHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 <body>
 <h1>Hello, %s (%v)</h1>
 This is my Tailscale device. Your device is %v.
-`, html.EscapeString(who), h.remoteAddr.IP, html.EscapeString(h.peerNode.ComputedName))
+`, html.EscapeString(who), h.remoteAddr.IP(), html.EscapeString(h.peerNode.ComputedName))
 
 	if h.isSelf {
 		fmt.Fprintf(w, "<p>You are the owner of this node.\n")
@@ -350,21 +518,22 @@ This is my Tailscale device. Your device is %v.
 }
 
 type incomingFile struct {
-	name    string // "foo.jpg"
-	started time.Time
-	size    int64     // or -1 if unknown; never 0
-	w       io.Writer // underlying writer
-	ph      *peerAPIHandler
+	name        string // "foo.jpg"
+	started     time.Time
+	size        int64     // or -1 if unknown; never 0
+	w           io.Writer // underlying writer
+	ph          *peerAPIHandler
+	partialPath string // non-empty in direct mode
 
 	mu         sync.Mutex
 	copied     int64
+	done       bool
 	lastNotify time.Time
-	finalPath  string // non-empty in direct mode, when file is done
 }
 
-func (f *incomingFile) markAndNotifyDone(finalPath string) {
+func (f *incomingFile) markAndNotifyDone() {
 	f.mu.Lock()
-	f.finalPath = finalPath
+	f.done = true
 	f.mu.Unlock()
 	b := f.ph.ps.b
 	b.sendFileNotify()
@@ -401,11 +570,12 @@ func (f *incomingFile) PartialFile() ipn.PartialFile {
 		Started:      f.started,
 		DeclaredSize: f.size,
 		Received:     f.copied,
-		FinalPath:    f.finalPath,
+		PartialPath:  f.partialPath,
+		Done:         f.done,
 	}
 }
 
-func (h *peerAPIHandler) put(w http.ResponseWriter, r *http.Request) {
+func (h *peerAPIHandler) handlePeerPut(w http.ResponseWriter, r *http.Request) {
 	if !h.isSelf {
 		http.Error(w, "not owner", http.StatusForbidden)
 		return
@@ -415,32 +585,50 @@ func (h *peerAPIHandler) put(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	if r.Method != "PUT" {
-		http.Error(w, "not method PUT", http.StatusMethodNotAllowed)
+		http.Error(w, "expected method PUT", http.StatusMethodNotAllowed)
 		return
 	}
 	if h.ps.rootDir == "" {
 		http.Error(w, "no rootdir", http.StatusInternalServerError)
 		return
 	}
-	baseName := path.Base(r.URL.Path)
+	rawPath := r.URL.EscapedPath()
+	suffix := strings.TrimPrefix(rawPath, "/v0/put/")
+	if suffix == rawPath {
+		http.Error(w, "misconfigured internals", 500)
+		return
+	}
+	if suffix == "" {
+		http.Error(w, "empty filename", 400)
+		return
+	}
+	if strings.Contains(suffix, "/") {
+		http.Error(w, "directories not supported", 400)
+		return
+	}
+	baseName, err := url.PathUnescape(suffix)
+	if err != nil {
+		http.Error(w, "bad path encoding", 400)
+		return
+	}
 	dstFile, ok := h.ps.diskPath(baseName)
 	if !ok {
 		http.Error(w, "bad filename", 400)
 		return
 	}
-	if h.ps.directFileMode {
-		dstFile += partialSuffix
-	}
-	f, err := os.Create(dstFile)
+	t0 := time.Now()
+	// TODO(bradfitz): prevent same filename being sent by two peers at once
+	partialFile := dstFile + partialSuffix
+	f, err := os.Create(partialFile)
 	if err != nil {
-		h.logf("put Create error: %v", err)
+		h.logf("put Create error: %v", redactErr(err))
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	var success bool
 	defer func() {
 		if !success {
-			os.Remove(dstFile)
+			os.Remove(partialFile)
 		}
 	}()
 	var finalSize int64
@@ -453,10 +641,14 @@ func (h *peerAPIHandler) put(w http.ResponseWriter, r *http.Request) {
 			w:       f,
 			ph:      h,
 		}
+		if h.ps.directFileMode {
+			inFile.partialPath = partialFile
+		}
 		h.ps.b.registerIncomingFile(inFile, true)
 		defer h.ps.b.registerIncomingFile(inFile, false)
 		n, err := io.Copy(inFile, r.Body)
 		if err != nil {
+			err = redactErr(err)
 			f.Close()
 			h.logf("put Copy error: %v", err)
 			http.Error(w, err.Error(), http.StatusInternalServerError)
@@ -464,24 +656,26 @@ func (h *peerAPIHandler) put(w http.ResponseWriter, r *http.Request) {
 		}
 		finalSize = n
 	}
-	if err := f.Close(); err != nil {
+	if err := redactErr(f.Close()); err != nil {
 		h.logf("put Close error: %v", err)
 		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	if h.ps.directFileMode {
-		finalPath := strings.TrimSuffix(dstFile, partialSuffix)
-		if err := os.Rename(dstFile, finalPath); err != nil {
-			h.logf("Rename error: %v", err)
-			http.Error(w, "error renaming file", http.StatusInternalServerError)
-			return
-		}
 		if inFile != nil { // non-zero length; TODO: notify even for zero length
-			inFile.markAndNotifyDone(finalPath)
+			inFile.markAndNotifyDone()
+		}
+	} else {
+		if err := os.Rename(partialFile, dstFile); err != nil {
+			err = redactErr(err)
+			h.logf("put final rename: %v", err)
+			http.Error(w, err.Error(), http.StatusInternalServerError)
+			return
 		}
 	}
 
-	h.logf("put of %s from %v/%v", approxSize(finalSize), h.remoteAddr.IP, h.peerNode.ComputedName)
+	d := time.Since(t0).Round(time.Second / 10)
+	h.logf("got put of %s in %v from %v/%v", approxSize(finalSize), d, h.remoteAddr.IP, h.peerNode.ComputedName)
 
 	// TODO: set modtime
 	// TODO: some real response
@@ -498,5 +692,21 @@ func approxSize(n int64) string {
 	if n <= 1<<20 {
 		return "<=1MB"
 	}
-	return fmt.Sprintf("~%dMB", n/1<<20)
+	return fmt.Sprintf("~%dMB", n>>20)
+}
+
+func (h *peerAPIHandler) handleServeGoroutines(w http.ResponseWriter, r *http.Request) {
+	if !h.isSelf {
+		http.Error(w, "not owner", http.StatusForbidden)
+		return
+	}
+	var buf []byte
+	for size := 4 << 10; size <= 2<<20; size *= 2 {
+		buf = make([]byte, size)
+		buf = buf[:runtime.Stack(buf, true)]
+		if len(buf) < size {
+			break
+		}
+	}
+	w.Write(buf)
 }

ipn/ipnlocal/peerapi_test.go

@@ -0,0 +1,573 @@
+// Copyright (c) 2021 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 ipnlocal
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"io/fs"
+	"io/ioutil"
+	"math/rand"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"testing"
+
+	"tailscale.com/tailcfg"
+)
+
+type peerAPITestEnv struct {
+	ph     *peerAPIHandler
+	rr     *httptest.ResponseRecorder
+	logBuf bytes.Buffer
+}
+
+func (e *peerAPITestEnv) logf(format string, a ...interface{}) {
+	fmt.Fprintf(&e.logBuf, format, a...)
+}
+
+type check func(*testing.T, *peerAPITestEnv)
+
+func checks(vv ...check) []check { return vv }
+
+func httpStatus(wantStatus int) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if res := e.rr.Result(); res.StatusCode != wantStatus {
+			t.Errorf("HTTP response code = %v; want %v", res.Status, wantStatus)
+		}
+	}
+}
+
+func bodyContains(sub string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if body := e.rr.Body.String(); !strings.Contains(body, sub) {
+			t.Errorf("HTTP response body does not contain %q; got: %s", sub, body)
+		}
+	}
+}
+
+func bodyNotContains(sub string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		if body := e.rr.Body.String(); strings.Contains(body, sub) {
+			t.Errorf("HTTP response body unexpectedly contains %q; got: %s", sub, body)
+		}
+	}
+}
+
+func fileHasSize(name string, size int) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		root := e.ph.ps.rootDir
+		if root == "" {
+			t.Errorf("no rootdir; can't check whether %q has size %v", name, size)
+			return
+		}
+		path := filepath.Join(root, name)
+		if fi, err := os.Stat(path); err != nil {
+			t.Errorf("fileHasSize(%q, %v): %v", name, size, err)
+		} else if fi.Size() != int64(size) {
+			t.Errorf("file %q has size %v; want %v", name, fi.Size(), size)
+		}
+	}
+}
+
+func fileHasContents(name string, want string) check {
+	return func(t *testing.T, e *peerAPITestEnv) {
+		root := e.ph.ps.rootDir
+		if root == "" {
+			t.Errorf("no rootdir; can't check contents of %q", name)
+			return
+		}
+		path := filepath.Join(root, name)
+		got, err := ioutil.ReadFile(path)
+		if err != nil {
+			t.Errorf("fileHasContents: %v", err)
+			return
+		}
+		if string(got) != want {
+			t.Errorf("file contents = %q; want %q", got, want)
+		}
+	}
+}
+
+func hexAll(v string) string {
+	var sb strings.Builder
+	for i := 0; i < len(v); i++ {
+		fmt.Fprintf(&sb, "%%%02x", v[i])
+	}
+	return sb.String()
+}
+
+func TestHandlePeerAPI(t *testing.T) {
+	tests := []struct {
+		name       string
+		isSelf     bool // the peer sending the request is owned by us
+		capSharing bool // self node has file sharing capabilty
+		omitRoot   bool // don't configure
+		req        *http.Request
+		checks     []check
+	}{
+		{
+			name:       "not_peer_api",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("GET", "/", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("This is my Tailscale device."),
+				bodyContains("You are the owner of this node."),
+			),
+		},
+		{
+			name:       "not_peer_api_not_owner",
+			isSelf:     false,
+			capSharing: true,
+			req:        httptest.NewRequest("GET", "/", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("This is my Tailscale device."),
+				bodyNotContains("You are the owner of this node."),
+			),
+		},
+		{
+			name:   "peer_api_goroutines_deny",
+			isSelf: false,
+			req:    httptest.NewRequest("GET", "/v0/goroutines", nil),
+			checks: checks(httpStatus(403)),
+		},
+		{
+			name:   "peer_api_goroutines",
+			isSelf: true,
+			req:    httptest.NewRequest("GET", "/v0/goroutines", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("ServeHTTP"),
+			),
+		},
+		{
+			name:       "reject_non_owner_put",
+			isSelf:     false,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusForbidden),
+				bodyContains("not owner"),
+			),
+		},
+		{
+			name:       "owner_without_cap",
+			isSelf:     true,
+			capSharing: false,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusForbidden),
+				bodyContains("file sharing not enabled by Tailscale admin"),
+			),
+		},
+		{
+			name:       "owner_with_cap_no_rootdir",
+			omitRoot:   true,
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(http.StatusInternalServerError),
+				bodyContains("no rootdir"),
+			),
+		},
+		{
+			name:       "bad_method",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("POST", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(405),
+				bodyContains("expected method PUT"),
+			),
+		},
+		{
+			name:       "put_zero_length",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", nil),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", 0),
+				fileHasContents("foo", ""),
+			),
+		},
+		{
+			name:       "put_non_zero_length_content_length",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", strings.NewReader("contents")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", len("contents")),
+				fileHasContents("foo", "contents"),
+			),
+		},
+		{
+			name:       "put_non_zero_length_chunked",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo", struct{ io.Reader }{strings.NewReader("contents")}),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasSize("foo", len("contents")),
+				fileHasContents("foo", "contents"),
+			),
+		},
+		{
+			name:       "bad_filename_partial",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo.partial", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_deleted",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo.deleted", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_dot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/.", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_empty",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("empty filename"),
+			),
+		},
+		{
+			name:       "bad_filename_slash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/foo/bar", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("directories not supported"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("."), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_slash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("/"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_backslash",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("\\"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dotdot",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll(".."), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "bad_filename_encoded_dotdot_out",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("foo/../../../../../etc/passwd"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_spaces_and_caps",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("Foo Bar.dat"), strings.NewReader("baz")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasContents("Foo Bar.dat", "baz"),
+			),
+		},
+		{
+			name:       "put_unicode",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("Томас и его друзья.mp3"), strings.NewReader("главный озорник")),
+			checks: checks(
+				httpStatus(200),
+				bodyContains("{}"),
+				fileHasContents("Томас и его друзья.mp3", "главный озорник"),
+			),
+		},
+		{
+			name:       "put_invalid_utf8",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+(hexAll("😜")[:3]), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_null",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/%00", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_non_printable",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/%01", nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_colon",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll("nul:"), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+		{
+			name:       "put_invalid_surrounding_whitespace",
+			isSelf:     true,
+			capSharing: true,
+			req:        httptest.NewRequest("PUT", "/v0/put/"+hexAll(" foo "), nil),
+			checks: checks(
+				httpStatus(400),
+				bodyContains("bad filename"),
+			),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var e peerAPITestEnv
+			lb := &LocalBackend{
+				logf:           e.logf,
+				capFileSharing: tt.capSharing,
+			}
+			e.ph = &peerAPIHandler{
+				isSelf: tt.isSelf,
+				peerNode: &tailcfg.Node{
+					ComputedName: "some-peer-name",
+				},
+				ps: &peerAPIServer{
+					b: lb,
+				},
+			}
+			var rootDir string
+			if !tt.omitRoot {
+				rootDir = t.TempDir()
+				e.ph.ps.rootDir = rootDir
+			}
+			e.rr = httptest.NewRecorder()
+			e.ph.ServeHTTP(e.rr, tt.req)
+			for _, f := range tt.checks {
+				f(t, &e)
+			}
+			if t.Failed() && rootDir != "" {
+				t.Logf("Contents of %s:", rootDir)
+				des, _ := fs.ReadDir(os.DirFS(rootDir), ".")
+				for _, de := range des {
+					fi, err := de.Info()
+					if err != nil {
+						t.Log(err)
+					} else {
+						t.Logf("  %v %5d %s", fi.Mode(), fi.Size(), de.Name())
+					}
+				}
+			}
+		})
+	}
+}
+
+// Windows likes to hold on to file descriptors for some indeterminate
+// amount of time after you close them and not let you delete them for
+// a bit. So test that we work around that sufficiently.
+func TestFileDeleteRace(t *testing.T) {
+	dir := t.TempDir()
+	ps := &peerAPIServer{
+		b: &LocalBackend{
+			logf:           t.Logf,
+			capFileSharing: true,
+		},
+		rootDir: dir,
+	}
+	ph := &peerAPIHandler{
+		isSelf: true,
+		peerNode: &tailcfg.Node{
+			ComputedName: "some-peer-name",
+		},
+		ps: ps,
+	}
+	buf := make([]byte, 2<<20)
+	for i := 0; i < 30; i++ {
+		rr := httptest.NewRecorder()
+		ph.ServeHTTP(rr, httptest.NewRequest("PUT", "/v0/put/foo.txt", bytes.NewReader(buf[:rand.Intn(len(buf))])))
+		if res := rr.Result(); res.StatusCode != 200 {
+			t.Fatal(res.Status)
+		}
+		wfs, err := ps.WaitingFiles()
+		if err != nil {
+			t.Fatal(err)
+		}
+		if len(wfs) != 1 {
+			t.Fatalf("waiting files = %d; want 1", len(wfs))
+		}
+
+		if err := ps.DeleteFile("foo.txt"); err != nil {
+			t.Fatal(err)
+		}
+		wfs, err = ps.WaitingFiles()
+		if err != nil {
+			t.Fatal(err)
+		}
+		if len(wfs) != 0 {
+			t.Fatalf("waiting files = %d; want 0", len(wfs))
+		}
+	}
+}
+
+// Tests "foo.jpg.deleted" marks (for Windows).
+func TestDeletedMarkers(t *testing.T) {
+	dir := t.TempDir()
+	ps := &peerAPIServer{
+		b: &LocalBackend{
+			logf:           t.Logf,
+			capFileSharing: true,
+		},
+		rootDir: dir,
+	}
+
+	nothingWaiting := func() {
+		t.Helper()
+		ps.knownEmpty.Set(false)
+		if ps.hasFilesWaiting() {
+			t.Fatal("unexpected files waiting")
+		}
+	}
+	touch := func(base string) {
+		t.Helper()
+		if err := touchFile(filepath.Join(dir, base)); err != nil {
+			t.Fatal(err)
+		}
+	}
+	wantEmptyTempDir := func() {
+		t.Helper()
+		if fis, err := ioutil.ReadDir(dir); err != nil {
+			t.Fatal(err)
+		} else if len(fis) > 0 && runtime.GOOS != "windows" {
+			for _, fi := range fis {
+				t.Errorf("unexpected file in tempdir: %q", fi.Name())
+			}
+		}
+	}
+
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	nothingWaiting()
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	wf, err := ps.WaitingFiles()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(wf) != 0 {
+		t.Fatalf("WaitingFiles = %d; want 0", len(wf))
+	}
+	wantEmptyTempDir()
+
+	touch("foo.jpg.deleted")
+	touch("foo.jpg")
+	if rc, _, err := ps.OpenFile("foo.jpg"); err == nil {
+		rc.Close()
+		t.Fatal("unexpected foo.jpg open")
+	}
+	wantEmptyTempDir()
+
+	// And verify basics still work in non-deleted cases.
+	touch("foo.jpg")
+	touch("bar.jpg.deleted")
+	if wf, err := ps.WaitingFiles(); err != nil {
+		t.Error(err)
+	} else if len(wf) != 1 {
+		t.Errorf("WaitingFiles = %d; want 1", len(wf))
+	} else if wf[0].Name != "foo.jpg" {
+		t.Errorf("unexpected waiting file %+v", wf[0])
+	}
+	if rc, _, err := ps.OpenFile("foo.jpg"); err != nil {
+		t.Fatal(err)
+	} else {
+		rc.Close()
+	}
+
+}

ipn/ipnlocal/state_test.go

@@ -0,0 +1,859 @@
+// Copyright (c) 2021 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 ipnlocal
+
+import (
+	"context"
+	"sync"
+	"testing"
+	"time"
+
+	qt "github.com/frankban/quicktest"
+
+	"tailscale.com/control/controlclient"
+	"tailscale.com/ipn"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/empty"
+	"tailscale.com/types/logger"
+	"tailscale.com/types/netmap"
+	"tailscale.com/types/persist"
+	"tailscale.com/types/wgkey"
+	"tailscale.com/wgengine"
+)
+
+// notifyThrottler receives notifications from an ipn.Backend, blocking
+// (with eventual timeout and t.Fatal) if there are too many and complaining
+// (also with t.Fatal) if they are too few.
+type notifyThrottler struct {
+	t *testing.T
+
+	// ch gets replaced frequently. Lock the mutex before getting or
+	// setting it, but not while waiting on it.
+	mu sync.Mutex
+	ch chan ipn.Notify
+}
+
+// expect tells the throttler to expect count upcoming notifications.
+func (nt *notifyThrottler) expect(count int) {
+	nt.mu.Lock()
+	nt.ch = make(chan ipn.Notify, count)
+	nt.mu.Unlock()
+}
+
+// put adds one notification into the throttler's queue.
+func (nt *notifyThrottler) put(n ipn.Notify) {
+	nt.mu.Lock()
+	ch := nt.ch
+	nt.mu.Unlock()
+
+	select {
+	case ch <- n:
+		return
+	default:
+		nt.t.Fatalf("put: channel full: %v", n)
+	}
+}
+
+// drain pulls the notifications out of the queue, asserting that there are
+// exactly count notifications that have been put so far.
+func (nt *notifyThrottler) drain(count int) []ipn.Notify {
+	nt.mu.Lock()
+	ch := nt.ch
+	nt.mu.Unlock()
+
+	nn := []ipn.Notify{}
+	for i := 0; i < count; i++ {
+		select {
+		case n := <-ch:
+			nn = append(nn, n)
+		case <-time.After(6 * time.Second):
+			nt.t.Fatalf("drain: channel empty after %d/%d", i, count)
+		}
+	}
+
+	// no more notifications expected
+	close(ch)
+
+	return nn
+}
+
+// mockControl is a mock implementation of controlclient.Client.
+// Much of the backend state machine depends on callbacks and state
+// in the controlclient.Client, so by controlling it, we can check that
+// the state machine works as expected.
+type mockControl struct {
+	opts       controlclient.Options
+	logf       logger.Logf
+	statusFunc func(controlclient.Status)
+
+	mu          sync.Mutex
+	calls       []string
+	authBlocked bool
+	persist     persist.Persist
+	machineKey  wgkey.Private
+}
+
+func newMockControl() *mockControl {
+	return &mockControl{
+		calls:       []string{},
+		authBlocked: true,
+	}
+}
+
+func (cc *mockControl) SetStatusFunc(fn func(controlclient.Status)) {
+	cc.statusFunc = fn
+}
+
+func (cc *mockControl) populateKeys() (newKeys bool) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	if cc.machineKey.IsZero() {
+		cc.logf("Copying machineKey.")
+		cc.machineKey, _ = cc.opts.GetMachinePrivateKey()
+		newKeys = true
+	}
+
+	if cc.persist.PrivateNodeKey.IsZero() {
+		cc.logf("Generating a new nodekey.")
+		cc.persist.OldPrivateNodeKey = cc.persist.PrivateNodeKey
+		cc.persist.PrivateNodeKey, _ = wgkey.NewPrivate()
+		newKeys = true
+	}
+
+	return newKeys
+}
+
+// send publishes a controlclient.Status notification upstream.
+// (In our tests here, upstream is the ipnlocal.Local instance.)
+func (cc *mockControl) send(err error, url string, loginFinished bool, nm *netmap.NetworkMap) {
+	if cc.statusFunc != nil {
+		s := controlclient.Status{
+			URL:     url,
+			NetMap:  nm,
+			Persist: &cc.persist,
+		}
+		if err != nil {
+			s.Err = err.Error()
+		}
+		if loginFinished {
+			s.LoginFinished = &empty.Message{}
+		}
+		cc.statusFunc(s)
+	}
+}
+
+// called records that a particular function name was called.
+func (cc *mockControl) called(s string) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	cc.calls = append(cc.calls, s)
+}
+
+// getCalls returns the list of functions that have been called since the
+// last time getCalls was run.
+func (cc *mockControl) getCalls() []string {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	r := cc.calls
+	cc.calls = []string{}
+	return r
+}
+
+// setAuthBlocked changes the return value of AuthCantContinue.
+// Auth is blocked if you haven't called Login, the control server hasn't
+// provided an auth URL, or it has provided an auth URL and you haven't
+// visited it yet.
+func (cc *mockControl) setAuthBlocked(blocked bool) {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	cc.authBlocked = blocked
+}
+
+// Shutdown disconnects the client.
+//
+// Note that in a normal controlclient, Shutdown would be the last thing you
+// do before discarding the object. In this mock, we don't actually discard
+// the object, but if you see a call to Shutdown, you should always see a
+// call to New right after it, if the object continues to be used.
+// (Note that "New" is the ccGen function here; it means ipn.Backend wanted
+// to create an entirely new controlclient.)
+func (cc *mockControl) Shutdown() {
+	cc.logf("Shutdown")
+	cc.called("Shutdown")
+}
+
+// Login starts a login process.
+// Note that in this mock, we don't automatically generate notifications
+// about the progress of the login operation. You have to call setAuthBlocked()
+// and send() as required by the test.
+func (cc *mockControl) Login(t *tailcfg.Oauth2Token, flags controlclient.LoginFlags) {
+	cc.logf("Login token=%v flags=%v", t, flags)
+	cc.called("Login")
+	newKeys := cc.populateKeys()
+
+	interact := (flags & controlclient.LoginInteractive) != 0
+	cc.logf("Login: interact=%v newKeys=%v", interact, newKeys)
+	cc.setAuthBlocked(interact || newKeys)
+}
+
+func (cc *mockControl) StartLogout() {
+	cc.logf("StartLogout")
+	cc.called("StartLogout")
+}
+
+func (cc *mockControl) Logout(ctx context.Context) error {
+	cc.logf("Logout")
+	cc.called("Logout")
+	return nil
+}
+
+func (cc *mockControl) SetPaused(paused bool) {
+	cc.logf("SetPaused=%v", paused)
+	if paused {
+		cc.called("pause")
+	} else {
+		cc.called("unpause")
+	}
+}
+
+func (cc *mockControl) AuthCantContinue() bool {
+	cc.mu.Lock()
+	defer cc.mu.Unlock()
+
+	return cc.authBlocked
+}
+
+func (cc *mockControl) SetHostinfo(hi *tailcfg.Hostinfo) {
+	cc.logf("SetHostinfo: %v", *hi)
+	cc.called("SetHostinfo")
+}
+
+func (cc *mockControl) SetNetInfo(ni *tailcfg.NetInfo) {
+	cc.called("SetNetinfo")
+	cc.logf("SetNetInfo: %v", *ni)
+	cc.called("SetNetInfo")
+}
+
+func (cc *mockControl) UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint) {
+	// validate endpoint information here?
+	cc.logf("UpdateEndpoints: lp=%v ep=%v", localPort, endpoints)
+	cc.called("UpdateEndpoints")
+}
+
+// A very precise test of the sequence of function calls generated by
+// ipnlocal.Local into its controlclient instance, and the events it
+// produces upstream into the UI.
+//
+// [apenwarr] Normally I'm not a fan of "mock" style tests, but the precise
+// sequence of this state machine is so important for writing our multiple
+// frontends, that it's worth validating it all in one place.
+//
+// Any changes that affect this test will most likely require carefully
+// re-testing all our GUIs (and the CLI) to make sure we didn't break
+// anything.
+//
+// Note also that this test doesn't have any timers, goroutines, or duplicate
+// detection. It expects messages to be produced in exactly the right order,
+// with no duplicates, without doing network activity (other than through
+// controlclient, which we fake, so there's no network activity there either).
+//
+// TODO: A few messages that depend on magicsock (which actually might have
+// network delays) are just ignored for now, which makes the test
+// predictable, but maybe a bit less thorough. This is more of an overall
+// state machine test than a test of the wgengine+magicsock integration.
+func TestStateMachine(t *testing.T) {
+	c := qt.New(t)
+
+	logf := t.Logf
+	store := new(ipn.MemoryStore)
+	e, err := wgengine.NewFakeUserspaceEngine(logf, 0)
+	if err != nil {
+		t.Fatalf("NewFakeUserspaceEngine: %v", err)
+	}
+
+	cc := newMockControl()
+	b, err := NewLocalBackend(logf, "logid", store, e)
+	if err != nil {
+		t.Fatalf("NewLocalBackend: %v", err)
+	}
+	b.SetControlClientGetterForTesting(func(opts controlclient.Options) (controlclient.Client, error) {
+		cc.mu.Lock()
+		cc.opts = opts
+		cc.logf = opts.Logf
+		cc.authBlocked = true
+		cc.persist = cc.opts.Persist
+		cc.mu.Unlock()
+
+		cc.logf("ccGen: new mockControl.")
+		cc.called("New")
+		return cc, nil
+	})
+
+	notifies := &notifyThrottler{t: t}
+	notifies.expect(0)
+
+	b.SetNotifyCallback(func(n ipn.Notify) {
+		if n.State != nil ||
+			n.Prefs != nil ||
+			n.BrowseToURL != nil ||
+			n.LoginFinished != nil {
+			logf("\n%v\n\n", n)
+			notifies.put(n)
+		} else {
+			logf("\n(ignored) %v\n\n", n)
+		}
+	})
+
+	// Check that it hasn't called us right away.
+	// The state machine should be idle until we call Start().
+	c.Assert(cc.getCalls(), qt.HasLen, 0)
+
+	// Start the state machine.
+	// Since !WantRunning by default, it'll create a controlclient,
+	// but not ask it to do anything yet.
+	t.Logf("\n\nStart")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: strictly, it should pause, not unpause, here, since !WantRunning.
+		c.Assert([]string{"New", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		prefs := *nn[0].Prefs
+		// Note: a totally fresh system has Prefs.LoggedOut=false by
+		// default. We are logged out, but not because the user asked
+		// for it, so it doesn't count as Prefs.LoggedOut==true.
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Restart the state machine.
+	// It's designed to handle frontends coming and going sporadically.
+	// Make the sure the restart not only works, but generates the same
+	// events as the first time, so UIs always know what to expect.
+	t.Logf("\n\nStart2")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: strictly, it should pause, not unpause, here, since !WantRunning.
+		c.Assert([]string{"Shutdown", "New", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Start non-interactive login with no token.
+	// This will ask controlclient to start its own Login() process,
+	// then wait for us to respond.
+	t.Logf("\n\nLogin (noninteractive)")
+	notifies.expect(0)
+	b.Login(nil)
+	{
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"Login"})
+		notifies.drain(0)
+		// Note: WantRunning isn't true yet. It'll switch to true
+		// after a successful login finishes.
+		// (This behaviour is needed so that b.Login() won't
+		// start connecting to an old account right away, if one
+		// exists when you launch another login.)
+	}
+
+	// Attempted non-interactive login with no key; indicate that
+	// the user needs to visit a login URL.
+	t.Logf("\n\nLogin (url response)")
+	notifies.expect(1)
+	url1 := "http://localhost:1/1"
+	cc.send(nil, url1, false, nil)
+	{
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{})
+
+		// ...but backend eats that notification, because the user
+		// didn't explicitly request interactive login yet, and
+		// we're already in NeedsLogin state.
+		nn := notifies.drain(1)
+
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+	}
+
+	// Now we'll try an interactive login.
+	// Since we provided an interactive URL earlier, this shouldn't
+	// ask control to do anything. Instead backend will emit an event
+	// indicating that the UI should browse to the given URL.
+	t.Logf("\n\nLogin (interactive)")
+	notifies.expect(1)
+	b.StartLoginInteractive()
+	{
+		nn := notifies.drain(1)
+		// BUG: UpdateEndpoints shouldn't be called yet.
+		// We're still not logged in so there's nothing we can do
+		// with it. (And empirically, it's providing an empty list
+		// of endpoints.)
+		c.Assert([]string{"UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(url1, qt.Equals, *nn[0].BrowseToURL)
+	}
+
+	// Sometimes users press the Login button again, in the middle of
+	// a login sequence. For example, they might have closed their
+	// browser window without logging in, or they waited too long and
+	// the login URL expired. If they start another interactive login,
+	// we must always get a *new* login URL first.
+	t.Logf("\n\nLogin2 (interactive)")
+	notifies.expect(0)
+	b.StartLoginInteractive()
+	{
+		notifies.drain(0)
+		// backend asks control for another login sequence
+		c.Assert([]string{"Login"}, qt.DeepEquals, cc.getCalls())
+	}
+
+	// Provide a new interactive login URL.
+	t.Logf("\n\nLogin2 (url response)")
+	notifies.expect(1)
+	url2 := "http://localhost:1/2"
+	cc.send(nil, url2, false, nil)
+	{
+		// BUG: UpdateEndpoints again, this is getting silly.
+		c.Assert([]string{"UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+
+		// This time, backend should emit it to the UI right away,
+		// because the UI is anxiously awaiting a new URL to visit.
+		nn := notifies.drain(1)
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(url2, qt.Equals, *nn[0].BrowseToURL)
+	}
+
+	// Pretend that the interactive login actually happened.
+	// Controlclient always sends the netmap and LoginFinished at the
+	// same time.
+	// The backend should propagate this upward for the UI.
+	t.Logf("\n\nLoginFinished")
+	notifies.expect(3)
+	cc.setAuthBlocked(false)
+	cc.persist.LoginName = "user1"
+	cc.send(nil, "", true, &netmap.NetworkMap{})
+	{
+		nn := notifies.drain(3)
+		// BUG: still too soon for UpdateEndpoints.
+		//
+		// Arguably it makes sense to unpause now, since the machine
+		// authorization status is part of the netmap.
+		//
+		// BUG: backend unblocks wgengine at this point, even though
+		// our machine key is not authorized. It probably should
+		// wait until it gets into Starting.
+		// TODO: (Currently this test doesn't detect that bug, but
+		// it's visible in the logs)
+		c.Assert([]string{"unpause", "UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user1")
+		c.Assert(ipn.NeedsMachineAuth, qt.Equals, *nn[2].State)
+	}
+
+	// Pretend that the administrator has authorized our machine.
+	t.Logf("\n\nMachineAuthorized")
+	notifies.expect(1)
+	// BUG: the real controlclient sends LoginFinished with every
+	// notification while it's in StateAuthenticated, but not StateSynced.
+	// We should send it exactly once, or every time we're authenticated,
+	// but the current code is brittle.
+	// (ie. I suspect it would be better to change false->true in send()
+	// below, and do the same in the real controlclient.)
+	cc.send(nil, "", false, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause", "UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+	}
+
+	// TODO: add a fake DERP server to our fake netmap, so we can
+	// transition to the Running state here.
+
+	// TODO: test what happens when the admin forcibly deletes our key.
+	// (ie. unsolicited logout)
+
+	// TODO: test what happens when our key expires, client side.
+	// (and when it gets close to expiring)
+
+	// The user changes their preference to !WantRunning.
+	t.Logf("\n\nWantRunning -> false")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+	}
+
+	// The user changes their preference to WantRunning after all.
+	t.Logf("\n\nWantRunning -> true")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: true},
+	})
+	{
+		nn := notifies.drain(2)
+		// BUG: UpdateEndpoints isn't needed here.
+		// BUG: Login isn't needed here. We never logged out.
+		c.Assert([]string{"Login", "unpause", "UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+	}
+
+	// Test the fast-path frontend reconnection.
+	// This one is very finicky, so we have to force State==Running.
+	// TODO: actually get to State==Running, rather than cheating.
+	//  That'll require spinning up a fake DERP server and putting it in
+	//  the netmap.
+	t.Logf("\n\nFastpath Start()")
+	notifies.expect(1)
+	b.state = ipn.Running
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		nn := notifies.drain(1)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[0].NetMap, qt.Not(qt.IsNil))
+		// BUG: Prefs should be sent too, or the UI could end up in
+		// a bad state. (iOS, the only current user of this feature,
+		// probably wouldn't notice because it happens to not display
+		// any prefs. Maybe exit nodes will look weird?)
+	}
+
+	// undo the state hack above.
+	b.state = ipn.Starting
+
+	// User wants to logout.
+	t.Logf("\n\nLogout (async)")
+	notifies.expect(2)
+	b.Logout()
+	{
+		nn := notifies.drain(2)
+		// BUG: now is not the time to unpause.
+		c.Assert([]string{"unpause", "StartLogout"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[0].State)
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsTrue)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Let's make the logout succeed.
+	t.Logf("\n\nLogout (async) - succeed")
+	notifies.expect(0)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// A second logout should do nothing, since the prefs haven't changed.
+	t.Logf("\n\nLogout2 (async)")
+	notifies.expect(0)
+	b.Logout()
+	{
+		notifies.drain(0)
+		// BUG: the backend has already called StartLogout, and we're
+		// still logged out. So it shouldn't call it again.
+		c.Assert([]string{"StartLogout"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Let's acknowledge the second logout too.
+	t.Logf("\n\nLogout2 (async) - succeed")
+	notifies.expect(0)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Try the synchronous logout feature.
+	t.Logf("\n\nLogout3 (sync)")
+	notifies.expect(0)
+	b.LogoutSync(context.Background())
+	// NOTE: This returns as soon as cc.Logout() returns, which is okay
+	// I guess, since that's supposed to be synchronous.
+	{
+		notifies.drain(0)
+		c.Assert([]string{"Logout"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Generate the third logout event.
+	t.Logf("\n\nLogout3 (sync) - succeed")
+	notifies.expect(0)
+	cc.setAuthBlocked(true)
+	cc.send(nil, "", false, nil)
+	{
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Shut down the backend.
+	t.Logf("\n\nShutdown")
+	notifies.expect(0)
+	b.Shutdown()
+	{
+		notifies.drain(0)
+		// BUG: I expect a transition to ipn.NoState here.
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"Shutdown"})
+	}
+
+	// Oh, you thought we were done? Ha! Now we have to test what
+	// happens if the user exits and restarts while logged out.
+	// Note that it's explicitly okay to call b.Start() over and over
+	// again, every time the frontend reconnects.
+
+	// TODO: test user switching between statekeys.
+
+	// The frontend restarts!
+	t.Logf("\n\nStart3")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// BUG: We already called Shutdown(), no need to do it again.
+		// BUG: Way too soon for UpdateEndpoints.
+		// BUG: don't unpause because we're not logged in.
+		c.Assert([]string{"Shutdown", "New", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsTrue)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
+	}
+
+	// Let's break the rules a little. Our control server accepts
+	// your invalid login attempt, with no need for an interactive login.
+	// (This simulates an admin reviving a key that you previously
+	// disabled.)
+	t.Logf("\n\nLoginFinished3")
+	notifies.expect(3)
+	cc.setAuthBlocked(false)
+	cc.persist.LoginName = "user2"
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(3)
+		c.Assert([]string{"unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		// Prefs after finishing the login, so LoginName updated.
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user2")
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.Starting, qt.Equals, *nn[2].State)
+	}
+
+	// Now we've logged in successfully. Let's disconnect.
+	t.Logf("\n\nWantRunning -> false")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"pause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+	}
+
+	// One more restart, this time with a valid key, but WantRunning=false.
+	t.Logf("\n\nStart4")
+	notifies.expect(2)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// NOTE: cc.Shutdown() is correct here, since we didn't call
+		// b.Shutdown() explicitly ourselves.
+		// BUG: UpdateEndpoints should be called here since we're not WantRunning.
+		// Note: unpause happens because ipn needs to get at least one netmap
+		//  on startup, otherwise UIs can't show the node list, login
+		//  name, etc when in state ipn.Stopped.
+		//  Arguably they shouldn't try. But they currently do.
+		c.Assert([]string{"Shutdown", "New", "UpdateEndpoints", "Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(2)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(ipn.Stopped, qt.Equals, *nn[1].State)
+	}
+
+	// Request connection.
+	// The state machine didn't call Login() earlier, so now it needs to.
+	t.Logf("\n\nWantRunning4 -> true")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: true},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+	}
+
+	// Disconnect.
+	t.Logf("\n\nStop")
+	notifies.expect(2)
+	b.EditPrefs(&ipn.MaskedPrefs{
+		WantRunningSet: true,
+		Prefs:          ipn.Prefs{WantRunning: false},
+	})
+	{
+		nn := notifies.drain(2)
+		c.Assert([]string{"unpause"}, qt.DeepEquals, cc.getCalls())
+		// BUG: I would expect Prefs to change first, and state after.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(ipn.Stopped, qt.Equals, *nn[0].State)
+	}
+
+	// We want to try logging in as a different user, while Stopped.
+	// First, start the login process (without logging out first).
+	t.Logf("\n\nLoginDifferent")
+	notifies.expect(2)
+	b.StartLoginInteractive()
+	url3 := "http://localhost:1/3"
+	cc.send(nil, url3, false, nil)
+	{
+		nn := notifies.drain(2)
+		// It might seem like WantRunning should switch to true here,
+		// but that would be risky since we already have a valid
+		// user account. It might try to reconnect to the old account
+		// before the new one is ready. So no change yet.
+		c.Assert([]string{"Login", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(nn[1].State, qt.Not(qt.IsNil))
+		c.Assert(*nn[0].BrowseToURL, qt.Equals, url3)
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[1].State)
+	}
+
+	// Now, let's say the interactive login completed, using a different
+	// user account than before.
+	t.Logf("\n\nLoginDifferent URL visited")
+	notifies.expect(3)
+	cc.persist.LoginName = "user3"
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(3)
+		c.Assert([]string{"unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert(nn[1].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[2].State, qt.Not(qt.IsNil))
+		// Prefs after finishing the login, so LoginName updated.
+		c.Assert(nn[1].Prefs.Persist.LoginName, qt.Equals, "user3")
+		c.Assert(nn[1].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[1].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.Starting, qt.Equals, *nn[2].State)
+	}
+
+	// The last test case is the most common one: restarting when both
+	// logged in and WantRunning.
+	t.Logf("\n\nStart5")
+	notifies.expect(1)
+	c.Assert(b.Start(ipn.Options{StateKey: ipn.GlobalDaemonStateKey}), qt.IsNil)
+	{
+		// NOTE: cc.Shutdown() is correct here, since we didn't call
+		// b.Shutdown() ourselves.
+		c.Assert([]string{"Shutdown", "New", "UpdateEndpoints", "Login"}, qt.DeepEquals, cc.getCalls())
+
+		nn := notifies.drain(1)
+		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
+		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		c.Assert(nn[0].Prefs.WantRunning, qt.IsTrue)
+		c.Assert(ipn.NoState, qt.Equals, b.State())
+	}
+
+	// Control server accepts our valid key from before.
+	t.Logf("\n\nLoginFinished5")
+	notifies.expect(1)
+	cc.setAuthBlocked(false)
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause"}, qt.DeepEquals, cc.getCalls())
+		// NOTE: No LoginFinished message since no interactive
+		// login was needed.
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
+		// NOTE: No prefs change this time. WantRunning stays true.
+		// We were in Starting in the first place, so that doesn't
+		// change either.
+		c.Assert(ipn.Starting, qt.Equals, b.State())
+	}
+}

ipn/ipnserver/server.go

@@ -6,7 +6,9 @@ package ipnserver
 
 import (
 	"bufio"
+	"bytes"
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
@@ -145,7 +147,7 @@ func (s *server) getConnIdentity(c net.Conn) (ci connIdentity, err error) {
 	if err != nil {
 		return ci, fmt.Errorf("parsing local remote: %w", err)
 	}
-	if !la.IP.IsLoopback() || !ra.IP.IsLoopback() {
+	if !la.IP().IsLoopback() || !ra.IP().IsLoopback() {
 		return ci, errors.New("non-loopback connection")
 	}
 	tab, err := netstat.Get()
@@ -251,8 +253,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 			return
 		}
 		defer c.Close()
-		serverToClient := func(b []byte) { ipn.WriteMsg(c, b) }
-		bs := ipn.NewBackendServer(logf, nil, serverToClient)
+		bs := ipn.NewBackendServer(logf, nil, jsonNotifier(c, s.logf))
 		_, occupied := err.(inUseOtherUserError)
 		if occupied {
 			bs.SendInUseOtherUserErrorMessage(err.Error())
@@ -287,7 +288,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 	defer s.removeAndCloseConn(c)
 	logf("[v1] incoming control connection")
 
-	if isReadonlyConn(ci, logf) {
+	if isReadonlyConn(ci, s.b.OperatorUserID(), logf) {
 		ctx = ipn.ReadonlyContextOf(ctx)
 	}
 
@@ -313,7 +314,7 @@ func (s *server) serveConn(ctx context.Context, c net.Conn, logf logger.Logf) {
 	}
 }
 
-func isReadonlyConn(ci connIdentity, logf logger.Logf) bool {
+func isReadonlyConn(ci connIdentity, operatorUID string, logf logger.Logf) bool {
 	if runtime.GOOS == "windows" {
 		// Windows doesn't need/use this mechanism, at least yet. It
 		// has a different last-user-wins auth model.
@@ -342,6 +343,10 @@ func isReadonlyConn(ci connIdentity, logf logger.Logf) bool {
 		logf("connection from userid %v; connection from non-root user matching daemon has access", uid)
 		return rw
 	}
+	if operatorUID != "" && uid == operatorUID {
+		logf("connection from userid %v; is configured operator", uid)
+		return rw
+	}
 	var adminGroupID string
 	switch runtime.GOOS {
 	case "darwin":
@@ -435,7 +440,7 @@ func (s *server) localAPIPermissions(ci connIdentity) (read, write bool) {
 		return false, false
 	}
 	if ci.IsUnixSock {
-		return true, !isReadonlyConn(ci, logger.Discard)
+		return true, !isReadonlyConn(ci, s.b.OperatorUserID(), logger.Discard)
 	}
 	return false, false
 }
@@ -472,9 +477,7 @@ func (s *server) addConn(c net.Conn, isHTTP bool) (ci connIdentity, err error) {
 	defer func() {
 		if doReset {
 			s.logf("identity changed; resetting server")
-			s.bsMu.Lock()
-			s.bs.Reset(context.TODO())
-			s.bsMu.Unlock()
+			s.b.ResetForClientDisconnect()
 		}
 	}()
 
@@ -524,9 +527,7 @@ func (s *server) removeAndCloseConn(c net.Conn) {
 			s.logf("client disconnected; staying alive in server mode")
 		} else {
 			s.logf("client disconnected; stopping server")
-			s.bsMu.Lock()
-			s.bs.Reset(context.TODO())
-			s.bsMu.Unlock()
+			s.b.ResetForClientDisconnect()
 		}
 	}
 	c.Close()
@@ -567,7 +568,9 @@ func (s *server) setServerModeUserLocked() {
 	}
 }
 
-func (s *server) writeToClients(b []byte) {
+var jsonEscapedZero = []byte(`\u0000`)
+
+func (s *server) writeToClients(n ipn.Notify) {
 	inServerMode := s.b.InServerMode()
 
 	s.mu.Lock()
@@ -584,14 +587,24 @@ func (s *server) writeToClients(b []byte) {
 		}
 	}
 
-	for c := range s.clients {
-		ipn.WriteMsg(c, b)
+	if len(s.clients) == 0 {
+		// Common case (at least on busy servers): nobody
+		// connected (no GUI, etc), so return before
+		// serializing JSON.
+		return
+	}
+
+	if b, ok := marshalNotify(n, s.logf); ok {
+		for c := range s.clients {
+			ipn.WriteMsg(c, b)
+		}
 	}
 }
 
 // Run runs a Tailscale backend service.
 // The getEngine func is called repeatedly, once per connection, until it returns an engine successfully.
 func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (wgengine.Engine, error), opts Options) error {
+	getEngine = getEngineUntilItWorksWrapper(getEngine)
 	runDone := make(chan struct{})
 	defer close(runDone)
 
@@ -652,46 +665,6 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 	eng, err := getEngine()
 	if err != nil {
 		logf("ipnserver: initial getEngine call: %v", err)
-
-		// Issue 1187: on Windows, in unattended mode,
-		// sometimes we try 5 times and fail to create the
-		// engine before the system's ready. Hack until the
-		// bug if fixed properly: if we're running in
-		// unattended mode on Windows, keep trying forever,
-		// waiting for the machine to be ready (networking to
-		// come up?) and then dial our own safesocket TCP
-		// listener to wake up the usual mechanism that lets
-		// us surface getEngine errors to UI clients. (We
-		// don't want to just call getEngine in a loop without
-		// the listener.Accept, as we do want to handle client
-		// connections so we can tell them about errors)
-
-		bootRaceWaitForEngine, bootRaceWaitForEngineCancel := context.WithTimeout(context.Background(), time.Minute)
-		if runtime.GOOS == "windows" && opts.AutostartStateKey != "" {
-			logf("ipnserver: in unattended mode, waiting for engine availability")
-			getEngine = getEngineUntilItWorksWrapper(getEngine)
-			// Wait for it to be ready.
-			go func() {
-				defer bootRaceWaitForEngineCancel()
-				t0 := time.Now()
-				for {
-					time.Sleep(10 * time.Second)
-					if _, err := getEngine(); err != nil {
-						logf("ipnserver: unattended mode engine load: %v", err)
-						continue
-					}
-					c, err := net.Dial("tcp", listen.Addr().String())
-					logf("ipnserver: engine created after %v; waking up Accept: Dial error: %v", time.Since(t0).Round(time.Second), err)
-					if err == nil {
-						c.Close()
-					}
-					break
-				}
-			}()
-		} else {
-			bootRaceWaitForEngineCancel()
-		}
-
 		for i := 1; ctx.Err() == nil; i++ {
 			c, err := listen.Accept()
 			if err != nil {
@@ -699,7 +672,6 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 				bo.BackOff(ctx, err)
 				continue
 			}
-			<-bootRaceWaitForEngine.Done()
 			logf("ipnserver: try%d: trying getEngine again...", i)
 			eng, err = getEngine()
 			if err == nil {
@@ -711,8 +683,7 @@ func Run(ctx context.Context, logf logger.Logf, logid string, getEngine func() (
 			errMsg := err.Error()
 			go func() {
 				defer c.Close()
-				serverToClient := func(b []byte) { ipn.WriteMsg(c, b) }
-				bs := ipn.NewBackendServer(logf, nil, serverToClient)
+				bs := ipn.NewBackendServer(logf, nil, jsonNotifier(c, logf))
 				bs.SendErrorMessage(errMsg)
 				time.Sleep(time.Second)
 			}()
@@ -1002,3 +973,25 @@ func peerPid(entries []netstat.Entry, la, ra netaddr.IPPort) int {
 	}
 	return 0
 }
+
+// jsonNotifier returns a notify-writer func that writes ipn.Notify
+// messages to w.
+func jsonNotifier(w io.Writer, logf logger.Logf) func(ipn.Notify) {
+	return func(n ipn.Notify) {
+		if b, ok := marshalNotify(n, logf); ok {
+			ipn.WriteMsg(w, b)
+		}
+	}
+}
+
+func marshalNotify(n ipn.Notify, logf logger.Logf) (b []byte, ok bool) {
+	b, err := json.Marshal(n)
+	if err != nil {
+		logf("ipnserver: [unexpected] error serializing JSON: %v", err)
+		return nil, false
+	}
+	if bytes.Contains(b, jsonEscapedZero) {
+		logf("[unexpected] zero byte in BackendServer.send notify message: %q", b)
+	}
+	return b, true
+}

ipn/ipnstate/ipnstate.go

@@ -65,6 +65,7 @@ type PeerStatusLite struct {
 }
 
 type PeerStatus struct {
+	ID        tailcfg.StableNodeID
 	PublicKey key.Public
 	HostName  string // HostInfo's Hostname (not a DNS name or necessarily unique)
 	DNSName   string
@@ -203,6 +204,9 @@ func (sb *StatusBuilder) AddPeer(peer key.Public, st *PeerStatus) {
 		return
 	}
 
+	if v := st.ID; v != "" {
+		e.ID = v
+	}
 	if v := st.HostName; v != "" {
 		e.HostName = v
 	}

ipn/message.go

@@ -88,31 +88,36 @@ type Command struct {
 
 type BackendServer struct {
 	logf          logger.Logf
-	b             Backend              // the Backend we are serving up
-	sendNotifyMsg func(jsonMsg []byte) // send a notification message
-	GotQuit       bool                 // a Quit command was received
+	b             Backend      // the Backend we are serving up
+	sendNotifyMsg func(Notify) // send a notification message
+	GotQuit       bool         // a Quit command was received
 }
 
-func NewBackendServer(logf logger.Logf, b Backend, sendNotifyMsg func(b []byte)) *BackendServer {
+// NewBackendServer creates a new BackendServer using b.
+//
+// If sendNotifyMsg is non-nil, it additionally sets the Backend's
+// notification callback to call the func with ipn.Notify messages in
+// JSON form. If nil, it does not change the notification callback.
+func NewBackendServer(logf logger.Logf, b Backend, sendNotifyMsg func(Notify)) *BackendServer {
 	bs := &BackendServer{
 		logf:          logf,
 		b:             b,
 		sendNotifyMsg: sendNotifyMsg,
 	}
-	b.SetNotifyCallback(bs.send)
+	// b may be nil if the BackendServer is being created just to
+	// encapsulate and send an error message.
+	if sendNotifyMsg != nil && b != nil {
+		b.SetNotifyCallback(bs.send)
+	}
 	return bs
 }
 
 func (bs *BackendServer) send(n Notify) {
+	if bs.sendNotifyMsg == nil {
+		return
+	}
 	n.Version = version.Long
-	b, err := json.Marshal(n)
-	if err != nil {
-		log.Fatalf("Failed json.Marshal(notify): %v\n%#v", err, n)
-	}
-	if bytes.Contains(b, jsonEscapedZero) {
-		log.Printf("[unexpected] zero byte in BackendServer.send notify message: %q", b)
-	}
-	bs.sendNotifyMsg(b)
+	bs.sendNotifyMsg(n)
 }
 
 func (bs *BackendServer) SendErrorMessage(msg string) {
@@ -143,11 +148,6 @@ func (bs *BackendServer) GotCommandMsg(ctx context.Context, b []byte) error {
 	return bs.GotCommand(ctx, cmd)
 }
 
-func (bs *BackendServer) GotFakeCommand(ctx context.Context, cmd *Command) error {
-	cmd.Version = version.Long
-	return bs.GotCommand(ctx, cmd)
-}
-
 // ErrMsgPermissionDenied is the Notify.ErrMessage value used an
 // operation was done from a user/context that didn't have permission.
 const ErrMsgPermissionDenied = "permission denied"
@@ -211,12 +211,6 @@ func (bs *BackendServer) GotCommand(ctx context.Context, cmd *Command) error {
 	return fmt.Errorf("BackendServer.Do: no command specified")
 }
 
-func (bs *BackendServer) Reset(ctx context.Context) error {
-	// Tell the backend we got a Logout command, which will cause it
-	// to forget all its authentication information.
-	return bs.GotFakeCommand(ctx, &Command{Logout: &NoArgs{}})
-}
-
 type BackendClient struct {
 	logf           logger.Logf
 	sendCommandMsg func(jsonb []byte)

ipn/message_test.go

@@ -7,6 +7,7 @@ package ipn
 import (
 	"bytes"
 	"context"
+	"encoding/json"
 	"testing"
 	"time"
 
@@ -74,7 +75,11 @@ func TestClientServer(t *testing.T) {
 			bc.GotNotifyMsg(b)
 		}
 	}()
-	serverToClient := func(b []byte) {
+	serverToClient := func(n Notify) {
+		b, err := json.Marshal(n)
+		if err != nil {
+			panic(err.Error())
+		}
 		serverToClientCh <- append([]byte{}, b...)
 	}
 	clientToServer := func(b []byte) {
@@ -87,6 +92,8 @@ func TestClientServer(t *testing.T) {
 		t.Logf("c: "+fmt, args...)
 	}
 	bs = NewBackendServer(slogf, b, serverToClient)
+	// Verify that this doesn't break bs's callback:
+	NewBackendServer(slogf, b, nil)
 	bc = NewBackendClient(clogf, clientToServer)
 
 	ch := make(chan Notify, 256)
@@ -180,3 +187,17 @@ func TestClientServer(t *testing.T) {
 	})
 	flushUntil(Running)
 }
+
+func TestNilBackend(t *testing.T) {
+	var called *Notify
+	bs := NewBackendServer(t.Logf, nil, func(n Notify) {
+		called = &n
+	})
+	bs.SendErrorMessage("Danger, Will Robinson!")
+	if called == nil {
+		t.Errorf("expect callback to be called, wasn't")
+	}
+	if called.ErrMessage == nil || *called.ErrMessage != "Danger, Will Robinson!" {
+		t.Errorf("callback got wrong error: %v", called.ErrMessage)
+	}
+}

ipn/prefs.go

@@ -25,9 +25,27 @@ import (
 
 //go:generate go run tailscale.com/cmd/cloner -type=Prefs -output=prefs_clone.go
 
+// DefaultControlURL returns the URL base of the control plane
+// ("coordination server") for use when no explicit one is configured.
+// The default control plane is the hosted version run by Tailscale.com.
+const DefaultControlURL = "https://login.tailscale.com"
+
 // Prefs are the user modifiable settings of the Tailscale node agent.
 type Prefs struct {
 	// ControlURL is the URL of the control server to use.
+	//
+	// If empty, the default for new installs, DefaultControlURL
+	// is used. It's set non-empty once the daemon has been started
+	// for the first time.
+	//
+	// TODO(apenwarr): Make it safe to update this with SetPrefs().
+	// Right now, you have to pass it in the initial prefs in Start(),
+	// which is the only code that actually uses the ControlURL value.
+	// It would be more consistent to restart controlclient
+	// automatically whenever this variable changes.
+	//
+	// Meanwhile, you have to provide this as part of Options.Prefs or
+	// Options.UpdatePrefs when calling Backend.Start().
 	ControlURL string
 
 	// RouteAll specifies whether to accept subnets advertised by
@@ -78,6 +96,14 @@ type Prefs struct {
 	// this node.
 	WantRunning bool
 
+	// LoggedOut indicates whether the user intends to be logged out.
+	// There are other reasons we may be logged out, including no valid
+	// keys.
+	// We need to remember this state so that, on next startup, we can
+	// generate the "Login" vs "Connect" buttons correctly, without having
+	// to contact the server to confirm our nodekey status first.
+	LoggedOut bool
+
 	// ShieldsUp indicates whether to block all incoming connections,
 	// regardless of the control-provided packet filter. If false, we
 	// use the packet filter as provided. If true, we block incoming
@@ -144,6 +170,10 @@ type Prefs struct {
 	// Tailscale, if at all.
 	NetfilterMode preftype.NetfilterMode
 
+	// OperatorUser is the local machine user name who is allowed to
+	// operate tailscaled without being root or using sudo.
+	OperatorUser string `json:",omitempty"`
+
 	// The Persist field is named 'Config' in the file for backward
 	// compatibility with earlier versions.
 	// TODO(apenwarr): We should move this out of here, it's not a pref.
@@ -164,6 +194,7 @@ type MaskedPrefs struct {
 	ExitNodeAllowLANAccessSet bool `json:",omitempty"`
 	CorpDNSSet                bool `json:",omitempty"`
 	WantRunningSet            bool `json:",omitempty"`
+	LoggedOutSet              bool `json:",omitempty"`
 	ShieldsUpSet              bool `json:",omitempty"`
 	AdvertiseTagsSet          bool `json:",omitempty"`
 	HostnameSet               bool `json:",omitempty"`
@@ -174,6 +205,7 @@ type MaskedPrefs struct {
 	AdvertiseRoutesSet        bool `json:",omitempty"`
 	NoSNATSet                 bool `json:",omitempty"`
 	NetfilterModeSet          bool `json:",omitempty"`
+	OperatorUserSet           bool `json:",omitempty"`
 }
 
 // ApplyEdits mutates p, assigning fields from m.Prefs for each MaskedPrefs
@@ -232,6 +264,9 @@ func (p *Prefs) pretty(goos string) string {
 		sb.WriteString("mesh=false ")
 	}
 	fmt.Fprintf(&sb, "dns=%v want=%v ", p.CorpDNS, p.WantRunning)
+	if p.LoggedOut {
+		sb.WriteString("loggedout=true ")
+	}
 	if p.ForceDaemon {
 		sb.WriteString("server=true ")
 	}
@@ -258,12 +293,15 @@ func (p *Prefs) pretty(goos string) string {
 	if goos == "linux" {
 		fmt.Fprintf(&sb, "nf=%v ", p.NetfilterMode)
 	}
-	if p.ControlURL != "" && p.ControlURL != "https://login.tailscale.com" {
+	if p.ControlURL != "" && p.ControlURL != DefaultControlURL {
 		fmt.Fprintf(&sb, "url=%q ", p.ControlURL)
 	}
 	if p.Hostname != "" {
 		fmt.Fprintf(&sb, "host=%q ", p.Hostname)
 	}
+	if p.OperatorUser != "" {
+		fmt.Fprintf(&sb, "op=%q ", p.OperatorUser)
+	}
 	if p.Persist != nil {
 		sb.WriteString(p.Persist.Pretty())
 	} else {
@@ -298,10 +336,12 @@ func (p *Prefs) Equals(p2 *Prefs) bool {
 		p.ExitNodeAllowLANAccess == p2.ExitNodeAllowLANAccess &&
 		p.CorpDNS == p2.CorpDNS &&
 		p.WantRunning == p2.WantRunning &&
+		p.LoggedOut == p2.LoggedOut &&
 		p.NotepadURLs == p2.NotepadURLs &&
 		p.ShieldsUp == p2.ShieldsUp &&
 		p.NoSNAT == p2.NoSNAT &&
 		p.NetfilterMode == p2.NetfilterMode &&
+		p.OperatorUser == p2.OperatorUser &&
 		p.Hostname == p2.Hostname &&
 		p.OSVersion == p2.OSVersion &&
 		p.DeviceModel == p2.DeviceModel &&
@@ -335,12 +375,19 @@ func compareStrings(a, b []string) bool {
 	return true
 }
 
+// NewPrefs returns the default preferences to use.
 func NewPrefs() *Prefs {
+	// Provide default values for options which might be missing
+	// from the json data for any reason. The json can still
+	// override them to false.
 	return &Prefs{
-		// Provide default values for options which might be missing
-		// from the json data for any reason. The json can still
-		// override them to false.
-		ControlURL:       "https://login.tailscale.com",
+		// ControlURL is explicitly not set to signal that
+		// it's not yet configured, which relaxes the CLI "up"
+		// safety net features. It will get set to DefaultControlURL
+		// on first up. Or, if not, DefaultControlURL will be used
+		// later anyway.
+		ControlURL: "",
+
 		RouteAll:         true,
 		AllowSingleHosts: true,
 		CorpDNS:          true,
@@ -349,6 +396,15 @@ func NewPrefs() *Prefs {
 	}
 }
 
+// ControlURLOrDefault returns the coordination server's URL base.
+// If not configured, DefaultControlURL is returned instead.
+func (p *Prefs) ControlURLOrDefault() string {
+	if p.ControlURL != "" {
+		return p.ControlURL
+	}
+	return DefaultControlURL
+}
+
 // PrefsFromBytes deserializes Prefs from a JSON blob. If
 // enforceDefaults is true, Prefs.RouteAll and Prefs.AllowSingleHosts
 // are forced on.

ipn/prefs_clone.go

@@ -41,6 +41,7 @@ var _PrefsNeedsRegeneration = Prefs(struct {
 	ExitNodeAllowLANAccess bool
 	CorpDNS                bool
 	WantRunning            bool
+	LoggedOut              bool
 	ShieldsUp              bool
 	AdvertiseTags          []string
 	Hostname               string
@@ -51,5 +52,6 @@ var _PrefsNeedsRegeneration = Prefs(struct {
 	AdvertiseRoutes        []netaddr.IPPrefix
 	NoSNAT                 bool
 	NetfilterMode          preftype.NetfilterMode
+	OperatorUser           string
 	Persist                *persist.Persist
 }{})

ipn/prefs_test.go

@@ -33,7 +33,29 @@ func fieldsOf(t reflect.Type) (fields []string) {
 func TestPrefsEqual(t *testing.T) {
 	tstest.PanicOnLog()
 
-	prefsHandles := []string{"ControlURL", "RouteAll", "AllowSingleHosts", "ExitNodeID", "ExitNodeIP", "ExitNodeAllowLANAccess", "CorpDNS", "WantRunning", "ShieldsUp", "AdvertiseTags", "Hostname", "OSVersion", "DeviceModel", "NotepadURLs", "ForceDaemon", "AdvertiseRoutes", "NoSNAT", "NetfilterMode", "Persist"}
+	prefsHandles := []string{
+		"ControlURL",
+		"RouteAll",
+		"AllowSingleHosts",
+		"ExitNodeID",
+		"ExitNodeIP",
+		"ExitNodeAllowLANAccess",
+		"CorpDNS",
+		"WantRunning",
+		"LoggedOut",
+		"ShieldsUp",
+		"AdvertiseTags",
+		"Hostname",
+		"OSVersion",
+		"DeviceModel",
+		"NotepadURLs",
+		"ForceDaemon",
+		"AdvertiseRoutes",
+		"NoSNAT",
+		"NetfilterMode",
+		"OperatorUser",
+		"Persist",
+	}
 	if have := fieldsOf(reflect.TypeOf(Prefs{})); !reflect.DeepEqual(have, prefsHandles) {
 		t.Errorf("Prefs.Equal check might be out of sync\nfields: %q\nhandled: %q\n",
 			have, prefsHandles)

logtail/filch/filch.go

@@ -15,6 +15,7 @@ import (
 	"sync"
 )
 
+//lint:ignore U1000 work around false positive: https://github.com/dominikh/go-tools/issues/983
 var stderrFD = 2 // a variable for testing
 
 type Options struct {

logtail/logtail.go

@@ -15,6 +15,7 @@ import (
 	"net/http"
 	"os"
 	"strconv"
+	"sync/atomic"
 	"time"
 
 	"tailscale.com/logtail/backoff"
@@ -72,7 +73,7 @@ func NewLogger(cfg Config, logf tslogger.Logf) *Logger {
 	}
 	l := &Logger{
 		stderr:         cfg.Stderr,
-		stderrLevel:    cfg.StderrLevel,
+		stderrLevel:    int64(cfg.StderrLevel),
 		httpc:          cfg.HTTPC,
 		url:            cfg.BaseURL + "/c/" + cfg.Collection + "/" + cfg.PrivateID.String(),
 		lowMem:         cfg.LowMemory,
@@ -103,7 +104,7 @@ func NewLogger(cfg Config, logf tslogger.Logf) *Logger {
 // logging facilities and uploading to a log server.
 type Logger struct {
 	stderr         io.Writer
-	stderrLevel    int
+	stderrLevel    int64 // accessed atomically
 	httpc          *http.Client
 	url            string
 	lowMem         bool
@@ -125,10 +126,8 @@ type Logger struct {
 // SetVerbosityLevel controls the verbosity level that should be
 // written to stderr. 0 is the default (not verbose). Levels 1 or higher
 // are increasingly verbose.
-//
-// It should not be changed concurrently with log writes.
 func (l *Logger) SetVerbosityLevel(level int) {
-	l.stderrLevel = level
+	atomic.StoreInt64(&l.stderrLevel, int64(level))
 }
 
 // SetLinkMonitor sets the optional the link monitor.
@@ -514,7 +513,7 @@ func (l *Logger) Write(buf []byte) (int, error) {
 		return 0, nil
 	}
 	level, buf := parseAndRemoveLogLevel(buf)
-	if l.stderr != nil && l.stderr != ioutil.Discard && level <= l.stderrLevel {
+	if l.stderr != nil && l.stderr != ioutil.Discard && int64(level) <= atomic.LoadInt64(&l.stderrLevel) {
 		if buf[len(buf)-1] == '\n' {
 			l.stderr.Write(buf)
 		} else {

net/dns/config.go

@@ -22,27 +22,26 @@ type Config struct {
 	// for queries that fall within that suffix.
 	// If a query doesn't match any entry in Routes, the
 	// DefaultResolvers are used.
+	// A Routes entry with no resolvers means the route should be
+	// authoritatively answered using the contents of Hosts.
 	Routes map[dnsname.FQDN][]netaddr.IPPort
 	// SearchDomains are DNS suffixes to try when expanding
 	// single-label queries.
 	SearchDomains []dnsname.FQDN
 	// Hosts maps DNS FQDNs to their IPs, which can be a mix of IPv4
 	// and IPv6.
-	// Queries matching entries in Hosts are resolved locally without
-	// recursing off-machine.
+	// Queries matching entries in Hosts are resolved locally by
+	// 100.100.100.100 without leaving the machine.
+	// Adding an entry to Hosts merely creates the record. If you want
+	// it to resolve, you also need to add appropriate routes to
+	// Routes.
 	Hosts map[dnsname.FQDN][]netaddr.IP
-	// AuthoritativeSuffixes is a list of fully-qualified DNS suffixes
-	// for which the in-process Tailscale resolver is authoritative.
-	// Queries for names within AuthoritativeSuffixes can only be
-	// fulfilled by entries in Hosts. Queries with no match in Hosts
-	// return NXDOMAIN.
-	AuthoritativeSuffixes []dnsname.FQDN
 }
 
 // needsAnyResolvers reports whether c requires a resolver to be set
 // at the OS level.
 func (c Config) needsOSResolver() bool {
-	return c.hasDefaultResolvers() || c.hasRoutes() || c.hasHosts()
+	return c.hasDefaultResolvers() || c.hasRoutes()
 }
 
 func (c Config) hasRoutes() bool {
@@ -52,7 +51,7 @@ func (c Config) hasRoutes() bool {
 // hasDefaultResolversOnly reports whether the only resolvers in c are
 // DefaultResolvers.
 func (c Config) hasDefaultResolversOnly() bool {
-	return c.hasDefaultResolvers() && !c.hasRoutes() && !c.hasHosts()
+	return c.hasDefaultResolvers() && !c.hasRoutes()
 }
 
 func (c Config) hasDefaultResolvers() bool {
@@ -63,44 +62,28 @@ func (c Config) hasDefaultResolvers() bool {
 // routes use the same resolvers, or nil if multiple sets of resolvers
 // are specified.
 func (c Config) singleResolverSet() []netaddr.IPPort {
-	var first []netaddr.IPPort
+	var (
+		prev            []netaddr.IPPort
+		prevInitialized bool
+	)
 	for _, resolvers := range c.Routes {
-		if first == nil {
-			first = resolvers
+		if !prevInitialized {
+			prev = resolvers
+			prevInitialized = true
 			continue
 		}
-		if !sameIPPorts(first, resolvers) {
+		if !sameIPPorts(prev, resolvers) {
 			return nil
 		}
 	}
-	return first
+	return prev
 }
 
-// hasHosts reports whether c requires resolution of MagicDNS hosts or
-// domains.
-func (c Config) hasHosts() bool {
-	return len(c.Hosts) > 0 || len(c.AuthoritativeSuffixes) > 0
-}
-
-// matchDomains returns the list of match suffixes needed by Routes,
-// AuthoritativeSuffixes. Hosts is not considered as we assume that
-// they're covered by AuthoritativeSuffixes for now.
+// matchDomains returns the list of match suffixes needed by Routes.
 func (c Config) matchDomains() []dnsname.FQDN {
-	ret := make([]dnsname.FQDN, 0, len(c.Routes)+len(c.AuthoritativeSuffixes))
-	seen := map[dnsname.FQDN]bool{}
-	for _, suffix := range c.AuthoritativeSuffixes {
-		if seen[suffix] {
-			continue
-		}
-		ret = append(ret, suffix)
-		seen[suffix] = true
-	}
+	ret := make([]dnsname.FQDN, 0, len(c.Routes))
 	for suffix := range c.Routes {
-		if seen[suffix] {
-			continue
-		}
 		ret = append(ret, suffix)
-		seen[suffix] = true
 	}
 	sort.Slice(ret, func(i, j int) bool {
 		return ret[i].WithTrailingDot() < ret[j].WithTrailingDot()

net/dns/debian_resolvconf.go

@@ -2,6 +2,8 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// +build linux freebsd openbsd
+
 package dns
 
 import (
@@ -89,6 +91,15 @@ func newDebianResolvconfManager(logf logger.Logf) (*resolvconfManager, error) {
 	return ret, nil
 }
 
+func (m *resolvconfManager) deleteTailscaleConfig() error {
+	cmd := exec.Command("resolvconf", "-d", resolvconfConfigName)
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		return fmt.Errorf("running %s: %s", cmd, out)
+	}
+	return nil
+}
+
 func (m *resolvconfManager) SetDNS(config OSConfig) error {
 	if !m.scriptInstalled {
 		m.logf("injecting resolvconf workaround script")
@@ -101,18 +112,24 @@ func (m *resolvconfManager) SetDNS(config OSConfig) error {
 		m.scriptInstalled = true
 	}
 
-	stdin := new(bytes.Buffer)
-	writeResolvConf(stdin, config.Nameservers, config.SearchDomains) // dns_direct.go
+	if config.IsZero() {
+		if err := m.deleteTailscaleConfig(); err != nil {
+			return err
+		}
+	} else {
+		stdin := new(bytes.Buffer)
+		writeResolvConf(stdin, config.Nameservers, config.SearchDomains) // dns_direct.go
 
-	// This resolvconf implementation doesn't support exclusive mode
-	// or interface priorities, so it will end up blending our
-	// configuration with other sources. However, this will get fixed
-	// up by the script we injected above.
-	cmd := exec.Command("resolvconf", "-a", resolvconfConfigName)
-	cmd.Stdin = stdin
-	out, err := cmd.CombinedOutput()
-	if err != nil {
-		return fmt.Errorf("running %s: %s", cmd, out)
+		// This resolvconf implementation doesn't support exclusive
+		// mode or interface priorities, so it will end up blending
+		// our configuration with other sources. However, this will
+		// get fixed up by the script we injected above.
+		cmd := exec.Command("resolvconf", "-a", resolvconfConfigName)
+		cmd.Stdin = stdin
+		out, err := cmd.CombinedOutput()
+		if err != nil {
+			return fmt.Errorf("running %s: %s", cmd, out)
+		}
 	}
 
 	return nil
@@ -156,10 +173,8 @@ func (m *resolvconfManager) GetBaseConfig() (OSConfig, error) {
 }
 
 func (m *resolvconfManager) Close() error {
-	cmd := exec.Command("resolvconf", "-d", resolvconfConfigName)
-	out, err := cmd.CombinedOutput()
-	if err != nil {
-		return fmt.Errorf("running %s: %s", cmd, out)
+	if err := m.deleteTailscaleConfig(); err != nil {
+		return err
 	}
 
 	if m.scriptInstalled {

net/dns/direct.go

@@ -2,6 +2,8 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// +build linux freebsd openbsd
+
 package dns
 
 import (
@@ -201,15 +203,53 @@ func (m directManager) backupConfig() error {
 	return os.Rename(resolvConf, backupConf)
 }
 
-func (m directManager) SetDNS(config OSConfig) error {
-	if err := m.backupConfig(); err != nil {
+func (m directManager) restoreBackup() error {
+	if _, err := os.Stat(backupConf); err != nil {
+		if os.IsNotExist(err) {
+			// No backup, nothing we can do.
+			return nil
+		}
+		return err
+	}
+	owned, err := m.ownedByTailscale()
+	if err != nil {
+		return err
+	}
+	if _, err := os.Stat(resolvConf); err != nil && !os.IsNotExist(err) {
+		return err
+	}
+	resolvConfExists := !os.IsNotExist(err)
+
+	if resolvConfExists && !owned {
+		// There's already a non-tailscale config in place, get rid of
+		// our backup.
+		os.Remove(backupConf)
+		return nil
+	}
+
+	// We own resolv.conf, and a backup exists.
+	if err := os.Rename(backupConf, resolvConf); err != nil {
 		return err
 	}
 
-	buf := new(bytes.Buffer)
-	writeResolvConf(buf, config.Nameservers, config.SearchDomains)
-	if err := atomicfile.WriteFile(resolvConf, buf.Bytes(), 0644); err != nil {
-		return err
+	return nil
+}
+
+func (m directManager) SetDNS(config OSConfig) error {
+	if config.IsZero() {
+		if err := m.restoreBackup(); err != nil {
+			return err
+		}
+	} else {
+		if err := m.backupConfig(); err != nil {
+			return err
+		}
+
+		buf := new(bytes.Buffer)
+		writeResolvConf(buf, config.Nameservers, config.SearchDomains)
+		if err := atomicfile.WriteFile(resolvConf, buf.Bytes(), 0644); err != nil {
+			return err
+		}
 	}
 
 	// We might have taken over a configuration managed by resolved,

net/dns/manager.go

@@ -5,7 +5,7 @@
 package dns
 
 import (
-	"strings"
+	"runtime"
 	"time"
 
 	"inet.af/netaddr"
@@ -51,23 +51,6 @@ func NewManager(logf logger.Logf, oscfg OSConfigurator, linkMon *monitor.Mon) *M
 	return m
 }
 
-// forceSplitDNSForTesting alters cfg to be a split DNS configuration
-// that only captures search paths. It's intended for testing split
-// DNS until the functionality is linked up in the admin panel.
-func forceSplitDNSForTesting(cfg *Config) {
-	if len(cfg.DefaultResolvers) == 0 {
-		return
-	}
-
-	if cfg.Routes == nil {
-		cfg.Routes = map[dnsname.FQDN][]netaddr.IPPort{}
-	}
-	for _, search := range cfg.SearchDomains {
-		cfg.Routes[search] = cfg.DefaultResolvers
-	}
-	cfg.DefaultResolvers = nil
-}
-
 func (m *Manager) Set(cfg Config) error {
 	m.logf("Set: %+v", cfg)
 
@@ -91,40 +74,40 @@ func (m *Manager) Set(cfg Config) error {
 
 // compileConfig converts cfg into a quad-100 resolver configuration
 // and an OS-level configuration.
-func (m *Manager) compileConfig(cfg Config) (resolver.Config, OSConfig, error) {
+func (m *Manager) compileConfig(cfg Config) (rcfg resolver.Config, ocfg OSConfig, err error) {
+	// The internal resolver always gets MagicDNS hosts and
+	// authoritative suffixes, even if we don't propagate MagicDNS to
+	// the OS.
+	rcfg.Hosts = cfg.Hosts
+	routes := map[dnsname.FQDN][]netaddr.IPPort{} // assigned conditionally to rcfg.Routes below.
+	for suffix, resolvers := range cfg.Routes {
+		if len(resolvers) == 0 {
+			rcfg.LocalDomains = append(rcfg.LocalDomains, suffix)
+		} else {
+			routes[suffix] = resolvers
+		}
+	}
+	// Similarly, the OS always gets search paths.
+	ocfg.SearchDomains = cfg.SearchDomains
+
 	// Deal with trivial configs first.
 	switch {
 	case !cfg.needsOSResolver():
 		// Set search domains, but nothing else. This also covers the
 		// case where cfg is entirely zero, in which case these
 		// configs clear all Tailscale DNS settings.
-		return resolver.Config{}, OSConfig{
-			SearchDomains: cfg.SearchDomains,
-		}, nil
+		return rcfg, ocfg, nil
 	case cfg.hasDefaultResolversOnly():
 		// Trivial CorpDNS configuration, just override the OS
 		// resolver.
-		return resolver.Config{}, OSConfig{
-			Nameservers:   toIPsOnly(cfg.DefaultResolvers),
-			SearchDomains: cfg.SearchDomains,
-		}, nil
+		ocfg.Nameservers = toIPsOnly(cfg.DefaultResolvers)
+		return rcfg, ocfg, nil
 	case cfg.hasDefaultResolvers():
 		// Default resolvers plus other stuff always ends up proxying
 		// through quad-100.
-		rcfg := resolver.Config{
-			Routes: map[dnsname.FQDN][]netaddr.IPPort{
-				".": cfg.DefaultResolvers,
-			},
-			Hosts:        cfg.Hosts,
-			LocalDomains: cfg.AuthoritativeSuffixes,
-		}
-		for suffix, resolvers := range cfg.Routes {
-			rcfg.Routes[suffix] = resolvers
-		}
-		ocfg := OSConfig{
-			Nameservers:   []netaddr.IP{tsaddr.TailscaleServiceIP()},
-			SearchDomains: cfg.SearchDomains,
-		}
+		rcfg.Routes = routes
+		rcfg.Routes["."] = cfg.DefaultResolvers
+		ocfg.Nameservers = []netaddr.IP{tsaddr.TailscaleServiceIP()}
 		return rcfg, ocfg, nil
 	}
 
@@ -132,64 +115,46 @@ func (m *Manager) compileConfig(cfg Config) (resolver.Config, OSConfig, error) {
 	// configurations. The possible cases don't return directly any
 	// more, because as a final step we have to handle the case where
 	// the OS can't do split DNS.
-	var rcfg resolver.Config
-	var ocfg OSConfig
 
-	if !cfg.hasHosts() && cfg.singleResolverSet() != nil && m.os.SupportsSplitDNS() {
+	// Workaround for
+	// https://github.com/tailscale/corp/issues/1662. Even though
+	// Windows natively supports split DNS, it only configures linux
+	// containers using whatever the primary is, and doesn't apply
+	// NRPT rules to DNS traffic coming from WSL.
+	//
+	// In order to make WSL work okay when the host Windows is using
+	// Tailscale, we need to set up quad-100 as a "full proxy"
+	// resolver, regardless of whether Windows itself can do split
+	// DNS. We still make Windows do split DNS itself when it can, but
+	// quad-100 will still have the full split configuration as well,
+	// and so can service WSL requests correctly.
+	//
+	// This bool is used in a couple of places below to implement this
+	// workaround.
+	isWindows := runtime.GOOS == "windows"
+	if cfg.singleResolverSet() != nil && m.os.SupportsSplitDNS() && !isWindows {
 		// Split DNS configuration requested, where all split domains
 		// go to the same resolvers. We can let the OS do it.
-		return resolver.Config{}, OSConfig{
-			Nameservers:   toIPsOnly(cfg.singleResolverSet()),
-			SearchDomains: cfg.SearchDomains,
-			MatchDomains:  cfg.matchDomains(),
-		}, nil
+		ocfg.Nameservers = toIPsOnly(cfg.singleResolverSet())
+		ocfg.MatchDomains = cfg.matchDomains()
+		return rcfg, ocfg, nil
 	}
 
 	// Split DNS configuration with either multiple upstream routes,
 	// or routes + MagicDNS, or just MagicDNS, or on an OS that cannot
 	// split-DNS. Install a split config pointing at quad-100.
-	rcfg = resolver.Config{
-		Hosts:        cfg.Hosts,
-		LocalDomains: cfg.AuthoritativeSuffixes,
-		Routes:       map[dnsname.FQDN][]netaddr.IPPort{},
-	}
-	for suffix, resolvers := range cfg.Routes {
-		rcfg.Routes[suffix] = resolvers
-	}
-	ocfg = OSConfig{
-		Nameservers:   []netaddr.IP{tsaddr.TailscaleServiceIP()},
-		SearchDomains: cfg.SearchDomains,
-	}
+	rcfg.Routes = routes
+	ocfg.Nameservers = []netaddr.IP{tsaddr.TailscaleServiceIP()}
 
 	// If the OS can't do native split-dns, read out the underlying
 	// resolver config and blend it into our config.
 	if m.os.SupportsSplitDNS() {
 		ocfg.MatchDomains = cfg.matchDomains()
-	} else {
+	}
+	if !m.os.SupportsSplitDNS() || isWindows {
 		bcfg, err := m.os.GetBaseConfig()
 		if err != nil {
-			// Temporary hack to make OSes where split-DNS isn't fully
-			// implemented yet not completely crap out, but instead
-			// fall back to quad-9 as a hardcoded "backup resolver".
-			//
-			// This codepath currently only triggers when opted into
-			// the split-DNS feature server side, and when at least
-			// one search domain is something within tailscale.com, so
-			// we don't accidentally leak unstable user DNS queries to
-			// quad-9 if we accidentally go down this codepath.
-			canUseHack := false
-			for _, dom := range cfg.SearchDomains {
-				if strings.HasSuffix(dom.WithoutTrailingDot(), ".tailscale.com") {
-					canUseHack = true
-					break
-				}
-			}
-			if !canUseHack {
-				return resolver.Config{}, OSConfig{}, err
-			}
-			bcfg = OSConfig{
-				Nameservers: []netaddr.IP{netaddr.IPv4(9, 9, 9, 9)},
-			}
+			return resolver.Config{}, OSConfig{}, err
 		}
 		rcfg.Routes["."] = toIPPorts(bcfg.Nameservers)
 		ocfg.SearchDomains = append(ocfg.SearchDomains, bcfg.SearchDomains...)
@@ -206,7 +171,7 @@ func (m *Manager) compileConfig(cfg Config) (resolver.Config, OSConfig, error) {
 func toIPsOnly(ipps []netaddr.IPPort) (ret []netaddr.IP) {
 	ret = make([]netaddr.IP, 0, len(ipps))
 	for _, ipp := range ipps {
-		ret = append(ret, ipp.IP)
+		ret = append(ret, ipp.IP())
 	}
 	return ret
 }
@@ -214,7 +179,7 @@ func toIPsOnly(ipps []netaddr.IPPort) (ret []netaddr.IP) {
 func toIPPorts(ips []netaddr.IP) (ret []netaddr.IPPort) {
 	ret = make([]netaddr.IPPort, 0, len(ips))
 	for _, ip := range ips {
-		ret = append(ret, netaddr.IPPort{IP: ip, Port: 53})
+		ret = append(ret, netaddr.IPPortFrom(ip, 53))
 	}
 	return ret
 }

net/dns/manager_linux.go

@@ -16,6 +16,7 @@ import (
 
 	"github.com/godbus/dbus/v5"
 	"tailscale.com/types/logger"
+	"tailscale.com/util/cmpver"
 )
 
 type kv struct {
@@ -56,26 +57,62 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (ret OSConfigurat
 		}
 		if err := dbusPing("org.freedesktop.NetworkManager", "/org/freedesktop/NetworkManager/DnsManager"); err != nil {
 			dbg("nm", "no")
-			return newResolvedManager(logf)
+			return newResolvedManager(logf, interfaceName)
 		}
 		dbg("nm", "yes")
 		if err := nmIsUsingResolved(); err != nil {
 			dbg("nm-resolved", "no")
-			return newResolvedManager(logf)
+			return newResolvedManager(logf, interfaceName)
 		}
 		dbg("nm-resolved", "yes")
-		return newNMManager(interfaceName)
+
+		// Version of NetworkManager before 1.26.6 programmed resolved
+		// incorrectly, such that NM's settings would always take
+		// precedence over other settings set by other resolved
+		// clients.
+		//
+		// If we're dealing with such a version, we have to set our
+		// DNS settings through NM to have them take.
+		//
+		// However, versions 1.26.6 later both fixed the resolved
+		// programming issue _and_ started ignoring DNS settings for
+		// "unmanaged" interfaces - meaning NM 1.26.6 and later
+		// actively ignore DNS configuration we give it. So, for those
+		// NM versions, we can and must use resolved directly.
+		old, err := nmVersionOlderThan("1.26.6")
+		if err != nil {
+			// Failed to figure out NM's version, can't make a correct
+			// decision.
+			return nil, fmt.Errorf("checking NetworkManager version: %v", err)
+		}
+		if old {
+			dbg("nm-old", "yes")
+			return newNMManager(interfaceName)
+		}
+		dbg("nm-old", "no")
+		return newResolvedManager(logf, interfaceName)
 	case "resolvconf":
 		dbg("rc", "resolvconf")
 		if err := resolvconfSourceIsNM(bs); err == nil {
 			dbg("src-is-nm", "yes")
 			if err := dbusPing("org.freedesktop.NetworkManager", "/org/freedesktop/NetworkManager/DnsManager"); err == nil {
 				dbg("nm", "yes")
-				return newNMManager(interfaceName)
+				old, err := nmVersionOlderThan("1.26.6")
+				if err != nil {
+					return nil, fmt.Errorf("checking NetworkManager version: %v", err)
+				}
+				if old {
+					dbg("nm-old", "yes")
+					return newNMManager(interfaceName)
+				} else {
+					dbg("nm-old", "no")
+				}
+			} else {
+				dbg("nm", "no")
 			}
-			dbg("nm", "no")
+		} else {
+			dbg("src-is-nm", "no")
 		}
-		dbg("src-is-nm", "no")
 		if _, err := exec.LookPath("resolvconf"); err != nil {
 			dbg("resolvconf", "no")
 			return newDirectManager()
@@ -89,7 +126,16 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (ret OSConfigurat
 			return newDirectManager()
 		}
 		dbg("nm", "yes")
-		return newNMManager(interfaceName)
+		old, err := nmVersionOlderThan("1.26.6")
+		if err != nil {
+			return nil, fmt.Errorf("checking NetworkManager version: %v", err)
+		}
+		if old {
+			dbg("nm-old", "yes")
+			return newNMManager(interfaceName)
+		}
+		dbg("nm-old", "no")
+		return newDirectManager()
 	default:
 		dbg("rc", "unknown")
 		return newDirectManager()
@@ -135,6 +181,27 @@ func resolvconfSourceIsNM(resolvDotConf []byte) error {
 	return nil
 }
 
+func nmVersionOlderThan(want string) (bool, error) {
+	conn, err := dbus.SystemBus()
+	if err != nil {
+		// DBus probably not running.
+		return false, err
+	}
+
+	nm := conn.Object("org.freedesktop.NetworkManager", dbus.ObjectPath("/org/freedesktop/NetworkManager"))
+	v, err := nm.GetProperty("org.freedesktop.NetworkManager.Version")
+	if err != nil {
+		return false, err
+	}
+
+	version, ok := v.Value().(string)
+	if !ok {
+		return false, fmt.Errorf("unexpected type %T for NM version", v.Value())
+	}
+
+	return cmpver.Compare(version, want) < 0, nil
+}
+
 func nmIsUsingResolved() error {
 	conn, err := dbus.SystemBus()
 	if err != nil {

net/dns/manager_test.go

@@ -5,6 +5,7 @@
 package dns
 
 import (
+	"runtime"
 	"testing"
 
 	"github.com/google/go-cmp/cmp"
@@ -45,6 +46,10 @@ func (c *fakeOSConfigurator) GetBaseConfig() (OSConfig, error) {
 func (c *fakeOSConfigurator) Close() error { return nil }
 
 func TestManager(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skipf("test's assumptions break because of https://github.com/tailscale/corp/issues/1662")
+	}
+
 	// Note: these tests assume that it's safe to switch the
 	// OSConfigurator's split-dns support on and off between Set
 	// calls. Empirically this is currently true, because we reprobe
@@ -71,6 +76,20 @@ func TestManager(t *testing.T) {
 				SearchDomains: fqdns("tailscale.com", "universe.tf"),
 			},
 		},
+		{
+			// Regression test for https://github.com/tailscale/tailscale/issues/1886
+			name: "hosts-only",
+			in: Config{
+				Hosts: hosts(
+					"dave.ts.com.", "1.2.3.4",
+					"bradfitz.ts.com.", "2.3.4.5"),
+			},
+			rs: resolver.Config{
+				Hosts: hosts(
+					"dave.ts.com.", "1.2.3.4",
+					"bradfitz.ts.com.", "2.3.4.5"),
+			},
+		},
 		{
 			name: "corp",
 			in: Config{
@@ -99,10 +118,10 @@ func TestManager(t *testing.T) {
 			in: Config{
 				DefaultResolvers: mustIPPs("1.1.1.1:53", "9.9.9.9:53"),
 				SearchDomains:    fqdns("tailscale.com", "universe.tf"),
+				Routes:           upstreams("ts.com", ""),
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
 			},
 			os: OSConfig{
 				Nameservers:   mustIPs("100.100.100.100"),
@@ -121,10 +140,10 @@ func TestManager(t *testing.T) {
 			in: Config{
 				DefaultResolvers: mustIPPs("1.1.1.1:53", "9.9.9.9:53"),
 				SearchDomains:    fqdns("tailscale.com", "universe.tf"),
+				Routes:           upstreams("ts.com", ""),
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
 			},
 			split: true,
 			os: OSConfig{
@@ -256,8 +275,8 @@ func TestManager(t *testing.T) {
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
-				SearchDomains:         fqdns("tailscale.com", "universe.tf"),
+				Routes:        upstreams("ts.com", ""),
+				SearchDomains: fqdns("tailscale.com", "universe.tf"),
 			},
 			bs: OSConfig{
 				Nameservers:   mustIPs("8.8.8.8"),
@@ -281,8 +300,8 @@ func TestManager(t *testing.T) {
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
-				SearchDomains:         fqdns("tailscale.com", "universe.tf"),
+				Routes:        upstreams("ts.com", ""),
+				SearchDomains: fqdns("tailscale.com", "universe.tf"),
 			},
 			split: true,
 			os: OSConfig{
@@ -300,12 +319,11 @@ func TestManager(t *testing.T) {
 		{
 			name: "routes-magic",
 			in: Config{
-				Routes: upstreams("corp.com", "2.2.2.2:53"),
+				Routes: upstreams("corp.com", "2.2.2.2:53", "ts.com", ""),
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
-				SearchDomains:         fqdns("tailscale.com", "universe.tf"),
+				SearchDomains: fqdns("tailscale.com", "universe.tf"),
 			},
 			bs: OSConfig{
 				Nameservers:   mustIPs("8.8.8.8"),
@@ -328,12 +346,13 @@ func TestManager(t *testing.T) {
 		{
 			name: "routes-magic-split",
 			in: Config{
-				Routes: upstreams("corp.com", "2.2.2.2:53"),
+				Routes: upstreams(
+					"corp.com", "2.2.2.2:53",
+					"ts.com", ""),
 				Hosts: hosts(
 					"dave.ts.com.", "1.2.3.4",
 					"bradfitz.ts.com.", "2.3.4.5"),
-				AuthoritativeSuffixes: fqdns("ts.com"),
-				SearchDomains:         fqdns("tailscale.com", "universe.tf"),
+				SearchDomains: fqdns("tailscale.com", "universe.tf"),
 			},
 			split: true,
 			os: OSConfig{
@@ -363,11 +382,12 @@ func TestManager(t *testing.T) {
 			if err := m.Set(test.in); err != nil {
 				t.Fatalf("m.Set: %v", err)
 			}
-			tr := cmp.Transformer("ipStr", func(ip netaddr.IP) string { return ip.String() })
-			if diff := cmp.Diff(f.OSConfig, test.os, tr, cmpopts.EquateEmpty()); diff != "" {
+			trIP := cmp.Transformer("ipStr", func(ip netaddr.IP) string { return ip.String() })
+			trIPPort := cmp.Transformer("ippStr", func(ipp netaddr.IPPort) string { return ipp.String() })
+			if diff := cmp.Diff(f.OSConfig, test.os, trIP, trIPPort, cmpopts.EquateEmpty()); diff != "" {
 				t.Errorf("wrong OSConfig (-got+want)\n%s", diff)
 			}
-			if diff := cmp.Diff(f.ResolverConfig, test.rs, tr, cmpopts.EquateEmpty()); diff != "" {
+			if diff := cmp.Diff(f.ResolverConfig, test.rs, trIP, trIPPort, cmpopts.EquateEmpty()); diff != "" {
 				t.Errorf("wrong resolver.Config (-got+want)\n%s", diff)
 			}
 		})
@@ -423,7 +443,12 @@ func upstreams(strs ...string) (ret map[dnsname.FQDN][]netaddr.IPPort) {
 	var key dnsname.FQDN
 	ret = map[dnsname.FQDN][]netaddr.IPPort{}
 	for _, s := range strs {
-		if ipp, err := netaddr.ParseIPPort(s); err == nil {
+		if s == "" {
+			if key == "" {
+				panic("IPPort provided before suffix")
+			}
+			ret[key] = nil
+		} else if ipp, err := netaddr.ParseIPPort(s); err == nil {
 			if key == "" {
 				panic("IPPort provided before suffix")
 			}

net/dns/manager_windows.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"fmt"
 	"os/exec"
+	"sort"
 	"strings"
 	"syscall"
 	"time"
@@ -343,10 +344,11 @@ func (m windowsManager) getBasePrimaryResolver() (resolvers []netaddr.IP, err er
 		return nil, err
 	}
 
-	var (
-		primary winipcfg.LUID
-		best    = ^uint32(0)
-	)
+	type candidate struct {
+		id     winipcfg.LUID
+		metric uint32
+	}
+	var candidates []candidate
 	for _, row := range ifrows {
 		if !row.Connected {
 			continue
@@ -354,29 +356,57 @@ func (m windowsManager) getBasePrimaryResolver() (resolvers []netaddr.IP, err er
 		if row.InterfaceLUID == tsLUID {
 			continue
 		}
-		if row.Metric < best {
-			primary = row.InterfaceLUID
-			best = row.Metric
-		}
+		candidates = append(candidates, candidate{row.InterfaceLUID, row.Metric})
 	}
-	if primary == 0 {
+	if len(candidates) == 0 {
 		// No resolvers set outside of Tailscale.
 		return nil, nil
 	}
 
-	ips, err := primary.DNS()
-	if err != nil {
-		return nil, err
-	}
-	for _, stdip := range ips {
-		if ip, ok := netaddr.FromStdIP(stdip); ok {
+	sort.Slice(candidates, func(i, j int) bool { return candidates[i].metric < candidates[j].metric })
+
+	for _, candidate := range candidates {
+		ips, err := candidate.id.DNS()
+		if err != nil {
+			return nil, err
+		}
+
+	ipLoop:
+		for _, stdip := range ips {
+			ip, ok := netaddr.FromStdIP(stdip)
+			if !ok {
+				continue
+			}
+			// Skip IPv6 site-local resolvers. These are an ancient
+			// and obsolete IPv6 RFC, which Windows still faithfully
+			// implements. The net result is that some low-metric
+			// interfaces can "have" DNS resolvers, but they're just
+			// site-local resolver IPs that don't go anywhere. So, we
+			// skip the site-local resolvers in order to find the
+			// first interface that has real DNS servers configured.
+			for _, sl := range siteLocalResolvers {
+				if ip.WithZone("") == sl {
+					continue ipLoop
+				}
+			}
 			resolvers = append(resolvers, ip)
 		}
+
+		if len(resolvers) > 0 {
+			// Found some resolvers, we're done.
+			break
+		}
 	}
 
 	return resolvers, nil
 }
 
+var siteLocalResolvers = []netaddr.IP{
+	netaddr.MustParseIP("fec0:0:0:ffff::1"),
+	netaddr.MustParseIP("fec0:0:0:ffff::2"),
+	netaddr.MustParseIP("fec0:0:0:ffff::3"),
+}
+
 func isWindows7() bool {
 	key, err := registry.OpenKey(registry.LOCAL_MACHINE, versionKey, registry.READ)
 	if err != nil {

net/dns/nm.go

@@ -16,6 +16,7 @@ import (
 
 	"github.com/godbus/dbus/v5"
 	"inet.af/netaddr"
+	"tailscale.com/net/interfaces"
 	"tailscale.com/util/dnsname"
 	"tailscale.com/util/endian"
 )
@@ -137,13 +138,53 @@ func (m *nmManager) trySet(ctx context.Context, config OSConfig) error {
 		}
 	}
 
-	general := settings["connection"]
-	general["llmnr"] = dbus.MakeVariant(0)
-	general["mdns"] = dbus.MakeVariant(0)
+	// NetworkManager wipes out IPv6 address configuration unless we
+	// tell it explicitly to keep it. Read out the current interface
+	// settings and mirror them out to NetworkManager.
+	var addrs6 []map[string]interface{}
+	addrs, _, err := interfaces.Tailscale()
+	if err == nil {
+		for _, a := range addrs {
+			if a.Is6() {
+				addrs6 = append(addrs6, map[string]interface{}{
+					"address": a.String(),
+					"prefix":  uint32(128),
+				})
+			}
+		}
+	}
+
+	seen := map[dnsname.FQDN]bool{}
+	var search []string
+	for _, dom := range config.SearchDomains {
+		if seen[dom] {
+			continue
+		}
+		seen[dom] = true
+		search = append(search, dom.WithTrailingDot())
+	}
+	for _, dom := range config.MatchDomains {
+		if seen[dom] {
+			continue
+		}
+		seen[dom] = true
+		search = append(search, "~"+dom.WithTrailingDot())
+	}
+	if len(config.MatchDomains) == 0 {
+		// Non-split routing requested, add an all-domains match.
+		search = append(search, "~.")
+	}
+
+	// Ideally we would like to disable LLMNR and mdns on the
+	// interface here, but older NetworkManagers don't understand
+	// those settings and choke on them, so we don't. Both LLMNR and
+	// mdns will fail since tailscale0 doesn't do multicast, so it's
+	// effectively fine. We used to try and enforce LLMNR and mdns
+	// settings here, but that led to #1870.
 
 	ipv4Map := settings["ipv4"]
 	ipv4Map["dns"] = dbus.MakeVariant(dnsv4)
-	ipv4Map["dns-search"] = dbus.MakeVariant(config.SearchDomains)
+	ipv4Map["dns-search"] = dbus.MakeVariant(search)
 	// We should only request priority if we have nameservers to set.
 	if len(dnsv4) == 0 {
 		ipv4Map["dns-priority"] = dbus.MakeVariant(lowerPriority)
@@ -174,12 +215,15 @@ func (m *nmManager) trySet(ctx context.Context, config OSConfig) error {
 	// (none of its business anyway, we handle our own default
 	// routing).
 	ipv6Map["method"] = dbus.MakeVariant("auto")
+	if len(addrs6) > 0 {
+		ipv6Map["address-data"] = dbus.MakeVariant(addrs6)
+	}
 	ipv6Map["ignore-auto-routes"] = dbus.MakeVariant(true)
 	ipv6Map["ignore-auto-dns"] = dbus.MakeVariant(true)
 	ipv6Map["never-default"] = dbus.MakeVariant(true)
 
 	ipv6Map["dns"] = dbus.MakeVariant(dnsv6)
-	ipv6Map["dns-search"] = dbus.MakeVariant(config.SearchDomains)
+	ipv6Map["dns-search"] = dbus.MakeVariant(search)
 	if len(dnsv6) == 0 {
 		ipv6Map["dns-priority"] = dbus.MakeVariant(lowerPriority)
 	} else if len(config.MatchDomains) > 0 {
@@ -206,7 +250,7 @@ func (m *nmManager) trySet(ctx context.Context, config OSConfig) error {
 	}
 
 	if call := device.CallWithContext(ctx, "org.freedesktop.NetworkManager.Device.Reapply", 0, settings, version, uint32(0)); call.Err != nil {
-		return fmt.Errorf("reapply: %w", err)
+		return fmt.Errorf("reapply: %w", call.Err)
 	}
 
 	return nil

net/dns/openresolv.go

@@ -2,6 +2,8 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// +build linux freebsd openbsd
+
 package dns
 
 import (
@@ -19,7 +21,20 @@ func newOpenresolvManager() (openresolvManager, error) {
 	return openresolvManager{}, nil
 }
 
+func (m openresolvManager) deleteTailscaleConfig() error {
+	cmd := exec.Command("resolvconf", "-f", "-d", "tailscale")
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		return fmt.Errorf("running %s: %s", cmd, out)
+	}
+	return nil
+}
+
 func (m openresolvManager) SetDNS(config OSConfig) error {
+	if config.IsZero() {
+		return m.deleteTailscaleConfig()
+	}
+
 	var stdin bytes.Buffer
 	writeResolvConf(&stdin, config.Nameservers, config.SearchDomains)
 
@@ -75,10 +90,5 @@ func (m openresolvManager) GetBaseConfig() (OSConfig, error) {
 }
 
 func (m openresolvManager) Close() error {
-	cmd := exec.Command("resolvconf", "-f", "-d", "tailscale")
-	out, err := cmd.CombinedOutput()
-	if err != nil {
-		return fmt.Errorf("running %s: %s", cmd, out)
-	}
-	return nil
+	return m.deleteTailscaleConfig()
 }

net/dns/osconfig.go

@@ -52,6 +52,10 @@ type OSConfig struct {
 	MatchDomains []dnsname.FQDN
 }
 
+func (o OSConfig) IsZero() bool {
+	return len(o.Nameservers) == 0 && len(o.SearchDomains) == 0 && len(o.MatchDomains) == 0
+}
+
 func (a OSConfig) Equal(b OSConfig) bool {
 	if len(a.Nameservers) != len(b.Nameservers) {
 		return false

net/dns/resolvconf.go

@@ -2,6 +2,8 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// +build linux freebsd openbsd
+
 package dns
 
 import (

net/dns/resolved.go

@@ -12,11 +12,11 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"net"
 
 	"github.com/godbus/dbus/v5"
 	"golang.org/x/sys/unix"
 	"inet.af/netaddr"
-	"tailscale.com/net/interfaces"
 	"tailscale.com/types/logger"
 	"tailscale.com/util/dnsname"
 )
@@ -85,17 +85,24 @@ func isResolvedActive() bool {
 // resolvedManager uses the systemd-resolved DBus API.
 type resolvedManager struct {
 	logf     logger.Logf
+	ifidx    int
 	resolved dbus.BusObject
 }
 
-func newResolvedManager(logf logger.Logf) (*resolvedManager, error) {
+func newResolvedManager(logf logger.Logf, interfaceName string) (*resolvedManager, error) {
 	conn, err := dbus.SystemBus()
 	if err != nil {
 		return nil, err
 	}
 
+	iface, err := net.InterfaceByName(interfaceName)
+	if err != nil {
+		return nil, err
+	}
+
 	return &resolvedManager{
 		logf:     logf,
+		ifidx:    iface.Index,
 		resolved: conn.Object("org.freedesktop.resolve1", dbus.ObjectPath("/org/freedesktop/resolve1")),
 	}, nil
 }
@@ -105,16 +112,6 @@ func (m *resolvedManager) SetDNS(config OSConfig) error {
 	ctx, cancel := context.WithTimeout(context.Background(), reconfigTimeout)
 	defer cancel()
 
-	// In principle, we could persist this in the manager struct
-	// if we knew that interface indices are persistent. This does not seem to be the case.
-	_, iface, err := interfaces.Tailscale()
-	if err != nil {
-		return fmt.Errorf("getting interface index: %w", err)
-	}
-	if iface == nil {
-		return errNotReady
-	}
-
 	var linkNameservers = make([]resolvedLinkNameserver, len(config.Nameservers))
 	for i, server := range config.Nameservers {
 		ip := server.As16()
@@ -131,9 +128,9 @@ func (m *resolvedManager) SetDNS(config OSConfig) error {
 		}
 	}
 
-	err = m.resolved.CallWithContext(
+	err := m.resolved.CallWithContext(
 		ctx, "org.freedesktop.resolve1.Manager.SetLinkDNS", 0,
-		iface.Index, linkNameservers,
+		m.ifidx, linkNameservers,
 	).Store()
 	if err != nil {
 		return fmt.Errorf("setLinkDNS: %w", err)
@@ -163,7 +160,7 @@ func (m *resolvedManager) SetDNS(config OSConfig) error {
 			RoutingOnly: true,
 		})
 	}
-	if len(config.MatchDomains) == 0 {
+	if len(config.MatchDomains) == 0 && len(config.Nameservers) > 0 {
 		// Caller requested full DNS interception, install a
 		// routing-only root domain.
 		linkDomains = append(linkDomains, resolvedLinkDomain{
@@ -174,13 +171,13 @@ func (m *resolvedManager) SetDNS(config OSConfig) error {
 
 	err = m.resolved.CallWithContext(
 		ctx, "org.freedesktop.resolve1.Manager.SetLinkDomains", 0,
-		iface.Index, linkDomains,
+		m.ifidx, linkDomains,
 	).Store()
 	if err != nil {
 		return fmt.Errorf("setLinkDomains: %w", err)
 	}
 
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDefaultRoute", 0, iface.Index, len(config.MatchDomains) == 0); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDefaultRoute", 0, m.ifidx, len(config.MatchDomains) == 0); call.Err != nil {
 		return fmt.Errorf("setLinkDefaultRoute: %w", err)
 	}
 
@@ -189,22 +186,22 @@ func (m *resolvedManager) SetDNS(config OSConfig) error {
 	// or something).
 
 	// Disable LLMNR, we don't do multicast.
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkLLMNR", 0, iface.Index, "no"); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkLLMNR", 0, m.ifidx, "no"); call.Err != nil {
 		m.logf("[v1] failed to disable LLMNR: %v", call.Err)
 	}
 
 	// Disable mdns.
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkMulticastDNS", 0, iface.Index, "no"); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkMulticastDNS", 0, m.ifidx, "no"); call.Err != nil {
 		m.logf("[v1] failed to disable mdns: %v", call.Err)
 	}
 
 	// We don't support dnssec consistently right now, force it off to
 	// avoid partial failures when we split DNS internally.
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDNSSEC", 0, iface.Index, "no"); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDNSSEC", 0, m.ifidx, "no"); call.Err != nil {
 		m.logf("[v1] failed to disable DNSSEC: %v", call.Err)
 	}
 
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDNSOverTLS", 0, iface.Index, "no"); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.SetLinkDNSOverTLS", 0, m.ifidx, "no"); call.Err != nil {
 		m.logf("[v1] failed to disable DoT: %v", call.Err)
 	}
 
@@ -227,15 +224,7 @@ func (m *resolvedManager) Close() error {
 	ctx, cancel := context.WithTimeout(context.Background(), reconfigTimeout)
 	defer cancel()
 
-	_, iface, err := interfaces.Tailscale()
-	if err != nil {
-		return fmt.Errorf("getting interface index: %w", err)
-	}
-	if iface == nil {
-		return errNotReady
-	}
-
-	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.RevertLink", 0, iface.Index); call.Err != nil {
+	if call := m.resolved.CallWithContext(ctx, "org.freedesktop.resolve1.Manager.RevertLink", 0, m.ifidx); call.Err != nil {
 		return fmt.Errorf("RevertLink: %w", call.Err)
 	}
 

net/dns/resolver/forwarder.go

@@ -13,7 +13,6 @@ import (
 	"hash/crc32"
 	"math/rand"
 	"net"
-	"os"
 	"sync"
 	"time"
 
@@ -39,8 +38,6 @@ const (
 
 var errNoUpstreams = errors.New("upstream nameservers not set")
 
-var aLongTimeAgo = time.Unix(0, 1)
-
 type forwardingRecord struct {
 	src       netaddr.IPPort
 	createdAt time.Time
@@ -303,8 +300,6 @@ type fwdConn struct {
 	// logf allows a fwdConn to log.
 	logf logger.Logf
 
-	// wg tracks the number of outstanding conn.Read and conn.Write calls.
-	wg sync.WaitGroup
 	// change allows calls to read to block until a the network connection has been replaced.
 	change *sync.Cond
 
@@ -352,15 +347,12 @@ func (c *fwdConn) send(packet []byte, dst netaddr.IPPort) {
 		}
 		c.mu.Unlock()
 
-		a := dst.UDPAddr()
-		c.wg.Add(1)
-		_, err := conn.WriteTo(packet, a)
-		c.wg.Done()
+		_, err := conn.WriteTo(packet, dst.UDPAddr())
 		if err == nil {
 			// Success
 			return
 		}
-		if errors.Is(err, os.ErrDeadlineExceeded) {
+		if errors.Is(err, net.ErrClosed) {
 			// We intentionally closed this connection.
 			// It has been replaced by a new connection. Try again.
 			continue
@@ -429,14 +421,12 @@ func (c *fwdConn) read(out []byte) int {
 		}
 		c.mu.Unlock()
 
-		c.wg.Add(1)
 		n, _, err := conn.ReadFrom(out)
-		c.wg.Done()
 		if err == nil {
 			// Success.
 			return n
 		}
-		if errors.Is(err, os.ErrDeadlineExceeded) {
+		if errors.Is(err, net.ErrClosed) {
 			// We intentionally closed this connection.
 			// It has been replaced by a new connection. Try again.
 			continue
@@ -468,10 +458,7 @@ func (c *fwdConn) closeConnLocked() {
 	if c.conn == nil {
 		return
 	}
-	// Unblock all readers/writers, wait for them, close the connection.
-	c.conn.SetDeadline(aLongTimeAgo)
-	c.wg.Wait()
-	c.conn.Close()
+	c.conn.Close() // unblocks all readers/writers, they'll pick up the next connection.
 	c.conn = nil
 }
 

net/dns/resolver/tsdns_test.go

@@ -433,8 +433,8 @@ func TestDelegateCollision(t *testing.T) {
 		qtype dns.Type
 		addr  netaddr.IPPort
 	}{
-		{"test.site.", dns.TypeA, netaddr.IPPort{IP: netaddr.IPv4(1, 1, 1, 1), Port: 1001}},
-		{"test.site.", dns.TypeAAAA, netaddr.IPPort{IP: netaddr.IPv4(1, 1, 1, 1), Port: 1002}},
+		{"test.site.", dns.TypeA, netaddr.IPPortFrom(netaddr.IPv4(1, 1, 1, 1), 1001)},
+		{"test.site.", dns.TypeAAAA, netaddr.IPPortFrom(netaddr.IPv4(1, 1, 1, 1), 1002)},
 	}
 
 	// packets will have the same dns txid.

net/interfaces/interfaces.go

@@ -26,7 +26,7 @@ var LoginEndpointForProxyDetermination = "https://login.tailscale.com/"
 // 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) {
+func Tailscale() ([]netaddr.IP, *net.Interface, error) {
 	ifs, err := net.Interfaces()
 	if err != nil {
 		return nil, nil, err
@@ -39,14 +39,18 @@ func Tailscale() (net.IP, *net.Interface, error) {
 		if err != nil {
 			continue
 		}
+		var tsIPs []netaddr.IP
 		for _, a := range addrs {
 			if ipnet, ok := a.(*net.IPNet); ok {
 				nip, ok := netaddr.FromStdIP(ipnet.IP)
 				if ok && tsaddr.IsTailscaleIP(nip) {
-					return ipnet.IP, &iface, nil
+					tsIPs = append(tsIPs, nip)
 				}
 			}
 		}
+		if len(tsIPs) > 0 {
+			return tsIPs, &iface, nil
+		}
 	}
 	return nil, nil, nil
 }
@@ -191,7 +195,7 @@ func ForeachInterface(fn func(Interface, []netaddr.IPPrefix)) error {
 			}
 		}
 		sort.Slice(pfxs, func(i, j int) bool {
-			return pfxs[i].IP.Less(pfxs[j].IP)
+			return pfxs[i].IP().Less(pfxs[j].IP())
 		})
 		fn(Interface{iface}, pfxs)
 	}
@@ -260,7 +264,7 @@ func (s *State) String() string {
 			fmt.Fprintf(&sb, "%s:[", ifName)
 			needSpace := false
 			for _, pfx := range s.InterfaceIPs[ifName] {
-				if !isInterestingIP(pfx.IP) {
+				if !isInterestingIP(pfx.IP()) {
 					continue
 				}
 				if needSpace {
@@ -363,7 +367,7 @@ func (s *State) AnyInterfaceUp() bool {
 
 func hasTailscaleIP(pfxs []netaddr.IPPrefix) bool {
 	for _, pfx := range pfxs {
-		if tsaddr.IsTailscaleIP(pfx.IP) {
+		if tsaddr.IsTailscaleIP(pfx.IP()) {
 			return true
 		}
 	}
@@ -403,11 +407,11 @@ func GetState() (*State, error) {
 			return
 		}
 		for _, pfx := range pfxs {
-			if pfx.IP.IsLoopback() || pfx.IP.IsLinkLocalUnicast() {
+			if pfx.IP().IsLoopback() || pfx.IP().IsLinkLocalUnicast() {
 				continue
 			}
-			s.HaveV6Global = s.HaveV6Global || isGlobalV6(pfx.IP)
-			s.HaveV4 = s.HaveV4 || pfx.IP.Is4()
+			s.HaveV6Global = s.HaveV6Global || isGlobalV6(pfx.IP())
+			s.HaveV4 = s.HaveV4 || pfx.IP().Is4()
 		}
 	}); err != nil {
 		return nil, err
@@ -443,7 +447,7 @@ func HTTPOfListener(ln net.Listener) string {
 	var goodIP string
 	var privateIP string
 	ForeachInterfaceAddress(func(i Interface, pfx netaddr.IPPrefix) {
-		ip := pfx.IP
+		ip := pfx.IP()
 		if isPrivateIP(ip) {
 			if privateIP == "" {
 				privateIP = ip.String()
@@ -480,7 +484,7 @@ func LikelyHomeRouterIP() (gateway, myIP netaddr.IP, ok bool) {
 		return
 	}
 	ForeachInterfaceAddress(func(i Interface, pfx netaddr.IPPrefix) {
-		ip := pfx.IP
+		ip := pfx.IP()
 		if !i.IsUp() || ip.IsZero() || !myIP.IsZero() {
 			return
 		}
@@ -524,7 +528,7 @@ var (
 // isInterestingIP.
 func anyInterestingIP(pfxs []netaddr.IPPrefix) bool {
 	for _, pfx := range pfxs {
-		if isInterestingIP(pfx.IP) {
+		if isInterestingIP(pfx.IP()) {
 			return true
 		}
 	}

net/interfaces/interfaces_darwin.go

@@ -28,6 +28,11 @@ func DefaultRouteInterface() (string, error) {
 	return iface.Name, nil
 }
 
+// fetchRoutingTable calls route.FetchRIB, fetching NET_RT_DUMP2.
+func fetchRoutingTable() (rib []byte, err error) {
+	return route.FetchRIB(syscall.AF_UNSPEC, syscall.NET_RT_DUMP2, 0)
+}
+
 func DefaultRouteInterfaceIndex() (int, error) {
 	// $ netstat -nr
 	// Routing tables
@@ -43,7 +48,7 @@ func DefaultRouteInterfaceIndex() (int, error) {
 	// c       RTF_PRCLONING    Protocol-specified generate new routes on use
 	// I       RTF_IFSCOPE      Route is associated with an interface scope
 
-	rib, err := route.FetchRIB(syscall.AF_UNSPEC, syscall.NET_RT_DUMP2, 0)
+	rib, err := fetchRoutingTable()
 	if err != nil {
 		return 0, fmt.Errorf("route.FetchRIB: %w", err)
 	}
@@ -83,7 +88,7 @@ func init() {
 }
 
 func likelyHomeRouterIPDarwinFetchRIB() (ret netaddr.IP, ok bool) {
-	rib, err := route.FetchRIB(syscall.AF_UNSPEC, syscall.NET_RT_DUMP2, 0)
+	rib, err := fetchRoutingTable()
 	if err != nil {
 		log.Printf("routerIP/FetchRIB: %v", err)
 		return ret, false

net/interfaces/interfaces_darwin_test.go

@@ -83,4 +83,14 @@ func likelyHomeRouterIPDarwinExec() (ret netaddr.IP, ok bool) {
 	return ret, !ret.IsZero()
 }
 
+func TestFetchRoutingTable(t *testing.T) {
+	// Issue 1345: this used to be flaky on darwin.
+	for i := 0; i < 20; i++ {
+		_, err := fetchRoutingTable()
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+}
+
 var errStopReadingNetstatTable = errors.New("found private gateway")

net/netcheck/netcheck.go

@@ -625,7 +625,7 @@ func (rs *reportState) stopTimers() {
 func (rs *reportState) addNodeLatency(node *tailcfg.DERPNode, ipp netaddr.IPPort, d time.Duration) {
 	var ipPortStr string
 	if ipp != (netaddr.IPPort{}) {
-		ipPortStr = net.JoinHostPort(ipp.IP.String(), fmt.Sprint(ipp.Port))
+		ipPortStr = net.JoinHostPort(ipp.IP().String(), fmt.Sprint(ipp.Port()))
 	}
 
 	rs.mu.Lock()
@@ -650,13 +650,13 @@ func (rs *reportState) addNodeLatency(node *tailcfg.DERPNode, ipp netaddr.IPPort
 	}
 
 	switch {
-	case ipp.IP.Is6():
+	case ipp.IP().Is6():
 		updateLatency(ret.RegionV6Latency, node.RegionID, d)
 		ret.IPv6 = true
 		ret.GlobalV6 = ipPortStr
 		// TODO: track MappingVariesByDestIP for IPv6
 		// too? Would be sad if so, but who knows.
-	case ipp.IP.Is4():
+	case ipp.IP().Is4():
 		updateLatency(ret.RegionV4Latency, node.RegionID, d)
 		ret.IPv4 = true
 		if rs.gotEP4 == "" {
@@ -1172,7 +1172,7 @@ func (c *Client) nodeAddr(ctx context.Context, n *tailcfg.DERPNode, proto probeP
 		if proto == probeIPv6 && ip.Is4() {
 			return nil
 		}
-		return netaddr.IPPort{IP: ip, Port: uint16(port)}.UDPAddr()
+		return netaddr.IPPortFrom(ip, uint16(port)).UDPAddr()
 	}
 
 	switch proto {
@@ -1182,7 +1182,7 @@ func (c *Client) nodeAddr(ctx context.Context, n *tailcfg.DERPNode, proto probeP
 			if !ip.Is4() {
 				return nil
 			}
-			return netaddr.IPPort{IP: ip, Port: uint16(port)}.UDPAddr()
+			return netaddr.IPPortFrom(ip, uint16(port)).UDPAddr()
 		}
 	case probeIPv6:
 		if n.IPv6 != "" {
@@ -1190,7 +1190,7 @@ func (c *Client) nodeAddr(ctx context.Context, n *tailcfg.DERPNode, proto probeP
 			if !ip.Is6() {
 				return nil
 			}
-			return netaddr.IPPort{IP: ip, Port: uint16(port)}.UDPAddr()
+			return netaddr.IPPortFrom(ip, uint16(port)).UDPAddr()
 		}
 	default:
 		return nil

net/netstat/netstat_windows.go

@@ -157,10 +157,9 @@ func ipport4(addr uint32, port uint16) netaddr.IPPort {
 	if !endian.Big {
 		addr = bits.ReverseBytes32(addr)
 	}
-	return netaddr.IPPort{
-		IP:   netaddr.IPv4(byte(addr>>24), byte(addr>>16), byte(addr>>8), byte(addr)),
-		Port: port,
-	}
+	return netaddr.IPPortFrom(
+		netaddr.IPv4(byte(addr>>24), byte(addr>>16), byte(addr>>8), byte(addr)),
+		port)
 }
 
 func ipport6(addr [16]byte, scope uint32, port uint16) netaddr.IPPort {
@@ -169,10 +168,7 @@ func ipport6(addr [16]byte, scope uint32, port uint16) netaddr.IPPort {
 		// TODO: something better here?
 		ip = ip.WithZone(fmt.Sprint(scope))
 	}
-	return netaddr.IPPort{
-		IP:   ip,
-		Port: port,
-	}
+	return netaddr.IPPortFrom(ip, port)
 }
 
 func port(v *uint32) uint16 {

net/packet/packet.go

@@ -12,7 +12,6 @@ import (
 
 	"inet.af/netaddr"
 	"tailscale.com/types/ipproto"
-	"tailscale.com/types/strbuilder"
 )
 
 const unknown = ipproto.Unknown
@@ -62,36 +61,17 @@ func (p *Parsed) String() string {
 		return "Unknown{???}"
 	}
 
-	sb := strbuilder.Get()
-	sb.WriteString(p.IPProto.String())
-	sb.WriteByte('{')
-	writeIPPort(sb, p.Src)
-	sb.WriteString(" > ")
-	writeIPPort(sb, p.Dst)
-	sb.WriteByte('}')
-	return sb.String()
-}
-
-// writeIPPort writes ipp.String() into sb, with fewer allocations.
-//
-// TODO: make netaddr more efficient in this area, and retire this func.
-func writeIPPort(sb *strbuilder.Builder, ipp netaddr.IPPort) {
-	if ipp.IP.Is4() {
-		raw := ipp.IP.As4()
-		sb.WriteUint(uint64(raw[0]))
-		sb.WriteByte('.')
-		sb.WriteUint(uint64(raw[1]))
-		sb.WriteByte('.')
-		sb.WriteUint(uint64(raw[2]))
-		sb.WriteByte('.')
-		sb.WriteUint(uint64(raw[3]))
-		sb.WriteByte(':')
-	} else {
-		sb.WriteByte('[')
-		sb.WriteString(ipp.IP.String()) // TODO: faster?
-		sb.WriteString("]:")
-	}
-	sb.WriteUint(uint64(ipp.Port))
+	// max is the maximum reasonable length of the string we are constructing.
+	// It's OK to overshoot, as the temp buffer is allocated on the stack.
+	const max = len("ICMPv6{[ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff%enp5s0]:65535 > [ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff%enp5s0]:65535}")
+	b := make([]byte, 0, max)
+	b = append(b, p.IPProto.String()...)
+	b = append(b, '{')
+	b = p.Src.AppendTo(b)
+	b = append(b, ' ', '>', ' ')
+	b = p.Dst.AppendTo(b)
+	b = append(b, '}')
+	return string(b)
 }
 
 // Decode extracts data from the packet in b into q.
@@ -142,8 +122,8 @@ func (q *Parsed) decode4(b []byte) {
 	}
 
 	// If it's valid IPv4, then the IP addresses are valid
-	q.Src.IP = netaddr.IPv4(b[12], b[13], b[14], b[15])
-	q.Dst.IP = netaddr.IPv4(b[16], b[17], b[18], b[19])
+	q.Src = q.Src.WithIP(netaddr.IPv4(b[12], b[13], b[14], b[15]))
+	q.Dst = q.Dst.WithIP(netaddr.IPv4(b[16], b[17], b[18], b[19]))
 
 	q.subofs = int((b[0] & 0x0F) << 2)
 	if q.subofs > q.length {
@@ -185,8 +165,8 @@ func (q *Parsed) decode4(b []byte) {
 				q.IPProto = unknown
 				return
 			}
-			q.Src.Port = 0
-			q.Dst.Port = 0
+			q.Src = q.Src.WithPort(0)
+			q.Dst = q.Dst.WithPort(0)
 			q.dataofs = q.subofs + icmp4HeaderLength
 			return
 		case ipproto.IGMP:
@@ -198,8 +178,8 @@ func (q *Parsed) decode4(b []byte) {
 				q.IPProto = unknown
 				return
 			}
-			q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-			q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+			q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+			q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 			q.TCPFlags = TCPFlag(sub[13]) & 0x3F
 			headerLength := (sub[12] & 0xF0) >> 2
 			q.dataofs = q.subofs + int(headerLength)
@@ -209,8 +189,8 @@ func (q *Parsed) decode4(b []byte) {
 				q.IPProto = unknown
 				return
 			}
-			q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-			q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+			q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+			q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 			q.dataofs = q.subofs + udpHeaderLength
 			return
 		case ipproto.SCTP:
@@ -218,8 +198,8 @@ func (q *Parsed) decode4(b []byte) {
 				q.IPProto = unknown
 				return
 			}
-			q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-			q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+			q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+			q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 			return
 		case ipproto.TSMP:
 			// Inter-tailscale messages.
@@ -265,8 +245,10 @@ func (q *Parsed) decode6(b []byte) {
 
 	// okay to ignore `ok` here, because IPs pulled from packets are
 	// always well-formed stdlib IPs.
-	q.Src.IP, _ = netaddr.FromStdIP(net.IP(b[8:24]))
-	q.Dst.IP, _ = netaddr.FromStdIP(net.IP(b[24:40]))
+	srcIP, _ := netaddr.FromStdIP(net.IP(b[8:24]))
+	dstIP, _ := netaddr.FromStdIP(net.IP(b[24:40]))
+	q.Src = q.Src.WithIP(srcIP)
+	q.Dst = q.Dst.WithIP(dstIP)
 
 	// We don't support any IPv6 extension headers. Don't try to
 	// be clever. Therefore, the IP subprotocol always starts at
@@ -290,16 +272,16 @@ func (q *Parsed) decode6(b []byte) {
 			q.IPProto = unknown
 			return
 		}
-		q.Src.Port = 0
-		q.Dst.Port = 0
+		q.Src = q.Src.WithPort(0)
+		q.Dst = q.Dst.WithPort(0)
 		q.dataofs = q.subofs + icmp6HeaderLength
 	case ipproto.TCP:
 		if len(sub) < tcpHeaderLength {
 			q.IPProto = unknown
 			return
 		}
-		q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-		q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+		q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+		q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 		q.TCPFlags = TCPFlag(sub[13]) & 0x3F
 		headerLength := (sub[12] & 0xF0) >> 2
 		q.dataofs = q.subofs + int(headerLength)
@@ -309,16 +291,16 @@ func (q *Parsed) decode6(b []byte) {
 			q.IPProto = unknown
 			return
 		}
-		q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-		q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+		q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+		q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 		q.dataofs = q.subofs + udpHeaderLength
 	case ipproto.SCTP:
 		if len(sub) < sctpHeaderLength {
 			q.IPProto = unknown
 			return
 		}
-		q.Src.Port = binary.BigEndian.Uint16(sub[0:2])
-		q.Dst.Port = binary.BigEndian.Uint16(sub[2:4])
+		q.Src = q.Src.WithPort(binary.BigEndian.Uint16(sub[0:2]))
+		q.Dst = q.Dst.WithPort(binary.BigEndian.Uint16(sub[2:4]))
 		return
 	case ipproto.TSMP:
 		// Inter-tailscale messages.
@@ -338,8 +320,8 @@ func (q *Parsed) IP4Header() IP4Header {
 	return IP4Header{
 		IPID:    ipid,
 		IPProto: q.IPProto,
-		Src:     q.Src.IP,
-		Dst:     q.Dst.IP,
+		Src:     q.Src.IP(),
+		Dst:     q.Dst.IP(),
 	}
 }
 
@@ -351,8 +333,8 @@ func (q *Parsed) IP6Header() IP6Header {
 	return IP6Header{
 		IPID:    ipid,
 		IPProto: q.IPProto,
-		Src:     q.Src.IP,
-		Dst:     q.Dst.IP,
+		Src:     q.Src.IP(),
+		Dst:     q.Dst.IP(),
 	}
 }
 
@@ -373,8 +355,8 @@ func (q *Parsed) UDP4Header() UDP4Header {
 	}
 	return UDP4Header{
 		IP4Header: q.IP4Header(),
-		SrcPort:   q.Src.Port,
-		DstPort:   q.Dst.Port,
+		SrcPort:   q.Src.Port(),
+		DstPort:   q.Dst.Port(),
 	}
 }
 

net/packet/packet_test.go

@@ -378,11 +378,9 @@ func TestParsedString(t *testing.T) {
 		})
 	}
 
-	var sink string
 	allocs := testing.AllocsPerRun(1000, func() {
-		sink = tests[0].qdecode.String()
+		sinkString = tests[0].qdecode.String()
 	})
-	_ = sink
 	if allocs != 1 {
 		t.Errorf("allocs = %v; want 1", allocs)
 	}
@@ -532,3 +530,33 @@ func TestMarshalResponse(t *testing.T) {
 		})
 	}
 }
+
+var sinkString string
+
+func BenchmarkString(b *testing.B) {
+	benches := []struct {
+		name string
+		buf  []byte
+	}{
+		{"tcp4", tcp4PacketBuffer},
+		{"tcp6", tcp6RequestBuffer},
+		{"udp4", udp4RequestBuffer},
+		{"udp6", udp6RequestBuffer},
+		{"icmp4", icmp4RequestBuffer},
+		{"icmp6", icmp6PacketBuffer},
+		{"igmp", igmpPacketBuffer},
+		{"unknown", unknownPacketBuffer},
+	}
+
+	for _, bench := range benches {
+		b.Run(bench.name, func(b *testing.B) {
+			b.ReportAllocs()
+			var p Parsed
+			p.Decode(bench.buf)
+			b.ResetTimer()
+			for i := 0; i < b.N; i++ {
+				sinkString = p.String()
+			}
+		})
+	}
+}

net/packet/tsmp.go

@@ -14,6 +14,7 @@ import (
 	"encoding/binary"
 	"errors"
 	"fmt"
+	"log"
 
 	"inet.af/netaddr"
 	"tailscale.com/net/flowtrack"
@@ -143,7 +144,7 @@ func (h TailscaleRejectedHeader) Marshal(buf []byte) error {
 	if len(buf) > maxPacketLength {
 		return errLargePacket
 	}
-	if h.Src.IP.Is4() {
+	if h.Src.IP().Is4() {
 		iph := IP4Header{
 			IPProto: ipproto.TSMP,
 			Src:     h.IPSrc,
@@ -151,7 +152,7 @@ func (h TailscaleRejectedHeader) Marshal(buf []byte) error {
 		}
 		iph.Marshal(buf)
 		buf = buf[ip4HeaderLength:]
-	} else if h.Src.IP.Is6() {
+	} else if h.Src.IP().Is6() {
 		iph := IP6Header{
 			IPProto: ipproto.TSMP,
 			Src:     h.IPSrc,
@@ -165,8 +166,8 @@ func (h TailscaleRejectedHeader) Marshal(buf []byte) error {
 	buf[0] = byte(TSMPTypeRejectedConn)
 	buf[1] = byte(h.Proto)
 	buf[2] = byte(h.Reason)
-	binary.BigEndian.PutUint16(buf[3:5], h.Src.Port)
-	binary.BigEndian.PutUint16(buf[5:7], h.Dst.Port)
+	binary.BigEndian.PutUint16(buf[3:5], h.Src.Port())
+	binary.BigEndian.PutUint16(buf[5:7], h.Dst.Port())
 
 	if h.hasFlags() {
 		var flags byte
@@ -190,10 +191,10 @@ func (pp *Parsed) AsTailscaleRejectedHeader() (h TailscaleRejectedHeader, ok boo
 	h = TailscaleRejectedHeader{
 		Proto:  ipproto.Proto(p[1]),
 		Reason: TailscaleRejectReason(p[2]),
-		IPSrc:  pp.Src.IP,
-		IPDst:  pp.Dst.IP,
-		Src:    netaddr.IPPort{IP: pp.Dst.IP, Port: binary.BigEndian.Uint16(p[3:5])},
-		Dst:    netaddr.IPPort{IP: pp.Src.IP, Port: binary.BigEndian.Uint16(p[5:7])},
+		IPSrc:  pp.Src.IP(),
+		IPDst:  pp.Dst.IP(),
+		Src:    netaddr.IPPortFrom(pp.Dst.IP(), binary.BigEndian.Uint16(p[3:5])),
+		Dst:    netaddr.IPPortFrom(pp.Src.IP(), binary.BigEndian.Uint16(p[5:7])),
 	}
 	if len(p) > 7 {
 		flags := p[7]
@@ -232,6 +233,7 @@ type TSMPPongReply struct {
 // AsTSMPPong returns pp as a TSMPPongReply and whether it is one.
 // The pong.IPHeader field is not populated.
 func (pp *Parsed) AsTSMPPong() (pong TSMPPongReply, ok bool) {
+	log.Println("TSMPPONG")
 	if pp.IPProto != ipproto.TSMP {
 		return
 	}

net/portmapper/portmapper.go

@@ -84,7 +84,7 @@ type pmpMapping struct {
 
 // externalValid reports whether m.external is valid, with both its IP and Port populated.
 func (m *pmpMapping) externalValid() bool {
-	return !m.external.IP.IsZero() && m.external.Port != 0
+	return !m.external.IP().IsZero() && m.external.Port() != 0
 }
 
 // release does a best effort fire-and-forget release of the PMP mapping m.
@@ -94,8 +94,8 @@ func (m *pmpMapping) release() {
 		return
 	}
 	defer uc.Close()
-	pkt := buildPMPRequestMappingPacket(m.internal.Port, m.external.Port, pmpMapLifetimeDelete)
-	uc.WriteTo(pkt, netaddr.IPPort{IP: m.gw, Port: pmpPort}.UDPAddr())
+	pkt := buildPMPRequestMappingPacket(m.internal.Port(), m.external.Port(), pmpMapLifetimeDelete)
+	uc.WriteTo(pkt, netaddr.IPPortFrom(m.gw, pmpPort).UDPAddr())
 }
 
 // NewClient returns a new portmapping client.
@@ -256,7 +256,7 @@ func (c *Client) CreateOrGetMapping(ctx context.Context) (external netaddr.IPPor
 	localPort := c.localPort
 	m := &pmpMapping{
 		gw:       gw,
-		internal: netaddr.IPPort{IP: myIP, Port: localPort},
+		internal: netaddr.IPPortFrom(myIP, localPort),
 	}
 
 	// prevPort is the port we had most previously, if any. We try
@@ -271,7 +271,7 @@ func (c *Client) CreateOrGetMapping(ctx context.Context) (external netaddr.IPPor
 			return m.external, nil
 		}
 		// The mapping might still be valid, so just try to renew it.
-		prevPort = m.external.Port
+		prevPort = m.external.Port()
 	}
 
 	// If we just did a Probe (e.g. via netchecker) but didn't
@@ -279,7 +279,7 @@ func (c *Client) CreateOrGetMapping(ctx context.Context) (external netaddr.IPPor
 	// again. Cuts down latency for most clients.
 	haveRecentPMP := c.sawPMPRecentlyLocked()
 	if haveRecentPMP {
-		m.external.IP = c.pmpPubIP
+		m.external = m.external.WithIP(c.pmpPubIP)
 	}
 	if c.lastProbe.After(now.Add(-5*time.Second)) && !haveRecentPMP {
 		c.mu.Unlock()
@@ -297,11 +297,11 @@ func (c *Client) CreateOrGetMapping(ctx context.Context) (external netaddr.IPPor
 	uc.SetReadDeadline(time.Now().Add(portMapServiceTimeout))
 	defer closeCloserOnContextDone(ctx, uc)()
 
-	pmpAddr := netaddr.IPPort{IP: gw, Port: pmpPort}
+	pmpAddr := netaddr.IPPortFrom(gw, pmpPort)
 	pmpAddru := pmpAddr.UDPAddr()
 
 	// Ask for our external address if needed.
-	if m.external.IP.IsZero() {
+	if m.external.IP().IsZero() {
 		if _, err := uc.WriteTo(pmpReqExternalAddrPacket, pmpAddru); err != nil {
 			return netaddr.IPPort{}, err
 		}
@@ -337,10 +337,10 @@ func (c *Client) CreateOrGetMapping(ctx context.Context) (external netaddr.IPPor
 				return netaddr.IPPort{}, NoMappingError{fmt.Errorf("PMP response Op=0x%x,Res=0x%x", pres.OpCode, pres.ResultCode)}
 			}
 			if pres.OpCode == pmpOpReply|pmpOpMapPublicAddr {
-				m.external.IP = pres.PublicAddr
+				m.external = m.external.WithIP(pres.PublicAddr)
 			}
 			if pres.OpCode == pmpOpReply|pmpOpMapUDP {
-				m.external.Port = pres.ExternalPort
+				m.external = m.external.WithPort(pres.ExternalPort)
 				d := time.Duration(pres.MappingValidSeconds) * time.Second
 				d /= 2 // renew in half the time
 				m.useUntil = time.Now().Add(d)
@@ -468,9 +468,9 @@ func (c *Client) Probe(ctx context.Context) (res ProbeResult, err error) {
 	defer cancel()
 	defer closeCloserOnContextDone(ctx, uc)()
 
-	pcpAddr := netaddr.IPPort{IP: gw, Port: pcpPort}.UDPAddr()
-	pmpAddr := netaddr.IPPort{IP: gw, Port: pmpPort}.UDPAddr()
-	upnpAddr := netaddr.IPPort{IP: gw, Port: upnpPort}.UDPAddr()
+	pcpAddr := netaddr.IPPortFrom(gw, pcpPort).UDPAddr()
+	pmpAddr := netaddr.IPPortFrom(gw, pmpPort).UDPAddr()
+	upnpAddr := netaddr.IPPortFrom(gw, upnpPort).UDPAddr()
 
 	// Don't send probes to services that we recently learned (for
 	// the same gw/myIP) are available. See

net/stun/stuntest/stuntest.go

@@ -26,11 +26,11 @@ type stunStats struct {
 	readIPv6 int
 }
 
-func Serve(t *testing.T) (addr *net.UDPAddr, cleanupFn func()) {
+func Serve(t testing.TB) (addr *net.UDPAddr, cleanupFn func()) {
 	return ServeWithPacketListener(t, nettype.Std{})
 }
 
-func ServeWithPacketListener(t *testing.T, ln nettype.PacketListener) (addr *net.UDPAddr, cleanupFn func()) {
+func ServeWithPacketListener(t testing.TB, ln nettype.PacketListener) (addr *net.UDPAddr, cleanupFn func()) {
 	t.Helper()
 
 	// TODO(crawshaw): use stats to test re-STUN logic
@@ -52,7 +52,7 @@ func ServeWithPacketListener(t *testing.T, ln nettype.PacketListener) (addr *net
 	}
 }
 
-func runSTUN(t *testing.T, pc net.PacketConn, stats *stunStats, done chan<- struct{}) {
+func runSTUN(t testing.TB, pc net.PacketConn, stats *stunStats, done chan<- struct{}) {
 	defer close(done)
 
 	var buf [64 << 10]byte

net/tsaddr/tsaddr.go

@@ -77,10 +77,10 @@ func Tailscale4To6Range() netaddr.IPPrefix {
 // used for ephemeral IPv6-only Tailscale nodes.
 func TailscaleEphemeral6Range() netaddr.IPPrefix {
 	// This IP range has no significance, beyond being a subset of
-	// TailscaleULARange. The bits from /48 to /104 were picked at
+	// TailscaleULARange. The bits from /48 to /64 were picked at
 	// random, with the only criterion being to not be the conflict
 	// with the Tailscale4To6Range above.
-	ulaEph6Range.Do(func() { mustPrefix(&ulaEph6Range.v, "fd7a:115c:a1e0:7234:6e44:306d:2100::/104") })
+	ulaEph6Range.Do(func() { mustPrefix(&ulaEph6Range.v, "fd7a:115c:a1e0:efe3::/64") })
 	return ulaEph6Range.v
 }
 
@@ -92,7 +92,7 @@ func TailscaleEphemeral6Range() netaddr.IPPrefix {
 // Currently used to work around a Windows limitation when programming
 // IPv6 routes in corner cases.
 func Tailscale4To6Placeholder() netaddr.IP {
-	return Tailscale4To6Range().IP
+	return Tailscale4To6Range().IP()
 }
 
 // Tailscale4To6 returns a Tailscale IPv6 address that maps 1:1 to the
@@ -102,7 +102,7 @@ func Tailscale4To6(ipv4 netaddr.IP) netaddr.IP {
 	if !ipv4.Is4() || !IsTailscaleIP(ipv4) {
 		return netaddr.IP{}
 	}
-	ret := Tailscale4To6Range().IP.As16()
+	ret := Tailscale4To6Range().IP().As16()
 	v4 := ipv4.As4()
 	copy(ret[13:], v4[1:])
 	return netaddr.IPFrom16(ret)
@@ -138,3 +138,50 @@ type onceIP struct {
 	sync.Once
 	v netaddr.IP
 }
+
+// NewContainsIPFunc returns a func that reports whether ip is in addrs.
+//
+// It's optimized for the cases of addrs being empty and addrs
+// containing 1 or 2 single-IP prefixes (such as one IPv4 address and
+// one IPv6 address).
+//
+// Otherwise the implementation is somewhat slow.
+func NewContainsIPFunc(addrs []netaddr.IPPrefix) func(ip netaddr.IP) bool {
+	// Specialize the three common cases: no address, just IPv4
+	// (or just IPv6), and both IPv4 and IPv6.
+	if len(addrs) == 0 {
+		return func(netaddr.IP) bool { return false }
+	}
+	// If any addr is more than a single IP, then just do the slow
+	// linear thing until
+	// https://github.com/inetaf/netaddr/issues/139 is done.
+	for _, a := range addrs {
+		if a.IsSingleIP() {
+			continue
+		}
+		acopy := append([]netaddr.IPPrefix(nil), addrs...)
+		return func(ip netaddr.IP) bool {
+			for _, a := range acopy {
+				if a.Contains(ip) {
+					return true
+				}
+			}
+			return false
+		}
+	}
+	// Fast paths for 1 and 2 IPs:
+	if len(addrs) == 1 {
+		a := addrs[0]
+		return func(ip netaddr.IP) bool { return ip == a.IP() }
+	}
+	if len(addrs) == 2 {
+		a, b := addrs[0], addrs[1]
+		return func(ip netaddr.IP) bool { return ip == a.IP() || ip == b.IP() }
+	}
+	// General case:
+	m := map[netaddr.IP]bool{}
+	for _, a := range addrs {
+		m[a.IP()] = true
+	}
+	return func(ip netaddr.IP) bool { return m[ip] }
+}

net/tsaddr/tsaddr_test.go

@@ -64,3 +64,32 @@ func TestIsUla(t *testing.T) {
 		}
 	}
 }
+
+func TestNewContainsIPFunc(t *testing.T) {
+	f := NewContainsIPFunc([]netaddr.IPPrefix{netaddr.MustParseIPPrefix("10.0.0.0/8")})
+	if f(netaddr.MustParseIP("8.8.8.8")) {
+		t.Fatal("bad")
+	}
+	if !f(netaddr.MustParseIP("10.1.2.3")) {
+		t.Fatal("bad")
+	}
+	f = NewContainsIPFunc([]netaddr.IPPrefix{netaddr.MustParseIPPrefix("10.1.2.3/32")})
+	if !f(netaddr.MustParseIP("10.1.2.3")) {
+		t.Fatal("bad")
+	}
+	f = NewContainsIPFunc([]netaddr.IPPrefix{
+		netaddr.MustParseIPPrefix("10.1.2.3/32"),
+		netaddr.MustParseIPPrefix("::2/128"),
+	})
+	if !f(netaddr.MustParseIP("::2")) {
+		t.Fatal("bad")
+	}
+	f = NewContainsIPFunc([]netaddr.IPPrefix{
+		netaddr.MustParseIPPrefix("10.1.2.3/32"),
+		netaddr.MustParseIPPrefix("10.1.2.4/32"),
+		netaddr.MustParseIPPrefix("::2/128"),
+	})
+	if !f(netaddr.MustParseIP("10.1.2.4")) {
+		t.Fatal("bad")
+	}
+}

net/tstun/fake.go

@@ -8,7 +8,7 @@ import (
 	"io"
 	"os"
 
-	"github.com/tailscale/wireguard-go/tun"
+	"golang.zx2c4.com/wireguard/tun"
 )
 
 type fakeTUN struct {

net/tstun/ifstatus_noop.go

@@ -9,7 +9,7 @@ package tstun
 import (
 	"time"
 
-	"github.com/tailscale/wireguard-go/tun"
+	"golang.zx2c4.com/wireguard/tun"
 	"tailscale.com/types/logger"
 )
 

net/tstun/ifstatus_windows.go

@@ -9,7 +9,7 @@ import (
 	"sync"
 	"time"
 
-	"github.com/tailscale/wireguard-go/tun"
+	"golang.zx2c4.com/wireguard/tun"
 	"golang.zx2c4.com/wireguard/windows/tunnel/winipcfg"
 	"tailscale.com/types/logger"
 )
@@ -93,7 +93,6 @@ func waitInterfaceUp(iface tun.Device, timeout time.Duration, logf logger.Logf)
 			iw.logf("TUN interface is up after %v", time.Since(t0))
 			return nil
 		case <-ticker.C:
-			break
 		}
 
 		if iw.isUp() {

net/tstun/tun.go

@@ -13,7 +13,7 @@ import (
 	"runtime"
 	"time"
 
-	"github.com/tailscale/wireguard-go/tun"
+	"golang.zx2c4.com/wireguard/tun"
 	"tailscale.com/types/logger"
 	"tailscale.com/version/distro"
 )

net/tstun/tun_notwindows.go

@@ -6,7 +6,7 @@
 
 package tstun
 
-import "github.com/tailscale/wireguard-go/tun"
+import "golang.zx2c4.com/wireguard/tun"
 
 func interfaceName(dev tun.Device) (string, error) {
 	return dev.Name()

net/tstun/tun_windows.go

@@ -5,9 +5,9 @@
 package tstun
 
 import (
-	"github.com/tailscale/wireguard-go/tun"
-	"github.com/tailscale/wireguard-go/tun/wintun"
 	"golang.org/x/sys/windows"
+	"golang.zx2c4.com/wireguard/tun"
+	"golang.zx2c4.com/wireguard/tun/wintun"
 	"golang.zx2c4.com/wireguard/windows/tunnel/winipcfg"
 )
 

net/tstun/wrap.go

@@ -9,13 +9,14 @@ package tstun
 import (
 	"errors"
 	"io"
+	"log"
 	"os"
 	"sync"
 	"sync/atomic"
 	"time"
 
-	"github.com/tailscale/wireguard-go/device"
-	"github.com/tailscale/wireguard-go/tun"
+	"golang.zx2c4.com/wireguard/device"
+	"golang.zx2c4.com/wireguard/tun"
 	"inet.af/netaddr"
 	"tailscale.com/net/packet"
 	"tailscale.com/types/ipproto"
@@ -90,7 +91,12 @@ type Wrapper struct {
 	// to discard an empty packet instead of sending it through t.outbound.
 	outbound chan []byte
 
-	// fitler stores the currently active package filter
+	// eventsUpDown yields up and down tun.Events that arrive on a Wrapper's events channel.
+	eventsUpDown chan tun.Event
+	// eventsOther yields non-up-and-down tun.Events that arrive on a Wrapper's events channel.
+	eventsOther chan tun.Event
+
+	// filter atomically stores the currently active packet filter
 	filter atomic.Value // of *filter.Filter
 	// filterFlags control the verbosity of logging packet drops/accepts.
 	filterFlags filter.RunFlags
@@ -130,11 +136,14 @@ func Wrap(logf logger.Logf, tdev tun.Device) *Wrapper {
 		closed:         make(chan struct{}),
 		errors:         make(chan error),
 		outbound:       make(chan []byte),
+		eventsUpDown:   make(chan tun.Event),
+		eventsOther:    make(chan tun.Event),
 		// TODO(dmytro): (highly rate-limited) hexdumps should happen on unknown packets.
 		filterFlags: filter.LogAccepts | filter.LogDrops,
 	}
 
 	go tun.poll()
+	go tun.pumpEvents()
 	// The buffer starts out consumed.
 	tun.bufferConsumed <- struct{}{}
 
@@ -160,8 +169,50 @@ func (t *Wrapper) Close() error {
 	return err
 }
 
+// pumpEvents copies events from t.tdev to t.eventsUpDown and t.eventsOther.
+// pumpEvents exits when t.tdev.events or t.closed is closed.
+// pumpEvents closes t.eventsUpDown and t.eventsOther when it exits.
+func (t *Wrapper) pumpEvents() {
+	defer close(t.eventsUpDown)
+	defer close(t.eventsOther)
+	src := t.tdev.Events()
+	for {
+		// Retrieve an event from the TUN device.
+		var event tun.Event
+		var ok bool
+		select {
+		case <-t.closed:
+			return
+		case event, ok = <-src:
+			if !ok {
+				return
+			}
+		}
+
+		// Pass along event to the correct recipient.
+		// Though event is a bitmask, in practice there is only ever one bit set at a time.
+		dst := t.eventsOther
+		if event&(tun.EventUp|tun.EventDown) != 0 {
+			dst = t.eventsUpDown
+		}
+		select {
+		case <-t.closed:
+			return
+		case dst <- event:
+		}
+	}
+}
+
+// EventsUpDown returns a TUN event channel that contains all Up and Down events.
+func (t *Wrapper) EventsUpDown() chan tun.Event {
+	return t.eventsUpDown
+}
+
+// Events returns a TUN event channel that contains all non-Up, non-Down events.
+// It is named Events because it is the set of events that we want to expose to wireguard-go,
+// and Events is the name specified by the wireguard-go tun.Device interface.
 func (t *Wrapper) Events() chan tun.Event {
-	return t.tdev.Events()
+	return t.eventsOther
 }
 
 func (t *Wrapper) File() *os.File {
@@ -226,6 +277,7 @@ func (t *Wrapper) poll() {
 var magicDNSIPPort = netaddr.MustParseIPPort("100.100.100.100:0")
 
 func (t *Wrapper) filterOut(p *packet.Parsed) filter.Response {
+	log.Println("FILTEROUT")
 	// Fake ICMP echo responses to MagicDNS (100.100.100.100).
 	if p.IsEchoRequest() && p.Dst == magicDNSIPPort {
 		header := p.ICMP4Header()
@@ -302,7 +354,7 @@ func (t *Wrapper) Read(buf []byte, offset int) (int, error) {
 	p.Decode(buf[offset : offset+n])
 
 	if m, ok := t.destIPActivity.Load().(map[netaddr.IP]func()); ok {
-		if fn := m[p.Dst.IP]; fn != nil {
+		if fn := m[p.Dst.IP()]; fn != nil {
 			fn()
 		}
 	}
@@ -326,6 +378,7 @@ func (t *Wrapper) Read(buf []byte, offset int) (int, error) {
 }
 
 func (t *Wrapper) filterIn(buf []byte) filter.Response {
+	log.Println("FILTERIN")
 	p := parsedPacketPool.Get().(*packet.Parsed)
 	defer parsedPacketPool.Put(p)
 	p.Decode(buf)
@@ -362,7 +415,7 @@ func (t *Wrapper) filterIn(buf []byte) filter.Response {
 		p.IPProto == ipproto.TCP &&
 		p.TCPFlags&packet.TCPSyn != 0 &&
 		t.PeerAPIPort != nil {
-		if port, ok := t.PeerAPIPort(p.Dst.IP); ok && port == p.Dst.Port {
+		if port, ok := t.PeerAPIPort(p.Dst.IP()); ok && port == p.Dst.Port() {
 			outcome = filter.Accept
 		}
 	}
@@ -375,8 +428,8 @@ func (t *Wrapper) filterIn(buf []byte) filter.Response {
 		// can show them a rejection history with reasons.
 		if p.IPVersion == 4 && p.IPProto == ipproto.TCP && p.TCPFlags&packet.TCPSyn != 0 && !t.disableTSMPRejected {
 			rj := packet.TailscaleRejectedHeader{
-				IPSrc:  p.Dst.IP,
-				IPDst:  p.Src.IP,
+				IPSrc:  p.Dst.IP(),
+				IPDst:  p.Src.IP(),
 				Src:    p.Src,
 				Dst:    p.Dst,
 				Proto:  p.IPProto,
@@ -407,13 +460,22 @@ func (t *Wrapper) filterIn(buf []byte) filter.Response {
 // like wireguard-go/tun.Device.Write.
 func (t *Wrapper) Write(buf []byte, offset int) (int, error) {
 	if !t.disableFilter {
-		res := t.filterIn(buf[offset:])
-		if res == filter.DropSilently {
+		if t.filterIn(buf[offset:]) != filter.Accept {
+			// If we're not accepting the packet, lie to wireguard-go and pretend
+			// that everything is okay with a nil error, so wireguard-go
+			// doesn't log about this Write "failure".
+			//
+			// We return len(buf), but the ill-defined wireguard-go/tun.Device.Write
+			// method doesn't specify how the offset affects the return value.
+			// In fact, the Linux implementation does one of two different things depending
+			// on how the /dev/net/tun was created. But fortunately the wireguard-go
+			// code ignores the int return and only looks at the error:
+			//
+			//     device/receive.go: _, err = device.tun.device.Write(....)
+			//
+			// TODO(bradfitz): fix upstream interface docs, implementation.
 			return len(buf), nil
 		}
-		if res != filter.Accept {
-			return 0, ErrFiltered
-		}
 	}
 
 	t.noteActivity()
@@ -473,11 +535,12 @@ func (t *Wrapper) InjectInboundCopy(packet []byte) error {
 }
 
 func (t *Wrapper) injectOutboundPong(pp *packet.Parsed, req packet.TSMPPingRequest) {
+	log.Println("INJECT OUTBOUND")
 	pong := packet.TSMPPongReply{
 		Data: req.Data,
 	}
 	if t.PeerAPIPort != nil {
-		pong.PeerAPIPort, _ = t.PeerAPIPort(pp.Dst.IP)
+		pong.PeerAPIPort, _ = t.PeerAPIPort(pp.Dst.IP())
 	}
 	switch pp.IPVersion {
 	case 4:
@@ -509,8 +572,10 @@ func (t *Wrapper) InjectOutbound(packet []byte) error {
 	}
 	select {
 	case <-t.closed:
+		log.Println("Closed")
 		return ErrClosed
 	case t.outbound <- packet:
+		log.Println("t.outbound <- packet")
 		return nil
 	}
 }

net/tstun/wrap_test.go

@@ -14,7 +14,7 @@ import (
 	"testing"
 	"unsafe"
 
-	"github.com/tailscale/wireguard-go/tun/tuntest"
+	"golang.zx2c4.com/wireguard/tun/tuntest"
 	"inet.af/netaddr"
 	"tailscale.com/net/packet"
 	"tailscale.com/types/ipproto"
@@ -82,7 +82,7 @@ func nets(nets ...string) (ret []netaddr.IPPrefix) {
 			if ip.Is6() {
 				bits = 128
 			}
-			ret = append(ret, netaddr.IPPrefix{IP: ip, Bits: bits})
+			ret = append(ret, netaddr.IPPrefixFrom(ip, bits))
 		} else {
 			pfx, err := netaddr.ParseIPPrefix(s)
 			if err != nil {
@@ -329,11 +329,14 @@ func TestFilter(t *testing.T) {
 			var filtered bool
 
 			if tt.dir == in {
+				// Use the side effect of updating the last
+				// activity atomic to determine whether the
+				// data was actually filtered.
+				// If it stays zero, nothing made it through
+				// to the wrapped TUN.
+				atomic.StoreInt64(&tun.lastActivityAtomic, 0)
 				_, err = tun.Write(tt.data, 0)
-				if err == ErrFiltered {
-					filtered = true
-					err = nil
-				}
+				filtered = atomic.LoadInt64(&tun.lastActivityAtomic) == 0
 			} else {
 				chtun.Outbound <- tt.data
 				n, err = tun.Read(buf[:], 0)

paths/paths.go

@@ -13,9 +13,9 @@ import (
 	"sync/atomic"
 )
 
-// IOSSharedDir is a string set by the iOS app on start
+// AppSharedDir is a string set by the iOS or Android app on start
 // containing a directory we can read/write in.
-var IOSSharedDir atomic.Value
+var AppSharedDir atomic.Value
 
 // DefaultTailscaledSocket returns the path to the tailscaled Unix socket
 // or the empty string if there's no reasonable default.
@@ -26,6 +26,13 @@ func DefaultTailscaledSocket() string {
 	if runtime.GOOS == "darwin" {
 		return "/var/run/tailscaled.socket"
 	}
+	if runtime.GOOS == "linux" {
+		// TODO(crawshaw): does this path change with DSM7?
+		const synologySock = "/volume1/@appstore/Tailscale/var/tailscaled.sock" // SYNOPKG_PKGDEST in scripts/installer
+		if fi, err := os.Stat(filepath.Dir(synologySock)); err == nil && fi.IsDir() {
+			return synologySock
+		}
+	}
 	if fi, err := os.Stat("/var/run"); err == nil && fi.IsDir() {
 		return "/var/run/tailscale/tailscaled.sock"
 	}