net/portmapper: fix UPnP probing, work against all ports

Prior to Tailscale 1.12 it detected UPnP on any port.
Starting with Tailscale 1.11.x, it stopped detecting UPnP on all ports.

Then start plumbing its discovered Location header port number to the
code that was assuming port 5000.

Fixes #2109

Signed-off-by: Brad Fitzpatrick <bradfitz@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/go_generate.yml

@@ -0,0 +1,34 @@
+name: go generate
+
+on:
+  push:
+    branches:
+      - main
+      - "release-branch/*"
+  pull_request:
+    branches:
+      - "*"
+
+jobs:
+  check:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Set up Go
+        uses: actions/setup-go@v1
+        with:
+          go-version: 1.16
+
+      - name: Check out code
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+
+      - name: check 'go generate' is clean
+        run: |
+          mkdir gentools
+          go build -o gentools/stringer golang.org/x/tools/cmd/stringer
+          PATH="$PATH:$(pwd)/gentools" go generate ./...
+          echo
+          echo
+          git diff --name-only --exit-code || (echo "The files above need updating. Please run 'go generate'."; exit 1)

.github/workflows/xe-experimental-vm-test.yml

@@ -0,0 +1,45 @@
+name: "integration-vms"
+
+on:
+  pull_request:
+    paths:
+      - "tstest/integration/vms/**"
+  release:
+    types: [ created ]
+
+jobs:
+  experimental-linux-vm-test:
+    # To set up a new runner, see tstest/integration/vms/runner.nix
+    runs-on: [ self-hosted, linux, vm_integration_test ]
+
+    if: "!contains(github.event.head_commit.message, '[ci skip]')"
+
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v1
+
+      - name: Download VM Images
+        run: go test ./tstest/integration/vms -run-vm-tests -run=Download -timeout=60m -no-s3
+        env:
+          XDG_CACHE_HOME: "/var/lib/ghrunner/cache"
+
+      - name: Run VM tests
+        run: go test ./tstest/integration/vms -v -run-vm-tests
+        env:
+          TMPDIR: "/tmp"
+          XDG_CACHE_HOME: "/var/lib/ghrunner/cache"
+
+      - 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'

Dockerfile

@@ -42,8 +42,7 @@ FROM golang:1.16-alpine AS build-env
 
 WORKDIR /go/src/tailscale
 
-COPY go.mod .
-COPY go.sum .
+COPY go.mod go.sum ./
 RUN go mod download
 
 COPY . .

VERSION.txt

@@ -1 +1 @@
-1.7.0
+1.13.0

api.md

@@ -471,15 +471,16 @@ Determines what rules match for a user on an ACL without saving the ACL to the s
 ##### Parameters
 
 ###### Query Parameters
-`user` - A user's email. The provided ACL is queried with this user to determine which rules match.
+`type` - can be 'user' or 'ipport'
+`previewFor` - if type=user, a user's email. If type=ipport, a IP address + port like "10.0.0.1:80".
+The provided ACL is queried with this paramater to determine which rules match.
 
 ###### POST Body
 ACL JSON or HuJSON (see https://tailscale.com/kb/1018/acls)
 
 ##### Example
 ```
-POST /api/v2/tailnet/example.com/acl/preiew
-curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl?user=user1@example.com' \
+curl 'https://api.tailscale.com/api/v2/tailnet/example.com/acl/preview?previewFor=user1@example.com&type=user' \
   -u "tskey-yourapikey123:" \
   --data-binary '// Example/default ACLs for unrestricted connections.
 {

build_dist.sh

@@ -11,6 +11,36 @@
 
 set -eu
 
-eval $(./version/version.sh)
+IFS=".$IFS" read -r major minor patch <VERSION.txt
+git_hash=$(git rev-parse HEAD)
+if ! git diff-index --quiet HEAD; then
+	git_hash="${git_hash}-dirty"
+fi
+base_hash=$(git rev-list --max-count=1 HEAD -- VERSION.txt)
+change_count=$(git rev-list --count HEAD "^$base_hash")
+short_hash=$(echo "$git_hash" | cut -c1-9)
 
-exec go build -tags xversion -ldflags "-X tailscale.com/version.Long=${VERSION_LONG} -X tailscale.com/version.Short=${VERSION_SHORT} -X tailscale.com/version.GitCommit=${VERSION_GIT_HASH}" "$@"
+if expr "$minor" : "[0-9]*[13579]$" >/dev/null; then
+	patch="$change_count"
+	change_suffix=""
+elif [ "$change_count" != "0" ]; then
+	change_suffix="-$change_count"
+else
+	change_suffix=""
+fi
+
+long_suffix="$change_suffix-t$short_hash"
+SHORT="$major.$minor.$patch"
+LONG="${SHORT}$long_suffix"
+GIT_HASH="$git_hash"
+
+if [ "$1" = "shellvars" ]; then
+	cat <<EOF
+VERSION_SHORT="$SHORT"
+VERSION_LONG="$LONG"
+VERSION_GIT_HASH="$GIT_HASH"
+EOF
+	exit 0
+fi
+
+exec go build -ldflags "-X tailscale.com/version.Long=${LONG} -X tailscale.com/version.Short=${SHORT} -X tailscale.com/version.GitCommit=${GIT_HASH}" "$@"

build_docker.sh

@@ -25,7 +25,7 @@
 
 set -eu
 
-eval $(./version/version.sh)
+eval $(./build_dist.sh shellvars)
 
 docker build \
   --build-arg VERSION_LONG=$VERSION_LONG \

client/tailscale/tailscale.go

@@ -24,6 +24,7 @@ import (
 	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/paths"
 	"tailscale.com/safesocket"
+	"tailscale.com/tailcfg"
 )
 
 // TailscaledSocket is the tailscaled Unix socket.
@@ -256,3 +257,39 @@ func Logout(ctx context.Context) error {
 	_, err := send(ctx, "POST", "/localapi/v0/logout", http.StatusNoContent, nil)
 	return err
 }
+
+// SetDNS adds a DNS TXT record for the given domain name, containing
+// the provided TXT value. The intended use case is answering
+// LetsEncrypt/ACME dns-01 challenges.
+//
+// The control plane will only permit SetDNS requests with very
+// specific names and values. The name should be
+// "_acme-challenge." + your node's MagicDNS name. It's expected that
+// clients cache the certs from LetsEncrypt (or whichever CA is
+// providing them) and only request new ones as needed; the control plane
+// rate limits SetDNS requests.
+//
+// This is a low-level interface; it's expected that most Tailscale
+// users use a higher level interface to getting/using TLS
+// certificates.
+func SetDNS(ctx context.Context, name, value string) error {
+	v := url.Values{}
+	v.Set("name", name)
+	v.Set("value", value)
+	_, err := send(ctx, "POST", "/localapi/v0/set-dns?"+v.Encode(), 200, nil)
+	return err
+}
+
+// CurrentDERPMap returns the current DERPMap that is being used by the local tailscaled.
+// It is intended to be used with netcheck to see availability of DERPs.
+func CurrentDERPMap(ctx context.Context) (*tailcfg.DERPMap, error) {
+	var derpMap tailcfg.DERPMap
+	res, err := send(ctx, "GET", "/localapi/v0/derpmap", 200, nil)
+	if err != nil {
+		return nil, err
+	}
+	if err = json.Unmarshal(res, &derpMap); err != nil {
+		return nil, fmt.Errorf("invalid derp map json: %w", err)
+	}
+	return &derpMap, nil
+}

cmd/addlicense/main.go

@@ -0,0 +1,77 @@
+// 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.
+
+// Program addlicense adds a license header to a file.
+// It is intended for use with 'go generate',
+// so it has a slightly weird usage.
+package main
+
+import (
+	"flag"
+	"fmt"
+	"os"
+	"os/exec"
+)
+
+var (
+	year = flag.Int("year", 0, "copyright year")
+	file = flag.String("file", "", "file to modify")
+)
+
+func usage() {
+	fmt.Fprintf(os.Stderr, `
+usage: addlicense -year YEAR -file FILE <subcommand args...>
+`[1:])
+
+	flag.PrintDefaults()
+	fmt.Fprintf(os.Stderr, `
+addlicense adds a Tailscale license to the beginning of file,
+using year as the copyright year.
+
+It is intended for use with 'go generate', so it also runs a subcommand,
+which presumably creates the file.
+
+Sample usage:
+
+addlicense -year 2021 -file pull_strings.go stringer -type=pull
+`[1:])
+	os.Exit(2)
+}
+
+func main() {
+	flag.Usage = usage
+	flag.Parse()
+	if len(flag.Args()) == 0 {
+		flag.Usage()
+	}
+	cmd := exec.Command(flag.Arg(0), flag.Args()[1:]...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	err := cmd.Run()
+	check(err)
+	b, err := os.ReadFile(*file)
+	check(err)
+	f, err := os.OpenFile(*file, os.O_TRUNC|os.O_WRONLY, 0644)
+	check(err)
+	_, err = fmt.Fprintf(f, license, *year)
+	check(err)
+	_, err = f.Write(b)
+	check(err)
+	err = f.Close()
+	check(err)
+}
+
+func check(err error) {
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+}
+
+var license = `
+// Copyright (c) %d 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.
+
+`[1:]

cmd/cloner/cloner.go

@@ -246,7 +246,9 @@ func gen(buf *bytes.Buffer, imports map[string]struct{}, name string, typ *types
 					writef("\t\tdst.%s[k] = append([]%s{}, src.%s[k]...)", fname, n, fname)
 					writef("\t}")
 				} else if containsPointers(ft.Elem()) {
-					writef("\t\t" + `panic("TODO map value pointers")`)
+					writef("\tfor k, v := range src.%s {", fname)
+					writef("\t\tdst.%s[k] = v.Clone()", fname)
+					writef("\t}")
 				} else {
 					writef("\tfor k, v := range src.%s {", fname)
 					writef("\t\tdst.%s[k] = v", fname)

cmd/derper/derper.go

@@ -12,8 +12,6 @@ import (
 	"errors"
 	"expvar"
 	"flag"
-	"fmt"
-	"html"
 	"io"
 	"io/ioutil"
 	"log"
@@ -35,7 +33,6 @@ import (
 	"tailscale.com/tsweb"
 	"tailscale.com/types/key"
 	"tailscale.com/types/wgkey"
-	"tailscale.com/version"
 )
 
 var (
@@ -49,6 +46,7 @@ var (
 	meshPSKFile   = flag.String("mesh-psk-file", defaultMeshPSKFile(), "if non-empty, path to file containing the mesh pre-shared key file. It should contain some hex string; whitespace is trimmed.")
 	meshWith      = flag.String("mesh-with", "", "optional comma-separated list of hostnames to mesh with; the server's own hostname can be in the list")
 	bootstrapDNS  = flag.String("bootstrap-dns-names", "", "optional comma-separated list of hostnames to make available at /bootstrap-dns")
+	verifyClients = flag.Bool("verify-clients", false, "verify clients to this DERP server through a local tailscaled instance.")
 )
 
 type config struct {
@@ -60,7 +58,12 @@ func loadConfig() config {
 		return config{PrivateKey: mustNewKey()}
 	}
 	if *configPath == "" {
-		log.Fatalf("derper: -c <config path> not specified")
+		if os.Getuid() == 0 {
+			*configPath = "/var/lib/derper/derper.key"
+		} else {
+			log.Fatalf("derper: -c <config path> not specified")
+		}
+		log.Printf("no config path specified; using %s", *configPath)
 	}
 	b, err := ioutil.ReadFile(*configPath)
 	switch {
@@ -125,6 +128,7 @@ func main() {
 	letsEncrypt := tsweb.IsProd443(*addr)
 
 	s := derp.NewServer(key.Private(cfg.PrivateKey), log.Printf)
+	s.SetVerifyClient(*verifyClients)
 
 	if *meshPSKFile != "" {
 		b, err := ioutil.ReadFile(*meshPSKFile)
@@ -143,8 +147,7 @@ func main() {
 	}
 	expvar.Publish("derp", s.ExpVar())
 
-	// Create our own mux so we don't expose /debug/ stuff to the world.
-	mux := tsweb.NewMux(debugHandler(s))
+	mux := http.NewServeMux()
 	mux.Handle("/derp", derphttp.Handler(s))
 	go refreshBootstrapDNSLoop()
 	mux.HandleFunc("/bootstrap-dns", handleBootstrapDNS)
@@ -164,6 +167,18 @@ func main() {
 			io.WriteString(w, "<p>Debug info at <a href='/debug/'>/debug/</a>.</p>\n")
 		}
 	}))
+	debug := tsweb.Debugger(mux)
+	debug.KV("TLS hostname", *hostname)
+	debug.KV("Mesh key", s.HasMeshKey())
+	debug.Handle("check", "Consistency check", http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		err := s.ConsistencyCheck()
+		if err != nil {
+			http.Error(w, err.Error(), 500)
+		} else {
+			io.WriteString(w, "derp.Server ConsistencyCheck okay")
+		}
+	}))
+	debug.Handle("traffic", "Traffic check", http.HandlerFunc(s.ServeDebugTraffic))
 
 	if *runSTUN {
 		go serveSTUN()
@@ -217,39 +232,6 @@ func main() {
 	}
 }
 
-func debugHandler(s *derp.Server) http.Handler {
-	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.RequestURI == "/debug/check" {
-			err := s.ConsistencyCheck()
-			if err != nil {
-				http.Error(w, err.Error(), 500)
-			} else {
-				io.WriteString(w, "derp.Server ConsistencyCheck okay")
-			}
-			return
-		}
-		f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
-		f(`<html><body>
-<h1>DERP debug</h1>
-<ul>
-`)
-		f("<li><b>Hostname:</b> %v</li>\n", html.EscapeString(*hostname))
-		f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
-		f("<li><b>Mesh Key:</b> %v</li>\n", s.HasMeshKey())
-		f("<li><b>Version:</b> %v</li>\n", html.EscapeString(version.Long))
-
-		f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
-   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
-   <li><a href="/debug/pprof/">/debug/pprof/</a></li>
-   <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
-   <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
-   <li><a href="/debug/check">/debug/check</a> internal consistency check</li>
-<ul>
-</html>
-`)
-	})
-}
-
 func serveSTUN() {
 	pc, err := net.ListenPacket("udp", ":3478")
 	if err != nil {

cmd/derper/mesh.go

@@ -9,7 +9,9 @@ import (
 	"errors"
 	"fmt"
 	"log"
+	"net"
 	"strings"
+	"time"
 
 	"tailscale.com/derp"
 	"tailscale.com/derp/derphttp"
@@ -39,6 +41,34 @@ func startMeshWithHost(s *derp.Server, host string) error {
 		return err
 	}
 	c.MeshKey = s.MeshKey()
+
+	// For meshed peers within a region, connect via VPC addresses.
+	c.SetURLDialer(func(ctx context.Context, network, addr string) (net.Conn, error) {
+		host, port, err := net.SplitHostPort(addr)
+		if err != nil {
+			return nil, err
+		}
+		var d net.Dialer
+		var r net.Resolver
+		if port == "443" && strings.HasSuffix(host, ".tailscale.com") {
+			base := strings.TrimSuffix(host, ".tailscale.com")
+			subCtx, cancel := context.WithTimeout(ctx, 2*time.Second)
+			defer cancel()
+			vpcHost := base + "-vpc.tailscale.com"
+			ips, _ := r.LookupIP(subCtx, "ip", vpcHost)
+			if len(ips) > 0 {
+				vpcAddr := net.JoinHostPort(ips[0].String(), port)
+				c, err := d.DialContext(subCtx, network, vpcAddr)
+				if err == nil {
+					log.Printf("connected to %v (%v) instead of %v", vpcHost, ips[0], base)
+					return c, nil
+				}
+				log.Printf("failed to connect to %v (%v): %v; trying non-VPC route", vpcHost, ips[0], err)
+			}
+		}
+		return d.DialContext(ctx, network, addr)
+	})
+
 	add := func(k key.Public) { s.AddPacketForwarder(k, c) }
 	remove := func(k key.Public) { s.RemovePacketForwarder(k, c) }
 	go c.RunWatchConnectionLoop(context.Background(), s.PublicKey(), logf, add, remove)

cmd/derpprobe/derpprobe.go

@@ -0,0 +1,354 @@
+// 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.
+
+// The derpprobe binary probes derpers.
+package main // import "tailscale.com/cmd/derper/derpprobe"
+
+import (
+	"bytes"
+	"context"
+	crand "crypto/rand"
+	"encoding/json"
+	"flag"
+	"fmt"
+	"html"
+	"io"
+	"log"
+	"net/http"
+	"sort"
+	"sync"
+	"time"
+
+	"tailscale.com/derp"
+	"tailscale.com/derp/derphttp"
+	"tailscale.com/tailcfg"
+	"tailscale.com/types/key"
+)
+
+var (
+	derpMapURL = flag.String("derp-map", "https://login.tailscale.com/derpmap/default", "URL to DERP map (https:// or file://)")
+	listen     = flag.String("listen", ":8030", "HTTP listen address")
+)
+
+var (
+	mu            sync.Mutex
+	state         = map[nodePair]pairStatus{}
+	lastDERPMap   *tailcfg.DERPMap
+	lastDERPMapAt time.Time
+)
+
+func main() {
+	flag.Parse()
+	go probeLoop()
+	log.Fatal(http.ListenAndServe(*listen, http.HandlerFunc(serve)))
+}
+
+type overallStatus struct {
+	good, bad []string
+}
+
+func (st *overallStatus) addBadf(format string, a ...interface{}) {
+	st.bad = append(st.bad, fmt.Sprintf(format, a...))
+}
+
+func (st *overallStatus) addGoodf(format string, a ...interface{}) {
+	st.good = append(st.good, fmt.Sprintf(format, a...))
+}
+
+func getOverallStatus() (o overallStatus) {
+	mu.Lock()
+	defer mu.Unlock()
+	if lastDERPMap == nil {
+		o.addBadf("no DERP map")
+		return
+	}
+	now := time.Now()
+	if age := now.Sub(lastDERPMapAt); age > time.Minute {
+		o.addBadf("DERPMap hasn't been successfully refreshed in %v", age.Round(time.Second))
+	}
+	for _, reg := range sortedRegions(lastDERPMap) {
+		for _, from := range reg.Nodes {
+			for _, to := range reg.Nodes {
+				pair := nodePair{from.Name, to.Name}
+				st, ok := state[pair]
+				age := now.Sub(st.at).Round(time.Second)
+				switch {
+				case !ok:
+					o.addBadf("no state for %v", pair)
+				case st.err != nil:
+					o.addBadf("%v: %v", pair, st.err)
+				case age > 90*time.Second:
+					o.addBadf("%v: update is %v old", pair, age)
+				default:
+					o.addGoodf("%v: %v, %v ago", pair, st.latency.Round(time.Millisecond), age)
+				}
+			}
+		}
+	}
+	return
+}
+
+func serve(w http.ResponseWriter, r *http.Request) {
+	st := getOverallStatus()
+	summary := "All good"
+	if len(st.bad) > 0 {
+		w.WriteHeader(500)
+		summary = fmt.Sprintf("%d problems", len(st.bad))
+	}
+	io.WriteString(w, "<html><head><style>.bad { font-weight: bold; color: #700; }</style></head>\n")
+	fmt.Fprintf(w, "<body><h1>derp probe</h1>\n%s:<ul>", summary)
+	for _, s := range st.bad {
+		fmt.Fprintf(w, "<li class=bad>%s</li>\n", html.EscapeString(s))
+	}
+	for _, s := range st.good {
+		fmt.Fprintf(w, "<li>%s</li>\n", html.EscapeString(s))
+	}
+	io.WriteString(w, "</ul></body></html>\n")
+}
+
+func sortedRegions(dm *tailcfg.DERPMap) []*tailcfg.DERPRegion {
+	ret := make([]*tailcfg.DERPRegion, 0, len(dm.Regions))
+	for _, r := range dm.Regions {
+		ret = append(ret, r)
+	}
+	sort.Slice(ret, func(i, j int) bool { return ret[i].RegionID < ret[j].RegionID })
+	return ret
+}
+
+type nodePair struct {
+	from, to string // DERPNode.Name
+}
+
+func (p nodePair) String() string { return fmt.Sprintf("(%s→%s)", p.from, p.to) }
+
+type pairStatus struct {
+	err     error
+	latency time.Duration
+	at      time.Time
+}
+
+func setDERPMap(dm *tailcfg.DERPMap) {
+	mu.Lock()
+	defer mu.Unlock()
+	lastDERPMap = dm
+	lastDERPMapAt = time.Now()
+}
+
+func setState(p nodePair, latency time.Duration, err error) {
+	mu.Lock()
+	defer mu.Unlock()
+	st := pairStatus{
+		err:     err,
+		latency: latency,
+		at:      time.Now(),
+	}
+	state[p] = st
+	if err != nil {
+		log.Printf("%+v error: %v", p, err)
+	} else {
+		log.Printf("%+v: %v", p, latency.Round(time.Millisecond))
+	}
+}
+
+func probeLoop() {
+	ticker := time.NewTicker(15 * time.Second)
+	for {
+		err := probe()
+		if err != nil {
+			log.Printf("probe: %v", err)
+		}
+		<-ticker.C
+	}
+}
+
+func probe() error {
+	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	defer cancel()
+	dm, err := getDERPMap(ctx)
+	if err != nil {
+		return err
+	}
+
+	var wg sync.WaitGroup
+	wg.Add(len(dm.Regions))
+	for _, reg := range dm.Regions {
+		reg := reg
+		go func() {
+			defer wg.Done()
+			for _, from := range reg.Nodes {
+				for _, to := range reg.Nodes {
+					latency, err := probeNodePair(ctx, dm, from, to)
+					setState(nodePair{from.Name, to.Name}, latency, err)
+				}
+			}
+		}()
+	}
+
+	wg.Wait()
+	return ctx.Err()
+}
+
+func probeNodePair(ctx context.Context, dm *tailcfg.DERPMap, from, to *tailcfg.DERPNode) (latency time.Duration, err error) {
+	// The passed in context is a minute for the whole region. The
+	// idea is that each node pair in the region will be done
+	// serially and regularly in the future, reusing connections
+	// (at least in the happy path). For now they don't reuse
+	// connections and probe at most once every 15 seconds. We
+	// bound the duration of a single node pair within a region
+	// so one bad one can't starve others.
+	ctx, cancel := context.WithTimeout(ctx, 10*time.Second)
+	defer cancel()
+
+	fromc, err := newConn(ctx, dm, from)
+	if err != nil {
+		return 0, err
+	}
+	defer fromc.Close()
+	toc, err := newConn(ctx, dm, to)
+	if err != nil {
+		return 0, err
+	}
+	defer toc.Close()
+
+	// Wait a bit for from's node to hear about to existing on the
+	// other node in the region, in the case where the two nodes
+	// are different.
+	if from.Name != to.Name {
+		time.Sleep(100 * time.Millisecond) // pretty arbitrary
+	}
+
+	// Make a random packet
+	pkt := make([]byte, 8)
+	crand.Read(pkt)
+
+	t0 := time.Now()
+
+	// Send the random packet.
+	sendc := make(chan error, 1)
+	go func() {
+		sendc <- fromc.Send(toc.SelfPublicKey(), pkt)
+	}()
+	select {
+	case <-ctx.Done():
+		return 0, fmt.Errorf("timeout sending via %q: %w", from.Name, ctx.Err())
+	case err := <-sendc:
+		if err != nil {
+			return 0, fmt.Errorf("error sending via %q: %w", from.Name, err)
+		}
+	}
+
+	// Receive the random packet.
+	recvc := make(chan interface{}, 1) // either derp.ReceivedPacket or error
+	go func() {
+		for {
+			m, err := toc.Recv()
+			if err != nil {
+				recvc <- err
+				return
+			}
+			switch v := m.(type) {
+			case derp.ReceivedPacket:
+				recvc <- v
+			default:
+				log.Printf("%v: ignoring Recv frame type %T", to.Name, v)
+				// Loop.
+			}
+		}
+	}()
+	select {
+	case <-ctx.Done():
+		return 0, fmt.Errorf("timeout receiving from %q: %w", to.Name, ctx.Err())
+	case v := <-recvc:
+		if err, ok := v.(error); ok {
+			return 0, fmt.Errorf("error receiving from %q: %w", to.Name, err)
+		}
+		p := v.(derp.ReceivedPacket)
+		if p.Source != fromc.SelfPublicKey() {
+			return 0, fmt.Errorf("got data packet from unexpected source, %v", p.Source)
+		}
+		if !bytes.Equal(p.Data, pkt) {
+			return 0, fmt.Errorf("unexpected data packet %q", p.Data)
+		}
+	}
+	return time.Since(t0), nil
+}
+
+func newConn(ctx context.Context, dm *tailcfg.DERPMap, n *tailcfg.DERPNode) (*derphttp.Client, error) {
+	priv := key.NewPrivate()
+	dc := derphttp.NewRegionClient(priv, log.Printf, func() *tailcfg.DERPRegion {
+		rid := n.RegionID
+		return &tailcfg.DERPRegion{
+			RegionID:   rid,
+			RegionCode: fmt.Sprintf("%s-%s", dm.Regions[rid].RegionCode, n.Name),
+			RegionName: dm.Regions[rid].RegionName,
+			Nodes:      []*tailcfg.DERPNode{n},
+		}
+	})
+	dc.IsProber = true
+	err := dc.Connect(ctx)
+	if err != nil {
+		return nil, err
+	}
+	errc := make(chan error, 1)
+	go func() {
+		m, err := dc.Recv()
+		if err != nil {
+			errc <- err
+			return
+		}
+		switch m.(type) {
+		case derp.ServerInfoMessage:
+			errc <- nil
+		default:
+			errc <- fmt.Errorf("unexpected first message type %T", errc)
+		}
+	}()
+	select {
+	case err := <-errc:
+		if err != nil {
+			go dc.Close()
+			return nil, err
+		}
+	case <-ctx.Done():
+		go dc.Close()
+		return nil, fmt.Errorf("timeout waiting for ServerInfoMessage: %w", ctx.Err())
+	}
+	return dc, nil
+}
+
+var httpOrFileClient = &http.Client{Transport: httpOrFileTransport()}
+
+func httpOrFileTransport() http.RoundTripper {
+	tr := http.DefaultTransport.(*http.Transport).Clone()
+	tr.RegisterProtocol("file", http.NewFileTransport(http.Dir("/")))
+	return tr
+}
+
+func getDERPMap(ctx context.Context) (*tailcfg.DERPMap, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", *derpMapURL, nil)
+	if err != nil {
+		return nil, err
+	}
+	res, err := httpOrFileClient.Do(req)
+	if err != nil {
+		mu.Lock()
+		defer mu.Unlock()
+		if lastDERPMap != nil && time.Since(lastDERPMapAt) < 10*time.Minute {
+			// Assume that control is restarting and use
+			// the same one for a bit.
+			return lastDERPMap, nil
+		}
+		return nil, err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		return nil, fmt.Errorf("fetching %s: %s", *derpMapURL, res.Status)
+	}
+	dm := new(tailcfg.DERPMap)
+	if err := json.NewDecoder(res.Body).Decode(dm); err != nil {
+		return nil, fmt.Errorf("decoding %s JSON: %v", *derpMapURL, err)
+	}
+	setDERPMap(dm)
+	return dm, nil
+}

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/microproxy/microproxy.go

@@ -55,9 +55,13 @@ func main() {
 		log.Fatalf("Couldn't parse URL %q: %v", *goVarsURL, err)
 	}
 
-	mux := tsweb.NewMux(http.HandlerFunc(debugHandler))
+	mux := http.NewServeMux()
+	tsweb.Debugger(mux) // registers /debug/*
 	mux.Handle("/metrics", tsweb.Protected(proxy))
-	mux.Handle("/varz", tsweb.Protected(tsweb.StdHandler(&goVarsHandler{*goVarsURL}, log.Printf)))
+	mux.Handle("/varz", tsweb.Protected(tsweb.StdHandler(&goVarsHandler{*goVarsURL}, tsweb.HandlerOptions{
+		Quiet200s: true,
+		Logf:      log.Printf,
+	})))
 
 	ch := &certHolder{
 		hostname: *hostname,
@@ -167,23 +171,3 @@ func (c *certHolder) loadLocked() error {
 	c.loaded = time.Now()
 	return nil
 }
-
-// debugHandler serves a page with links to tsweb-managed debug URLs
-// at /debug/.
-func debugHandler(w http.ResponseWriter, r *http.Request) {
-	f := func(format string, args ...interface{}) { fmt.Fprintf(w, format, args...) }
-	f(`<html><body>
-<h1>microproxy debug</h1>
-<ul>
-`)
-	f("<li><b>Hostname:</b> %v</li>\n", *hostname)
-	f("<li><b>Uptime:</b> %v</li>\n", tsweb.Uptime())
-	f(`<li><a href="/debug/vars">/debug/vars</a> (Go)</li>
-   <li><a href="/debug/varz">/debug/varz</a> (Prometheus)</li>
-   <li><a href="/debug/pprof/">/debug/pprof/</a></li>
-   <li><a href="/debug/pprof/goroutine?debug=1">/debug/pprof/goroutine</a> (collapsed)</li>
-   <li><a href="/debug/pprof/goroutine?debug=2">/debug/pprof/goroutine</a> (full)</li>
-<ul>
-</html>
-`)
-}

cmd/mkpkg/main.go

@@ -21,6 +21,9 @@ import (
 // into a map of filePathOnDisk -> filePathInPackage.
 func parseFiles(s string) (map[string]string, error) {
 	ret := map[string]string{}
+	if len(s) == 0 {
+		return ret, nil
+	}
 	for _, f := range strings.Split(s, ",") {
 		fs := strings.Split(f, ":")
 		if len(fs) != 2 {

cmd/speedtest/speedtest.go

@@ -0,0 +1,121 @@
+// 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.
+
+// Program speedtest provides the speedtest command. The reason to keep it separate from
+// the normal tailscale cli is because it is not yet ready to go in the tailscale binary.
+// It will be included in the tailscale cli after it has been added to tailscaled.
+
+// Example usage for client command: go run cmd/speedtest -host 127.0.0.1:20333 -t 5s
+// This will connect to the server on 127.0.0.1:20333 and start a 5 second download speedtest.
+// Example usage for server command: go run cmd/speedtest -s -host :20333
+// This will start a speedtest server on port 20333.
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"net"
+	"os"
+	"strconv"
+	"text/tabwriter"
+	"time"
+
+	"github.com/peterbourgon/ff/v2/ffcli"
+	"tailscale.com/net/speedtest"
+)
+
+// Runs the speedtest command as a commandline program
+func main() {
+	args := os.Args[1:]
+	if err := speedtestCmd.Parse(args); err != nil {
+		fmt.Fprintln(os.Stderr, err.Error())
+		os.Exit(1)
+	}
+
+	err := speedtestCmd.Run(context.Background())
+	if errors.Is(err, flag.ErrHelp) {
+		fmt.Fprintln(os.Stderr, speedtestCmd.ShortUsage)
+		os.Exit(2)
+	}
+	if err != nil {
+		fmt.Fprintln(os.Stderr, err.Error())
+		os.Exit(1)
+	}
+}
+
+// speedtestCmd is the root command. It runs either the server or client depending on the
+// flags passed to it.
+var speedtestCmd = &ffcli.Command{
+	Name:       "speedtest",
+	ShortUsage: "speedtest [-host <host:port>] [-s] [-r] [-t <test duration>]",
+	ShortHelp:  "Run a speed test",
+	FlagSet: (func() *flag.FlagSet {
+		fs := flag.NewFlagSet("speedtest", flag.ExitOnError)
+		fs.StringVar(&speedtestArgs.host, "host", ":20333", "host:port pair to connect to or listen on")
+		fs.DurationVar(&speedtestArgs.testDuration, "t", speedtest.DefaultDuration, "duration of the speed test")
+		fs.BoolVar(&speedtestArgs.runServer, "s", false, "run a speedtest server")
+		fs.BoolVar(&speedtestArgs.reverse, "r", false, "run in reverse mode (server sends, client receives)")
+		return fs
+	})(),
+	Exec: runSpeedtest,
+}
+
+var speedtestArgs struct {
+	host         string
+	testDuration time.Duration
+	runServer    bool
+	reverse      bool
+}
+
+func runSpeedtest(ctx context.Context, args []string) error {
+
+	if _, _, err := net.SplitHostPort(speedtestArgs.host); err != nil {
+		var addrErr *net.AddrError
+		if errors.As(err, &addrErr) && addrErr.Err == "missing port in address" {
+			// if no port is provided, append the default port
+			speedtestArgs.host = net.JoinHostPort(speedtestArgs.host, strconv.Itoa(speedtest.DefaultPort))
+		}
+	}
+
+	if speedtestArgs.runServer {
+		listener, err := net.Listen("tcp", speedtestArgs.host)
+		if err != nil {
+			return err
+		}
+
+		fmt.Printf("listening on %v\n", listener.Addr())
+
+		return speedtest.Serve(listener)
+	}
+
+	// Ensure the duration is within the allowed range
+	if speedtestArgs.testDuration < speedtest.MinDuration || speedtestArgs.testDuration > speedtest.MaxDuration {
+		return fmt.Errorf("test duration must be within %v and %v", speedtest.MinDuration, speedtest.MaxDuration)
+	}
+
+	dir := speedtest.Download
+	if speedtestArgs.reverse {
+		dir = speedtest.Upload
+	}
+
+	fmt.Printf("Starting a %s test with %s\n", dir, speedtestArgs.host)
+	results, err := speedtest.RunClient(dir, speedtestArgs.testDuration, speedtestArgs.host)
+	if err != nil {
+		return err
+	}
+
+	w := tabwriter.NewWriter(os.Stdout, 12, 0, 0, ' ', tabwriter.TabIndent)
+	fmt.Println("Results:")
+	fmt.Fprintln(w, "Interval\t\tTransfer\t\tBandwidth\t\t")
+	for _, r := range results {
+		if r.Total {
+			fmt.Fprintln(w, "-------------------------------------------------------------------------")
+		}
+		fmt.Fprintf(w, "%.2f-%.2f\tsec\t%.4f\tMBits\t%.4f\tMbits/sec\t\n", r.IntervalStart.Seconds(), r.IntervalEnd.Seconds(), r.MegaBits(), r.MBitsPerSecond())
+	}
+	w.Flush()
+	return nil
+}

cmd/tailscale/cli/auth-redirect.html

@@ -0,0 +1,57 @@
+<html>
+<head>
+	<title>Redirecting...</title>
+	<style>
+		html,
+		body {
+			height: 100%;
+		}
+
+		html {
+			background-color: rgb(249, 247, 246);
+			font-family: ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji";
+			line-height: 1.5;
+			-webkit-text-size-adjust: 100%;
+			-webkit-font-smoothing: antialiased;
+			-moz-osx-font-smoothing: grayscale;
+		}
+
+		body {
+			display: flex;
+			flex-direction: column;
+			align-items: center;
+			justify-content: center;
+		}
+
+		.spinner {
+			margin-bottom: 2rem;
+			border: 4px rgba(112, 110, 109, 0.5) solid;
+			border-left-color: transparent;
+			border-radius: 9999px;
+			width: 4rem;
+			height: 4rem;
+			-webkit-animation: spin 700ms linear infinite;
+      animation: spin 800ms linear infinite;
+		}
+
+		.label {
+			color: rgb(112, 110, 109);
+			padding-left: 0.4rem;
+		}
+
+		@-webkit-keyframes spin {
+			to {
+				transform: rotate(360deg);
+			}
+		}
+
+		@keyframes spin {
+			to {
+				transform: rotate(360deg);
+			}
+		}
+	</style>
+</head> <body>
+	<div class="spinner"></div>
+	<div class="label">Redirecting...</div>
+</body>

cmd/tailscale/cli/cli_test.go

@@ -13,129 +13,96 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/google/go-cmp/cmp"
 	"inet.af/netaddr"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
-	"tailscale.com/types/logger"
+	"tailscale.com/types/persist"
 	"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
-		curUser  string // os.Getenv("USER") on the client side
-		goos     string // empty means "linux"
-		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{
-				ControlURL:  ipn.DefaultControlURL,
-				WantRunning: false,
-				Hostname:    "foo",
-				CorpDNS:     true,
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:  ipn.DefaultControlURL,
-					WantRunning: true,
-					CorpDNS:     true,
-				},
-				ControlURLSet:  true,
-				WantRunningSet: true,
-				CorpDNSSet:     true,
+				ControlURL:       ipn.DefaultControlURL,
+				WantRunning:      false,
+				Hostname:         "foo",
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
 			},
 			want: accidentalUpPrefix + " --accept-dns --hostname=foo",
 		},
 		{
-			name:    "hostname_changing_explicitly",
-			flagSet: f("hostname"),
+			name:  "hostname_changing_explicitly",
+			flags: []string{"--hostname=bar"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:  ipn.DefaultControlURL,
-				WantRunning: false,
-				Hostname:    "foo",
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:  ipn.DefaultControlURL,
-					WantRunning: true,
-					Hostname:    "bar",
-				},
-				ControlURLSet:  true,
-				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{
-				ControlURL:  ipn.DefaultControlURL,
-				WantRunning: false,
-				Hostname:    "foo",
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:  ipn.DefaultControlURL,
-					WantRunning: true,
-					Hostname:    "",
-				},
-				ControlURLSet:  true,
-				WantRunningSet: true,
-				HostnameSet:    true,
-			},
-			want: "",
-		},
-		{
-			name:    "empty_slice_equals_nil_slice",
-			flagSet: f("hostname"),
-			curPrefs: &ipn.Prefs{
-				ControlURL:      ipn.DefaultControlURL,
-				AdvertiseRoutes: []netaddr.IPPrefix{},
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:      ipn.DefaultControlURL,
-					AdvertiseRoutes: nil,
-				},
-				ControlURLSet: true,
+				ControlURL:       ipn.DefaultControlURL,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				AllowSingleHosts: true,
+				Hostname:         "foo",
 			},
 			want: "",
 		},
@@ -143,228 +110,174 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			// Issue 1725: "tailscale up --authkey=..." (or other non-empty flags) works from
 			// a fresh server's initial prefs.
 			name:     "up_with_default_prefs",
-			flagSet:  f("authkey"),
+			flags:    []string{"--authkey=foosdlkfjskdljf"},
 			curPrefs: ipn.NewPrefs(),
-			mp: &ipn.MaskedPrefs{
-				Prefs:          *defaultPrefsFromUpArgs(t, "linux"),
-				WantRunningSet: true,
-			},
-			want: "",
+			want:     "",
 		},
 		{
-			name:    "implicit_operator_change",
-			flagSet: f("hostname"),
+			name:  "implicit_operator_change",
+			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:   ipn.DefaultControlURL,
-				OperatorUser: "alice",
+				ControlURL:       ipn.DefaultControlURL,
+				OperatorUser:     "alice",
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
 			},
 			curUser: "eve",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-				},
-				ControlURLSet: true,
-			},
-			want: accidentalUpPrefix + " --hostname= --operator=alice",
+			want:    accidentalUpPrefix + " --hostname=foo --operator=alice",
 		},
 		{
-			name:    "implicit_operator_matches_shell_user",
-			flagSet: f("hostname"),
+			name:  "implicit_operator_matches_shell_user",
+			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:   ipn.DefaultControlURL,
-				OperatorUser: "alice",
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+				OperatorUser:     "alice",
 			},
 			curUser: "alice",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-				},
-				ControlURLSet: true,
-			},
-			want: "",
+			want:    "",
 		},
 		{
-			name:    "error_advertised_routes_exit_node_removed",
-			flagSet: f("advertise-routes"),
+			name:  "error_advertised_routes_exit_node_removed",
+			flags: []string{"--advertise-routes=10.0.42.0/24"},
 			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
+				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"),
 				},
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("10.0.42.0/24"),
-					},
-				},
-				AdvertiseRoutesSet: true,
-			},
 			want: accidentalUpPrefix + " --advertise-routes=10.0.42.0/24 --advertise-exit-node",
 		},
 		{
-			name:    "advertised_routes_exit_node_removed",
-			flagSet: f("advertise-routes", "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,
+				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"),
 				},
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("10.0.42.0/24"),
-					},
-				},
-				AdvertiseRoutesSet: true,
-			},
 			want: "",
 		},
 		{
-			name:    "advertised_routes_includes_the_0_routes", // but no --advertise-exit-node
-			flagSet: f("advertise-routes"),
+			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,
+				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"),
 				},
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("11.1.43.0/24"),
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-						netaddr.MustParseIPPrefix("::/0"),
-					},
-				},
-				AdvertiseRoutesSet: true,
+			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:    "advertised_routes_includes_only_one_0_route", // and no --advertise-exit-node
-			flagSet: f("advertise-routes"),
+			name:  "advertise_exit_node_over_existing_routes",
+			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
-				AdvertiseRoutes: []netaddr.IPPrefix{
-					netaddr.MustParseIPPrefix("10.0.42.0/24"),
-					netaddr.MustParseIPPrefix("0.0.0.0/0"),
-					netaddr.MustParseIPPrefix("::/0"),
-				},
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("11.1.43.0/24"),
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-					},
-				},
-				AdvertiseRoutesSet: true,
-			},
-			want: accidentalUpPrefix + " --advertise-routes=11.1.43.0/24,0.0.0.0/0 --advertise-exit-node",
-		},
-		{
-			name:    "advertise_exit_node", // Issue 1859
-			flagSet: f("advertise-exit-node"),
-			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-						netaddr.MustParseIPPrefix("::/0"),
-					},
-				},
-				// Note: without setting "AdvertiseRoutesSet", as
-				// updateMaskedPrefsFromUpFlag doesn't set that.
-			},
-			want: "",
-		},
-		{
-			name:    "advertise_exit_node_over_existing_routes",
-			flagSet: f("advertise-exit-node"),
-			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
 				AdvertiseRoutes: []netaddr.IPPrefix{
 					netaddr.MustParseIPPrefix("1.2.0.0/16"),
 				},
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-						netaddr.MustParseIPPrefix("::/0"),
-					},
-				},
-				// Note: without setting "AdvertiseRoutesSet", as
-				// updateMaskedPrefsFromUpFlag doesn't set that.
-			},
 			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
 		},
 		{
-			name:    "advertise_exit_node_over_existing_routes_and_exit_node",
-			flagSet: f("advertise-exit-node"),
+			name:  "advertise_exit_node_over_existing_routes_and_exit_node",
+			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
+				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"),
 				},
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-						netaddr.MustParseIPPrefix("::/0"),
-					},
-				},
-				// Note: without setting "AdvertiseRoutesSet", as
-				// updateMaskedPrefsFromUpFlag doesn't set that.
-			},
 			want: accidentalUpPrefix + " --advertise-exit-node --advertise-routes=1.2.0.0/16",
 		},
 		{
-			name:    "exit_node_clearing", // Issue 1777
-			flagSet: f("exit-node"),
+			name:  "exit_node_clearing", // Issue 1777
+			flags: []string{"--exit-node="},
 			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+
 				ExitNodeID: "fooID",
 			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					ExitNodeIP: netaddr.IP{},
-				},
-				ExitNodeIPSet: true,
-			},
 			want: "",
 		},
 		{
-			name:    "remove_all_implicit",
-			flagSet: f("force-reauth"),
+			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:          true,
+				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",
@@ -376,101 +289,142 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 				OperatorUser:  "alice",
 			},
 			curUser: "eve",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:  ipn.DefaultControlURL,
-					WantRunning: true,
-				},
-			},
-			want: accidentalUpPrefix + " --force-reauth --accept-routes --exit-node=100.64.5.6 --accept-dns --shields-up --advertise-tags=tag:foo,tag:bar --hostname=myhostname --unattended --advertise-routes=10.0.0.0/16 --netfilter-mode=nodivert --operator=alice",
+			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:    "remove_all_implicit_except_hostname",
-			flagSet: f("hostname"),
+			name:  "loggedout_is_implicit",
+			flags: []string{"--hostname=foo"},
 			curPrefs: &ipn.Prefs{
-				WantRunning:      true,
 				ControlURL:       ipn.DefaultControlURL,
-				RouteAll:         true,
-				AllowSingleHosts: false,
-				ExitNodeIP:       netaddr.MustParseIP("100.64.5.6"),
+				LoggedOut:        true,
+				AllowSingleHosts: true,
 				CorpDNS:          true,
-				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",
+				NetfilterMode:    preftype.NetfilterOn,
 			},
-			curUser: "eve",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL:  ipn.DefaultControlURL,
-					WantRunning: true,
-					Hostname:    "newhostname",
-				},
-				HostnameSet: true,
-			},
-			want: accidentalUpPrefix + " --hostname=newhostname --accept-routes --exit-node=100.64.5.6 --accept-dns --shields-up --advertise-tags=tag:foo,tag:bar --unattended --advertise-routes=10.0.0.0/16 --netfilter-mode=nodivert --operator=alice",
-		},
-		{
-			name:    "loggedout_is_implicit",
-			flagSet: f("advertise-exit-node"),
-			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
-				LoggedOut:  true,
-			},
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("0.0.0.0/0"),
-					},
-				},
-				AdvertiseRoutesSet: true,
-			},
-			// not an error. LoggedOut is implicit.
-			want: "",
+			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",
-			flagSet: f("advertise-exit-node"),
+			name:  "make_windows_exit_node",
+			flags: []string{"--advertise-exit-node"},
 			curPrefs: &ipn.Prefs{
-				ControlURL: ipn.DefaultControlURL,
-				NoSNAT:     true, // assume this no-op accidental pre-1.8 value
+				ControlURL:       ipn.DefaultControlURL,
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+
+				// And assume this no-op accidental pre-1.8 value:
+				NoSNAT: true,
 			},
 			goos: "windows",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					AdvertiseRoutes: []netaddr.IPPrefix{
-						netaddr.MustParseIPPrefix("192.168.0.0/16"),
-					},
+			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"),
 				},
-				AdvertiseRoutesSet: true,
+			},
+			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",
+		},
+		{
+			name:  "ignore_login_server_synonym",
+			flags: []string{"--login-server=https://controlplane.tailscale.com"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       "https://login.tailscale.com",
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
 			},
 			want: "", // not an error
 		},
 		{
-			name:    "ignore_netfilter_change_non_linux",
-			flagSet: f("accept-dns"),
+			name:  "ignore_login_server_synonym_on_other_change",
+			flags: []string{"--netfilter-mode=off"},
 			curPrefs: &ipn.Prefs{
-				ControlURL:    ipn.DefaultControlURL,
-				NetfilterMode: preftype.NetfilterNoDivert, // we never had this bug, but pretend it got set non-zero on Windows somehow
+				ControlURL:       "https://login.tailscale.com",
+				AllowSingleHosts: true,
+				CorpDNS:          false,
+				NetfilterMode:    preftype.NetfilterOn,
 			},
-			goos: "windows",
-			mp: &ipn.MaskedPrefs{
-				Prefs: ipn.Prefs{
-					ControlURL: ipn.DefaultControlURL,
-					CorpDNS:    false,
-				},
-				CorpDNSSet: true,
-			},
-			want: "", // not an error
+			want: accidentalUpPrefix + " --netfilter-mode=off --accept-dns=false",
 		},
 	}
 	for _, tt := range tests {
@@ -479,8 +433,20 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 			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, goos, tt.curUser); err != nil {
+			if err := checkForAccidentalSettingReverts(newPrefs, tt.curPrefs, upCheckEnv{
+				goos:          goos,
+				flagSet:       flagSet,
+				curExitNodeIP: tt.curExitNodeIP,
+			}); err != nil {
 				got = err.Error()
 			}
 			if strings.TrimSpace(got) != tt.want {
@@ -490,16 +456,6 @@ func TestCheckForAccidentalSettingReverts(t *testing.T) {
 	}
 }
 
-func defaultPrefsFromUpArgs(t testing.TB, goos string) *ipn.Prefs {
-	upArgs := upArgsFromOSArgs(goos)
-	prefs, err := prefsFromUpArgs(upArgs, logger.Discard, new(ipnstate.Status), "linux")
-	if err != nil {
-		t.Fatalf("defaultPrefsFromUpArgs: %v", err)
-	}
-	prefs.WantRunning = true
-	return prefs
-}
-
 func upArgsFromOSArgs(goos string, flagArgs ...string) (args upArgsT) {
 	fs := newUpFlagSet(goos, &args)
 	fs.Parse(flagArgs) // populates args
@@ -575,7 +531,7 @@ func TestPrefsFromUpArgs(t *testing.T) {
 			args: upArgsT{
 				exitNodeIP: "foo",
 			},
-			wantErr: `invalid IP address "foo" for --exit-node: unable to parse IP`,
+			wantErr: `invalid IP address "foo" for --exit-node: ParseIP("foo"): unable to parse IP`,
 		},
 		{
 			name: "error_exit_node_allow_lan_without_exit_node",
@@ -695,10 +651,17 @@ func TestPrefsFromUpArgs(t *testing.T) {
 }
 
 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 _, ok := flagForPref[prefName]; ok {
+		if prefHasFlag[prefName] {
 			continue
 		}
 		switch prefName {
@@ -716,3 +679,120 @@ func TestPrefFlagMapping(t *testing.T) {
 		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)
+			}
+		})
+	}
+}
+
+func TestUpdatePrefs(t *testing.T) {
+	tests := []struct {
+		name     string
+		flags    []string // argv to be parsed into env.flagSet and env.upArgs
+		curPrefs *ipn.Prefs
+		env      upCheckEnv // empty goos means "linux"
+
+		wantSimpleUp   bool
+		wantJustEditMP *ipn.MaskedPrefs
+		wantErrSubtr   string
+	}{
+		{
+			name:  "bare_up_means_up",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL:  ipn.DefaultControlURL,
+				WantRunning: false,
+				Hostname:    "foo",
+			},
+		},
+		{
+			name:  "just_up",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: ipn.DefaultControlURL,
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env: upCheckEnv{
+				backendState: "Stopped",
+			},
+			wantSimpleUp: true,
+		},
+		{
+			name:  "just_edit",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: ipn.DefaultControlURL,
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+		},
+		{
+			name:  "control_synonym",
+			flags: []string{},
+			curPrefs: &ipn.Prefs{
+				ControlURL: "https://login.tailscale.com",
+				Persist:    &persist.Persist{LoginName: "crawshaw.github"},
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+		},
+		{
+			name:  "change_login_server",
+			flags: []string{"--login-server=https://localhost:1000"},
+			curPrefs: &ipn.Prefs{
+				ControlURL:       "https://login.tailscale.com",
+				Persist:          &persist.Persist{LoginName: "crawshaw.github"},
+				AllowSingleHosts: true,
+				CorpDNS:          true,
+				NetfilterMode:    preftype.NetfilterOn,
+			},
+			env:            upCheckEnv{backendState: "Running"},
+			wantSimpleUp:   true,
+			wantJustEditMP: &ipn.MaskedPrefs{WantRunningSet: true},
+			wantErrSubtr:   "can't change --login-server without --force-reauth",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if tt.env.goos == "" {
+				tt.env.goos = "linux"
+			}
+			tt.env.flagSet = newUpFlagSet(tt.env.goos, &tt.env.upArgs)
+			tt.env.flagSet.Parse(tt.flags)
+
+			newPrefs, err := prefsFromUpArgs(tt.env.upArgs, t.Logf, new(ipnstate.Status), tt.env.goos)
+			if err != nil {
+				t.Fatal(err)
+			}
+			simpleUp, justEditMP, err := updatePrefs(newPrefs, tt.curPrefs, tt.env)
+			if err != nil {
+				if tt.wantErrSubtr != "" {
+					if !strings.Contains(err.Error(), tt.wantErrSubtr) {
+						t.Fatalf("want error %q, got: %v", tt.wantErrSubtr, err)
+					}
+					return
+				}
+				t.Fatal(err)
+			}
+			if simpleUp != tt.wantSimpleUp {
+				t.Fatalf("simpleUp=%v, want %v", simpleUp, tt.wantSimpleUp)
+			}
+			if justEditMP != nil {
+				justEditMP.Prefs = ipn.Prefs{} // uninteresting
+			}
+			if !reflect.DeepEqual(justEditMP, tt.wantJustEditMP) {
+				t.Fatalf("justEditMP: %v", cmp.Diff(justEditMP, tt.wantJustEditMP))
+			}
+		})
+	}
+}

cmd/tailscale/cli/debug.go

@@ -31,6 +31,7 @@ var debugCmd = &ffcli.Command{
 		fs.BoolVar(&debugArgs.goroutines, "daemon-goroutines", false, "If true, dump the tailscaled daemon's goroutines")
 		fs.BoolVar(&debugArgs.ipn, "ipn", false, "If true, subscribe to IPN notifications")
 		fs.BoolVar(&debugArgs.prefs, "prefs", false, "If true, dump active prefs")
+		fs.BoolVar(&debugArgs.derpMap, "derp", false, "If true, dump DERP map")
 		fs.BoolVar(&debugArgs.pretty, "pretty", false, "If true, pretty-print output (for --prefs)")
 		fs.BoolVar(&debugArgs.netMap, "netmap", true, "whether to include netmap in --ipn mode")
 		fs.BoolVar(&debugArgs.localCreds, "local-creds", false, "print how to connect to local tailscaled")
@@ -44,6 +45,7 @@ var debugArgs struct {
 	goroutines bool
 	ipn        bool
 	netMap     bool
+	derpMap    bool
 	file       string
 	prefs      bool
 	pretty     bool
@@ -87,6 +89,18 @@ func runDebug(ctx context.Context, args []string) error {
 		os.Stdout.Write(goroutines)
 		return nil
 	}
+	if debugArgs.derpMap {
+		dm, err := tailscale.CurrentDERPMap(ctx)
+		if err != nil {
+			return fmt.Errorf(
+				"failed to get local derp map, instead `curl %s/derpmap/default`: %w", ipn.DefaultControlURL, err,
+			)
+		}
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "\t")
+		enc.Encode(dm)
+		return nil
+	}
 	if debugArgs.ipn {
 		c, bc, ctx, cancel := connect(ctx)
 		defer cancel()

cmd/tailscale/cli/diag.go

@@ -0,0 +1,55 @@
+// 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 linux windows darwin
+
+package cli
+
+import (
+	"fmt"
+	"path/filepath"
+	"runtime"
+	"strings"
+
+	ps "github.com/mitchellh/go-ps"
+)
+
+// fixTailscaledConnectError is called when the local tailscaled has
+// been determined unreachable due to the provided origErr value. It
+// returns either the same error or a better one to help the user
+// understand why tailscaled isn't running for their platform.
+func fixTailscaledConnectError(origErr error) error {
+	procs, err := ps.Processes()
+	if err != nil {
+		return fmt.Errorf("failed to connect to local Tailscaled process and failed to enumerate processes while looking for it")
+	}
+	found := false
+	for _, proc := range procs {
+		base := filepath.Base(proc.Executable())
+		if base == "tailscaled" {
+			found = true
+			break
+		}
+		if runtime.GOOS == "darwin" && base == "IPNExtension" {
+			found = true
+			break
+		}
+		if runtime.GOOS == "windows" && strings.EqualFold(base, "tailscaled.exe") {
+			found = true
+			break
+		}
+	}
+	if !found {
+		switch runtime.GOOS {
+		case "windows":
+			return fmt.Errorf("failed to connect to local tailscaled process; is the Tailscale service running?")
+		case "darwin":
+			return fmt.Errorf("failed to connect to local Tailscale service; is Tailscale running?")
+		case "linux":
+			return fmt.Errorf("failed to connect to local tailscaled; it doesn't appear to be running (sudo systemctl start tailscaled ?)")
+		}
+		return fmt.Errorf("failed to connect to local tailscaled process; it doesn't appear to be running")
+	}
+	return fmt.Errorf("failed to connect to local tailscaled (which appears to be running). Got error: %w", origErr)
+}

cmd/tailscale/cli/diag_other.go

@@ -0,0 +1,16 @@
+// 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 !linux,!windows,!darwin
+
+package cli
+
+import "fmt"
+
+// The github.com/mitchellh/go-ps package doesn't work on all platforms,
+// so just don't diagnose connect failures.
+
+func fixTailscaledConnectError(origErr error) error {
+	return fmt.Errorf("failed to connect to local tailscaled process (is it running?); got: %w", origErr)
+}

cmd/tailscale/cli/file.go

@@ -74,7 +74,6 @@ func runCp(ctx context.Context, args []string) error {
 		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]
@@ -97,14 +96,12 @@ func runCp(ctx context.Context, args []string) error {
 		return err
 	}
 
-	peerAPIBase, lastSeen, isOffline, err := discoverPeerAPIBase(ctx, ip)
+	peerAPIBase, 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 {
@@ -182,29 +179,26 @@ func runCp(ctx context.Context, args []string) error {
 	return nil
 }
 
-func discoverPeerAPIBase(ctx context.Context, ipStr string) (base string, lastSeen time.Time, isOffline bool, err error) {
+func discoverPeerAPIBase(ctx context.Context, ipStr string) (base string, isOffline bool, err error) {
 	ip, err := netaddr.ParseIP(ipStr)
 	if err != nil {
-		return "", time.Time{}, false, err
+		return "", false, err
 	}
 	fts, err := tailscale.FileTargets(ctx)
 	if err != nil {
-		return "", time.Time{}, false, err
+		return "", false, err
 	}
 	for _, ft := range fts {
 		n := ft.Node
 		for _, a := range n.Addresses {
-			if a.IP != ip {
+			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 ft.PeerAPIURL, isOffline, nil
 		}
 	}
-	return "", time.Time{}, false, fileTargetErrorDetail(ctx, ip)
+	return "", false, fileTargetErrorDetail(ctx, ip)
 }
 
 // fileTargetErrorDetail returns a non-nil error saying why ip is an
@@ -274,8 +268,6 @@ func (r *slowReader) Read(p []byte) (n int, err error) {
 	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")
@@ -301,7 +293,7 @@ func runCpTargets(ctx context.Context, args []string) error {
 		if detail != "" {
 			detail = "\t" + detail
 		}
-		fmt.Printf("%s\t%s%s\n", n.Addresses[0].IP, n.ComputedName, detail)
+		fmt.Printf("%s\t%s%s\n", n.Addresses[0].IP(), n.ComputedName, detail)
 	}
 	return nil
 }

cmd/tailscale/cli/netcheck.go

@@ -9,14 +9,18 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
+	"io/ioutil"
 	"log"
+	"net/http"
 	"os"
 	"sort"
 	"strings"
 	"time"
 
 	"github.com/peterbourgon/ff/v2/ffcli"
-	"tailscale.com/derp/derpmap"
+	"tailscale.com/client/tailscale"
+	"tailscale.com/ipn"
 	"tailscale.com/net/netcheck"
 	"tailscale.com/net/portmapper"
 	"tailscale.com/tailcfg"
@@ -46,7 +50,7 @@ var netcheckArgs struct {
 func runNetcheck(ctx context.Context, args []string) error {
 	c := &netcheck.Client{
 		UDPBindAddr: os.Getenv("TS_DEBUG_NETCHECK_UDP_BIND"),
-		PortMapper:  portmapper.NewClient(logger.WithPrefix(log.Printf, "portmap: ")),
+		PortMapper:  portmapper.NewClient(logger.WithPrefix(log.Printf, "portmap: "), nil),
 	}
 	if netcheckArgs.verbose {
 		c.Logf = logger.WithPrefix(log.Printf, "netcheck: ")
@@ -59,7 +63,13 @@ func runNetcheck(ctx context.Context, args []string) error {
 		fmt.Fprintln(os.Stderr, "# Warning: this JSON format is not yet considered a stable interface")
 	}
 
-	dm := derpmap.Prod()
+	dm, err := tailscale.CurrentDERPMap(ctx)
+	if err != nil {
+		dm, err = prodDERPMap(ctx, http.DefaultClient)
+		if err != nil {
+			return err
+		}
+	}
 	for {
 		t0 := time.Now()
 		report, err := c.GetReport(ctx, dm)
@@ -176,3 +186,27 @@ func portMapping(r *netcheck.Report) string {
 	}
 	return strings.Join(got, ", ")
 }
+
+func prodDERPMap(ctx context.Context, httpc *http.Client) (*tailcfg.DERPMap, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", ipn.DefaultControlURL+"/derpmap/default", nil)
+	if err != nil {
+		return nil, fmt.Errorf("create prodDERPMap request: %w", err)
+	}
+	res, err := httpc.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap failed: %w", err)
+	}
+	defer res.Body.Close()
+	b, err := ioutil.ReadAll(io.LimitReader(res.Body, 1<<20))
+	if err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap failed: %w", err)
+	}
+	if res.StatusCode != 200 {
+		return nil, fmt.Errorf("fetch prodDERPMap: %v: %s", res.Status, b)
+	}
+	var derpMap tailcfg.DERPMap
+	if err = json.Unmarshal(b, &derpMap); err != nil {
+		return nil, fmt.Errorf("fetch prodDERPMap: %w", err)
+	}
+	return &derpMap, nil
+}

cmd/tailscale/cli/ping.go

@@ -139,6 +139,9 @@ 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/status.go

@@ -23,6 +23,7 @@ import (
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/net/interfaces"
+	"tailscale.com/tstime/mono"
 	"tailscale.com/util/dnsname"
 )
 
@@ -57,7 +58,7 @@ var statusArgs struct {
 func runStatus(ctx context.Context, args []string) error {
 	st, err := tailscale.Status(ctx)
 	if err != nil {
-		return err
+		return fixTailscaledConnectError(err)
 	}
 	if statusArgs.json {
 		if statusArgs.active {
@@ -193,7 +194,7 @@ func runStatus(ctx context.Context, args []string) error {
 //
 // TODO: have the server report this bool instead.
 func peerActive(ps *ipnstate.PeerStatus) bool {
-	return !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
+	return !ps.LastWrite.IsZero() && mono.Since(ps.LastWrite) < 2*time.Minute
 }
 
 func dnsOrQuoteHostname(st *ipnstate.Status, ps *ipnstate.PeerStatus) string {

cmd/tailscale/cli/up.go

@@ -13,7 +13,6 @@ import (
 	"reflect"
 	"runtime"
 	"sort"
-	"strconv"
 	"strings"
 	"sync"
 
@@ -52,7 +51,14 @@ flag is also used.
 	Exec:    runUp,
 }
 
-var upFlagSet = newUpFlagSet(runtime.GOOS, &upArgs)
+func effectiveGOOS() string {
+	if v := os.Getenv("TS_DEBUG_UP_FLAG_GOOS"); v != "" {
+		return v
+	}
+	return runtime.GOOS
+}
+
+var upFlagSet = newUpFlagSet(effectiveGOOS(), &upArgs)
 
 func newUpFlagSet(goos string, upArgs *upArgsT) *flag.FlagSet {
 	upf := flag.NewFlagSet("up", flag.ExitOnError)
@@ -64,13 +70,13 @@ func newUpFlagSet(goos string, upArgs *upArgsT) *flag.FlagSet {
 	upf.BoolVar(&upArgs.acceptRoutes, "accept-routes", false, "accept routes advertised by other Tailscale nodes")
 	upf.BoolVar(&upArgs.acceptDNS, "accept-dns", true, "accept DNS configuration from the admin panel")
 	upf.BoolVar(&upArgs.singleRoutes, "host-routes", true, "install host routes to other Tailscale nodes")
-	upf.StringVar(&upArgs.exitNodeIP, "exit-node", "", "Tailscale IP of the exit node for internet traffic")
+	upf.StringVar(&upArgs.exitNodeIP, "exit-node", "", "Tailscale IP of the exit node for internet traffic, or empty string to not use an exit node")
 	upf.BoolVar(&upArgs.exitNodeAllowLANAccess, "exit-node-allow-lan-access", false, "Allow direct access to the local network when routing traffic via an exit node")
 	upf.BoolVar(&upArgs.shieldsUp, "shields-up", false, "don't allow incoming connections")
 	upf.StringVar(&upArgs.advertiseTags, "advertise-tags", "", "comma-separated ACL tags to request; each must start with \"tag:\" (e.g. \"tag:eng,tag:montreal,tag:ssh\")")
 	upf.StringVar(&upArgs.authKey, "authkey", "", "node authorization key")
 	upf.StringVar(&upArgs.hostname, "hostname", "", "hostname to use instead of the one provided by the OS")
-	upf.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.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\") or empty string to not advertise routes")
 	upf.BoolVar(&upArgs.advertiseDefaultRoute, "advertise-exit-node", false, "offer to be an exit node for internet traffic for the tailnet")
 	if safesocket.GOOSUsesPeerCreds(goos) {
 		upf.StringVar(&upArgs.opUser, "operator", "", "Unix username to allow to operate on tailscaled without sudo")
@@ -165,10 +171,10 @@ func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goo
 		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
@@ -231,7 +237,9 @@ func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goo
 			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:
 			return nil, fmt.Errorf("invalid value --netfilter-mode=%q", upArgs.netfilterMode)
 		}
@@ -239,6 +247,53 @@ func prefsFromUpArgs(upArgs upArgsT, warnf logger.Logf, st *ipnstate.Status, goo
 	return prefs, nil
 }
 
+// updatePrefs updates prefs based on curPrefs
+//
+// It returns a non-nil justEditMP if we're already running and none of
+// the flags require a restart, so we can just do an EditPrefs call and
+// change the prefs at runtime (e.g. changing hostname, changing
+// advertised tags, routes, etc).
+//
+// It returns simpleUp if we're running a simple "tailscale up" to
+// transition to running from a previously-logged-in but down state,
+// without changing any settings.
+func updatePrefs(prefs, curPrefs *ipn.Prefs, env upCheckEnv) (simpleUp bool, justEditMP *ipn.MaskedPrefs, err error) {
+	if !env.upArgs.reset {
+		applyImplicitPrefs(prefs, curPrefs, env.user)
+
+		if err := checkForAccidentalSettingReverts(prefs, curPrefs, env); err != nil {
+			return false, nil, err
+		}
+	}
+
+	controlURLChanged := curPrefs.ControlURL != prefs.ControlURL &&
+		!(ipn.IsLoginServerSynonym(curPrefs.ControlURL) && ipn.IsLoginServerSynonym(prefs.ControlURL))
+	if controlURLChanged && env.backendState == ipn.Running.String() && !env.upArgs.forceReauth {
+		return false, nil, fmt.Errorf("can't change --login-server without --force-reauth")
+	}
+
+	simpleUp = env.flagSet.NFlag() == 0 &&
+		curPrefs.Persist != nil &&
+		curPrefs.Persist.LoginName != "" &&
+		env.backendState != ipn.NeedsLogin.String()
+
+	justEdit := env.backendState == ipn.Running.String() &&
+		!env.upArgs.forceReauth &&
+		!env.upArgs.reset &&
+		env.upArgs.authKey == "" &&
+		!controlURLChanged
+	if justEdit {
+		justEditMP = new(ipn.MaskedPrefs)
+		justEditMP.WantRunningSet = true
+		justEditMP.Prefs = *prefs
+		env.flagSet.Visit(func(f *flag.Flag) {
+			updateMaskedPrefsFromUpFlag(justEditMP, f.Name)
+		})
+	}
+
+	return simpleUp, justEditMP, nil
+}
+
 func runUp(ctx context.Context, args []string) error {
 	if len(args) > 0 {
 		fatalf("too many non-flag arguments: %q", args)
@@ -246,7 +301,7 @@ func runUp(ctx context.Context, args []string) error {
 
 	st, err := tailscale.Status(ctx)
 	if err != nil {
-		fatalf("can't fetch status from tailscaled: %v", err)
+		return fixTailscaledConnectError(err)
 	}
 	origAuthURL := st.AuthURL
 
@@ -267,7 +322,7 @@ func runUp(ctx context.Context, args []string) error {
 	}
 
 	if distro.Get() == distro.Synology {
-		notSupported := "not yet supported on Synology; see https://github.com/tailscale/tailscale/issues/451"
+		notSupported := "not supported on Synology; see https://github.com/tailscale/tailscale/issues/1995"
 		if upArgs.acceptRoutes {
 			return errors.New("--accept-routes is " + notSupported)
 		}
@@ -279,7 +334,7 @@ func runUp(ctx context.Context, args []string) error {
 		}
 	}
 
-	prefs, err := prefsFromUpArgs(upArgs, warnf, st, runtime.GOOS)
+	prefs, err := prefsFromUpArgs(upArgs, warnf, st, effectiveGOOS())
 	if err != nil {
 		fatalf("%s", err)
 	}
@@ -295,48 +350,23 @@ 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, runtime.GOOS, os.Getenv("USER")); err != nil {
-			fatalf("%s", err)
-		}
+	env := upCheckEnv{
+		goos:          effectiveGOOS(),
+		user:          os.Getenv("USER"),
+		flagSet:       upFlagSet,
+		upArgs:        upArgs,
+		backendState:  st.BackendState,
+		curExitNodeIP: exitNodeIP(prefs, st),
 	}
-
-	controlURLChanged := curPrefs.ControlURL != prefs.ControlURL
-	if controlURLChanged && st.BackendState == ipn.Running.String() && !upArgs.forceReauth {
-		fatalf("can't change --login-server without --force-reauth")
+	simpleUp, justEditMP, err := updatePrefs(prefs, curPrefs, env)
+	if err != nil {
+		fatalf("%s", err)
 	}
-
-	// 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
-	// advertised tags, routes, etc)
-	justEdit := st.BackendState == ipn.Running.String() &&
-		!upArgs.forceReauth &&
-		!upArgs.reset &&
-		upArgs.authKey == "" &&
-		!controlURLChanged
-	if justEdit {
-		_, err := tailscale.EditPrefs(ctx, mp)
+	if justEditMP != nil {
+		_, err := tailscale.EditPrefs(ctx, justEditMP)
 		return err
 	}
 
-	// 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 != "" &&
-		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.
 	c, bc, pumpCtx, cancel := connect(ctx)
@@ -361,7 +391,7 @@ func runUp(ctx context.Context, args []string) error {
 		if n.ErrMessage != nil {
 			msg := *n.ErrMessage
 			if msg == ipn.ErrMsgPermissionDenied {
-				switch runtime.GOOS {
+				switch effectiveGOOS() {
 				case "windows":
 					msg += " (Tailscale service in use by other user?)"
 				default:
@@ -377,7 +407,7 @@ func runUp(ctx context.Context, args []string) error {
 				startLoginInteractive()
 			case ipn.NeedsMachineAuth:
 				printed = true
-				fmt.Fprintf(os.Stderr, "\nTo authorize your machine, visit (as admin):\n\n\t%s/admin/machines\n\n", upArgs.server)
+				fmt.Fprintf(os.Stderr, "\nTo authorize your machine, visit (as admin):\n\n\t%s\n\n", prefs.AdminPageURL())
 			case ipn.Starting, ipn.Running:
 				// Done full authentication process
 				if printed {
@@ -435,7 +465,7 @@ func runUp(ctx context.Context, args []string) error {
 		// Windows service (~tailscaled) is the one that computes the
 		// StateKey based on the connection identity. So for now, just
 		// do as the Windows GUI's always done:
-		if runtime.GOOS == "windows" {
+		if effectiveGOOS() == "windows" {
 			// The Windows service will set this as needed based
 			// on our connection's identity.
 			opts.StateKey = ""
@@ -464,14 +494,20 @@ func runUp(ctx context.Context, args []string) error {
 }
 
 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")
@@ -479,7 +515,6 @@ func init() {
 	addPrefFlagMapping("netfilter-mode", "NetfilterMode")
 	addPrefFlagMapping("shields-up", "ShieldsUp")
 	addPrefFlagMapping("snat-subnet-routes", "NoSNAT")
-	addPrefFlagMapping("exit-node", "ExitNodeIP", "ExitNodeID")
 	addPrefFlagMapping("exit-node-allow-lan-access", "ExitNodeAllowLANAccess")
 	addPrefFlagMapping("unattended", "ForceDaemon")
 	addPrefFlagMapping("operator", "OperatorUser")
@@ -489,8 +524,6 @@ 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))
@@ -498,21 +531,27 @@ 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(fmt.Sprintf("internal error: unhandled flag %q", flagName))
-	}
+	panic(fmt.Sprintf("internal error: unhandled flag %q", flagName))
 }
 
 const accidentalUpPrefix = "Error: changing settings via 'tailscale up' requires mentioning all\n" +
@@ -521,9 +560,20 @@ const accidentalUpPrefix = "Error: changing settings via 'tailscale up' requires
 	"all non-default settings:\n\n" +
 	"\ttailscale up"
 
-// checkForAccidentalSettingReverts checks for people running
-// "tailscale up" with a subset of the flags they originally ran it
-// with.
+// upCheckEnv are extra parameters describing the environment as
+// needed by checkForAccidentalSettingReverts and friends.
+type upCheckEnv struct {
+	goos          string
+	user          string
+	flagSet       *flag.FlagSet
+	upArgs        upArgsT
+	backendState  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
@@ -535,178 +585,183 @@ const accidentalUpPrefix = "Error: changing settings via 'tailscale up' requires
 //
 // 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, goos, curUser string) error {
-	if len(flagSet) == 0 {
-		// A bare "tailscale up" is a special case to just
-		// mean bringing the network up without any changes.
-		return nil
-	}
+func checkForAccidentalSettingReverts(newPrefs, curPrefs *ipn.Prefs, env upCheckEnv) error {
 	if curPrefs.ControlURL == "" {
 		// Don't validate things on initial "up" before a control URL has been set.
 		return nil
 	}
-	curWithExplicitEdits := curPrefs.Clone()
-	curWithExplicitEdits.ApplyEdits(mp)
 
-	prefType := reflect.TypeOf(ipn.Prefs{})
+	flagIsSet := map[string]bool{}
+	env.flagSet.Visit(func(f *flag.Flag) {
+		flagIsSet[f.Name] = true
+	})
 
-	// 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()
+	if len(flagIsSet) == 0 {
+		// A bare "tailscale up" is a special case to just
+		// mean bringing the network up without any changes.
+		return nil
+	}
+
+	// flagsCur is what flags we'd need to use to keep the exact
+	// settings as-is.
+	flagsCur := prefsToFlags(env, curPrefs)
+	flagsNew := prefsToFlags(env, newPrefs)
 
 	var missing []string
-	flagExplicitValue := map[string]interface{}{} // e.g. "accept-dns" => true (from flagSet)
-	for i := 0; i < prefType.NumField(); i++ {
-		prefName := prefType.Field(i).Name
-		// Persist is a legacy field used for storing keys, which
-		// probably should never have been part of Prefs. It's
-		// likely to migrate elsewhere eventually.
-		if prefName == "Persist" {
+	for flagName := range flagsCur {
+		valCur, valNew := flagsCur[flagName], flagsNew[flagName]
+		if flagIsSet[flagName] {
 			continue
 		}
-		// LoggedOut is a preference, but running the "up" command
-		// always implies that the user now prefers LoggedOut->false.
-		if prefName == "LoggedOut" {
+		if reflect.DeepEqual(valCur, valNew) {
 			continue
 		}
-		flagName, hasFlag := flagForPref[prefName]
-
-		// Special case for advertise-exit-node; which is a
-		// flag but doesn't have a corresponding pref.  The
-		// flag augments advertise-routes, so we have to infer
-		// the imaginary pref's current value from the routes.
-		if prefName == "AdvertiseRoutes" &&
-			hasExitNodeRoutes(curPrefs.AdvertiseRoutes) &&
-			!hasExitNodeRoutes(curWithExplicitEdits.AdvertiseRoutes) &&
-			!flagSet["advertise-exit-node"] {
-			missing = append(missing, "--advertise-exit-node")
-		}
-
-		if hasFlag && flagSet[flagName] {
-			flagExplicitValue[flagName] = ev.Field(i).Interface()
+		if flagName == "login-server" && ipn.IsLoginServerSynonym(valCur) && ipn.IsLoginServerSynonym(valNew) {
 			continue
 		}
-
-		if prefName == "AdvertiseRoutes" &&
-			(len(curPrefs.AdvertiseRoutes) == 0 ||
-				hasExitNodeRoutes(curPrefs.AdvertiseRoutes) && len(curPrefs.AdvertiseRoutes) == 2) &&
-			hasExitNodeRoutes(mp.Prefs.AdvertiseRoutes) &&
-			len(mp.Prefs.AdvertiseRoutes) == 2 &&
-			flagSet["advertise-exit-node"] {
-			continue
-		}
-
-		// Get explicit value and implicit value
-		ex, im := ev.Field(i), iv.Field(i)
-		switch ex.Kind() {
-		case reflect.String, reflect.Slice:
-			if ex.Kind() == reflect.Slice && ex.Len() == 0 && im.Len() == 0 {
-				// Treat nil and non-nil empty slices as equivalent.
-				continue
-			}
-		}
-		exi, imi := ex.Interface(), im.Interface()
-
-		if reflect.DeepEqual(exi, imi) {
-			continue
-		}
-		switch flagName {
-		case "operator":
-			if imi == "" && exi == curUser {
-				// Don't require setting operator if the current user matches
-				// the configured operator.
-				continue
-			}
-		case "snat-subnet-routes", "netfilter-mode":
-			if goos != "linux" {
-				// Issue 1833: we used to accidentally set the NoSNAT
-				// pref for non-Linux nodes. It only affects Linux, so
-				// ignore it if it changes. Likewise, ignore
-				// Linux-only netfilter-mode on non-Linux.
-				continue
-			}
-		}
-		switch flagName {
-		case "":
-			return 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 prefName == "ExitNodeIP" {
-				missing = append(missing, fmtFlagValueArg("exit-node", fmtSettingVal(exi)))
-			}
-		case "advertise-routes":
-			routes := withoutExitNodes(exi.([]netaddr.IPPrefix))
-			missing = append(missing, fmtFlagValueArg("advertise-routes", fmtSettingVal(routes)))
-		default:
-			missing = append(missing, fmtFlagValueArg(flagName, fmtSettingVal(exi)))
-		}
+		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
+	env.flagSet.Visit(func(f *flag.Flag) {
+		type isBool interface {
+			IsBoolFlag() bool
+		}
+		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)
+			}
+		} else {
+			explicit = append(explicit, fmtFlagValueArg(f.Name, f.Value.String()))
+		}
+	})
+
 	var sb strings.Builder
 	sb.WriteString(accidentalUpPrefix)
 
-	var flagSetSorted []string
-	for f := range flagSet {
-		flagSetSorted = append(flagSetSorted, f)
-	}
-	sort.Strings(flagSetSorted)
-	for _, flagName := range flagSetSorted {
-		if ev, ok := flagExplicitValue[flagName]; ok {
-			fmt.Fprintf(&sb, " %s", fmtFlagValueArg(flagName, fmtSettingVal(ev)))
-		} else {
-			fmt.Fprintf(&sb, " --%s", flagName)
-		}
-	}
-	for _, a := range missing {
+	for _, a := range append(explicit, missing...) {
 		fmt.Fprintf(&sb, " %s", a)
 	}
 	sb.WriteString("\n\n")
 	return errors.New(sb.String())
 }
 
-func fmtFlagValueArg(flagName, val string) string {
-	if val == "true" {
-		// TODO: check flagName's type to see if its Pref is of type bool
+// 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
+	}
+}
+
+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(val))
-}
-
-func fmtSettingVal(v interface{}) string {
-	switch v := v.(type) {
-	case bool:
-		return strconv.FormatBool(v)
-	case string:
-		return v
-	case preftype.NetfilterMode:
-		return v.String()
-	case []string:
-		return strings.Join(v, ",")
-	case []netaddr.IPPrefix:
-		var sb strings.Builder
-		for i, r := range v {
-			if i > 0 {
-				sb.WriteByte(',')
-			}
-			sb.WriteString(r.String())
-		}
-		return sb.String()
-	}
-	return fmt.Sprint(v)
+	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() {
+		if r.Bits() == 0 {
+			if r.IP().Is4() {
 				v4 = true
-			} else if r.IP.Is6() {
+			} else if r.IP().Is6() {
 				v6 = true
 			}
 		}
@@ -723,9 +778,33 @@ func withoutExitNodes(rr []netaddr.IPPrefix) []netaddr.IPPrefix {
 	}
 	var out []netaddr.IPPrefix
 	for _, r := range rr {
-		if r.Bits > 0 {
+		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

@@ -9,12 +9,16 @@ import (
 	"context"
 	_ "embed"
 	"encoding/json"
+	"encoding/xml"
 	"flag"
 	"fmt"
 	"html/template"
+	"io/ioutil"
 	"log"
+	"net"
 	"net/http"
 	"net/http/cgi"
+	"net/url"
 	"os/exec"
 	"runtime"
 	"strings"
@@ -24,6 +28,7 @@ import (
 	"tailscale.com/ipn"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/preftype"
+	"tailscale.com/util/groupmember"
 	"tailscale.com/version/distro"
 )
 
@@ -33,6 +38,9 @@ var webHTML string
 //go:embed web.css
 var webCSS string
 
+//go:embed auth-redirect.html
+var authenticationRedirectHTML string
+
 var tmpl *template.Template
 
 func init() {
@@ -53,6 +61,14 @@ var webCmd = &ffcli.Command{
 	ShortUsage: "web [flags]",
 	ShortHelp:  "Run a web server for controlling Tailscale",
 
+	LongHelp: strings.TrimSpace(`
+"tailscale web" runs a webserver for controlling the Tailscale daemon.
+
+It's primarily intended for use on Synology, QNAP, and other
+NAS devices where a web interface is the natural place to control
+Tailscale, as opposed to a CLI or a native app.
+`),
+
 	FlagSet: (func() *flag.FlagSet {
 		webf := flag.NewFlagSet("web", flag.ExitOnError)
 		webf.StringVar(&webArgs.listen, "listen", "localhost:8088", "listen address; use port 0 for automatic")
@@ -79,26 +95,128 @@ func runWeb(ctx context.Context, args []string) error {
 		}
 		return nil
 	}
+
+	log.Printf("web server running on: %s", urlOfListenAddr(webArgs.listen))
 	return http.ListenAndServe(webArgs.listen, http.HandlerFunc(webHandler))
 }
 
-func auth() (string, error) {
-	if distro.Get() == distro.Synology {
-		cmd := exec.Command("/usr/syno/synoman/webman/modules/authenticate.cgi")
-		out, err := cmd.CombinedOutput()
-		if err != nil {
-			return "", fmt.Errorf("auth: %v: %s", err, out)
-		}
-		return string(out), nil
+// urlOfListenAddr parses a given listen address into a formatted URL
+func urlOfListenAddr(addr string) string {
+	host, port, _ := net.SplitHostPort(addr)
+	if host == "" {
+		host = "127.0.0.1"
 	}
+	return fmt.Sprintf("http://%s", net.JoinHostPort(host, port))
+}
 
+// authorize returns the name of the user accessing the web UI after verifying
+// whether the user has access to the web UI. The function will write the
+// error to the provided http.ResponseWriter.
+// Note: This is different from a tailscale user, and is typically the local
+// user on the node.
+func authorize(w http.ResponseWriter, r *http.Request) (string, error) {
+	switch distro.Get() {
+	case distro.Synology:
+		user, err := synoAuthn()
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusUnauthorized)
+			return "", err
+		}
+		if err := authorizeSynology(user); err != nil {
+			http.Error(w, err.Error(), http.StatusForbidden)
+			return "", err
+		}
+		return user, nil
+	case distro.QNAP:
+		user, resp, err := qnapAuthn(r)
+		if err != nil {
+			http.Error(w, err.Error(), http.StatusUnauthorized)
+			return "", err
+		}
+		if resp.IsAdmin == 0 {
+			http.Error(w, err.Error(), http.StatusForbidden)
+			return "", err
+		}
+		return user, nil
+	}
 	return "", nil
 }
 
-func synoTokenRedirect(w http.ResponseWriter, r *http.Request) bool {
-	if distro.Get() != distro.Synology {
-		return false
+// authorizeSynology checks whether the provided user has access to the web UI
+// by consulting the membership of the "administrators" group.
+func authorizeSynology(name string) error {
+	yes, err := groupmember.IsMemberOfGroup("administrators", name)
+	if err != nil {
+		return err
 	}
+	if !yes {
+		return fmt.Errorf("not a member of administrators group")
+	}
+	return nil
+}
+
+type qnapAuthResponse struct {
+	AuthPassed int    `xml:"authPassed"`
+	IsAdmin    int    `xml:"isAdmin"`
+	AuthSID    string `xml:"authSid"`
+	ErrorValue int    `xml:"errorValue"`
+}
+
+func qnapAuthn(r *http.Request) (string, *qnapAuthResponse, error) {
+	user, err := r.Cookie("NAS_USER")
+	if err != nil {
+		return "", nil, err
+	}
+	token, err := r.Cookie("qtoken")
+	if err != nil {
+		return "", nil, err
+	}
+	query := url.Values{
+		"qtoken": []string{token.Value},
+		"user":   []string{user.Value},
+	}
+	u := url.URL{
+		Scheme:   r.URL.Scheme,
+		Host:     r.URL.Host,
+		Path:     "/cgi-bin/authLogin.cgi",
+		RawQuery: query.Encode(),
+	}
+	resp, err := http.Get(u.String())
+	if err != nil {
+		return "", nil, err
+	}
+	defer resp.Body.Close()
+	out, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		return "", nil, err
+	}
+	authResp := &qnapAuthResponse{}
+	if err := xml.Unmarshal(out, authResp); err != nil {
+		return "", nil, err
+	}
+	if authResp.AuthPassed == 0 {
+		return "", nil, fmt.Errorf("not authenticated")
+	}
+	return user.Value, authResp, nil
+}
+
+func synoAuthn() (string, error) {
+	cmd := exec.Command("/usr/syno/synoman/webman/modules/authenticate.cgi")
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		return "", fmt.Errorf("auth: %v: %s", err, out)
+	}
+	return strings.TrimSpace(string(out)), nil
+}
+
+func authRedirect(w http.ResponseWriter, r *http.Request) bool {
+	if distro.Get() == distro.Synology {
+		return synoTokenRedirect(w, r)
+	}
+	return false
+}
+
+func synoTokenRedirect(w http.ResponseWriter, r *http.Request) bool {
 	if r.Header.Get("X-Syno-Token") != "" {
 		return false
 	}
@@ -132,75 +250,13 @@ req.send(null);
 </body></html>
 `
 
-const authenticationRedirectHTML = `
-<html>
-<head>
-	<title>Redirecting...</title>
-	<style>
-		html,
-		body {
-			height: 100%;
-		}
-
-		html {
-			background-color: rgb(249, 247, 246);
-			font-family: ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, "Noto Sans", sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol", "Noto Color Emoji";
-			line-height: 1.5;
-			-webkit-text-size-adjust: 100%;
-			-webkit-font-smoothing: antialiased;
-			-moz-osx-font-smoothing: grayscale;
-		}
-
-		body {
-			display: flex;
-			flex-direction: column;
-			align-items: center;
-			justify-content: center;
-		}
-
-		.spinner {
-			margin-bottom: 2rem;
-			border: 4px rgba(112, 110, 109, 0.5) solid;
-			border-left-color: transparent;
-			border-radius: 9999px;
-			width: 4rem;
-			height: 4rem;
-			-webkit-animation: spin 700ms linear infinite;
-      animation: spin 800ms linear infinite;
-		}
-
-		.label {
-			color: rgb(112, 110, 109);
-			padding-left: 0.4rem;
-		}
-
-		@-webkit-keyframes spin {
-			to {
-				transform: rotate(360deg);
-			}
-		}
-
-		@keyframes spin {
-			to {
-				transform: rotate(360deg);
-			}
-		}
-	</style>
-</head>
-<body>
-	<div class="spinner"></div>
-	<div class="label">Redirecting...</div>
-</body>
-`
-
 func webHandler(w http.ResponseWriter, r *http.Request) {
-	if synoTokenRedirect(w, r) {
+	if authRedirect(w, r) {
 		return
 	}
 
-	user, err := auth()
+	user, err := authorize(w, r)
 	if err != nil {
-		http.Error(w, err.Error(), http.StatusForbidden)
 		return
 	}
 
@@ -214,7 +270,8 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 		w.Header().Set("Content-Type", "application/json")
 		url, err := tailscaleUpForceReauth(r.Context())
 		if err != nil {
-			json.NewEncoder(w).Encode(mi{"error": err})
+			w.WriteHeader(http.StatusInternalServerError)
+			json.NewEncoder(w).Encode(mi{"error": err.Error()})
 			return
 		}
 		json.NewEncoder(w).Encode(mi{"url": url})
@@ -223,7 +280,7 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 
 	st, err := tailscale.Status(r.Context())
 	if err != nil {
-		http.Error(w, err.Error(), 500)
+		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 
@@ -241,7 +298,7 @@ func webHandler(w http.ResponseWriter, r *http.Request) {
 
 	buf := new(bytes.Buffer)
 	if err := tmpl.Execute(buf, data); err != nil {
-		http.Error(w, err.Error(), 500)
+		http.Error(w, err.Error(), http.StatusInternalServerError)
 		return
 	}
 	w.Write(buf.Bytes())
@@ -320,6 +377,10 @@ func tailscaleUpForceReauth(ctx context.Context) (authURL string, retErr error)
 	})
 	bc.StartLoginInteractive()
 
+	<-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_test.go

@@ -0,0 +1,44 @@
+// 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 "testing"
+
+func TestUrlOfListenAddr(t *testing.T) {
+	tests := []struct {
+		name     string
+		in, want string
+	}{
+		{
+			name: "TestLocalhost",
+			in:   "localhost:8088",
+			want: "http://localhost:8088",
+		},
+		{
+			name: "TestNoHost",
+			in:   ":8088",
+			want: "http://127.0.0.1:8088",
+		},
+		{
+			name: "TestExplicitHost",
+			in:   "127.0.0.2:8088",
+			want: "http://127.0.0.2:8088",
+		},
+		{
+			name: "TestIPv6",
+			in:   "[::1]:8088",
+			want: "http://[::1]:8088",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			url := urlOfListenAddr(tt.in)
+			if url != tt.want {
+				t.Errorf("expected url: %q, got: %q", tt.want, url)
+			}
+		})
+	}
+}

cmd/tailscale/depaware.txt

@@ -1,10 +1,18 @@
 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/kballard/go-shellquote                            from tailscale.com/cmd/tailscale/cli
+     💣 github.com/mitchellh/go-ps                                   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/tailscale/goupnp                                  from github.com/tailscale/goupnp/dcps/internetgateway2
+        github.com/tailscale/goupnp/dcps/internetgateway2            from tailscale.com/net/portmapper
+        github.com/tailscale/goupnp/httpu                            from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/scpd                             from github.com/tailscale/goupnp
+        github.com/tailscale/goupnp/soap                             from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/ssdp                             from github.com/tailscale/goupnp
         github.com/tcnksm/go-httpstat                                from tailscale.com/net/netcheck
         github.com/toqueteos/webbrowser                              from tailscale.com/cmd/tailscale/cli
      💣 go4.org/intern                                               from inet.af/netaddr
@@ -14,13 +22,14 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         inet.af/netaddr                                              from tailscale.com/cmd/tailscale/cli+
         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                               from tailscale.com/cmd/tailscale/cli+
         tailscale.com/client/tailscale/apitype                       from tailscale.com/client/tailscale+
         tailscale.com/cmd/tailscale/cli                              from tailscale.com/cmd/tailscale
+        tailscale.com/control/controlknobs                           from tailscale.com/net/portmapper
         tailscale.com/derp                                           from tailscale.com/derp/derphttp
         tailscale.com/derp/derphttp                                  from tailscale.com/net/netcheck
-        tailscale.com/derp/derpmap                                   from tailscale.com/cmd/tailscale/cli
         tailscale.com/disco                                          from tailscale.com/derp
+        tailscale.com/hostinfo                                       from tailscale.com/net/interfaces
         tailscale.com/ipn                                            from tailscale.com/cmd/tailscale/cli+
         tailscale.com/ipn/ipnstate                                   from tailscale.com/cmd/tailscale/cli+
         tailscale.com/metrics                                        from tailscale.com/derp
@@ -40,6 +49,8 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         tailscale.com/syncs                                          from tailscale.com/net/interfaces+
         tailscale.com/tailcfg                                        from tailscale.com/cmd/tailscale/cli+
    W    tailscale.com/tsconst                                        from tailscale.com/net/interfaces
+     💣 tailscale.com/tstime/mono                                    from tailscale.com/cmd/tailscale/cli+
+        tailscale.com/tstime/rate                                    from tailscale.com/wgengine/filter
         tailscale.com/types/empty                                    from tailscale.com/ipn
         tailscale.com/types/ipproto                                  from tailscale.com/net/flowtrack+
         tailscale.com/types/key                                      from tailscale.com/derp+
@@ -48,14 +59,14 @@ 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+
    W    tailscale.com/util/endian                                    from tailscale.com/net/netns
-   L    tailscale.com/util/lineread                                  from tailscale.com/net/interfaces
+        tailscale.com/util/groupmember                               from tailscale.com/cmd/tailscale/cli
+        tailscale.com/util/lineread                                  from tailscale.com/net/interfaces+
         tailscale.com/version                                        from tailscale.com/cmd/tailscale/cli+
-        tailscale.com/version/distro                                 from tailscale.com/cmd/tailscale/cli
+        tailscale.com/version/distro                                 from tailscale.com/cmd/tailscale/cli+
         tailscale.com/wgengine/filter                                from tailscale.com/types/netmap
         golang.org/x/crypto/blake2b                                  from golang.org/x/crypto/nacl/box
         golang.org/x/crypto/chacha20                                 from golang.org/x/crypto/chacha20poly1305
@@ -75,7 +86,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         golang.org/x/net/idna                                        from golang.org/x/net/http/httpguts+
         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/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 tailscale.com/net/netns+
@@ -118,13 +129,14 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         debug/macho                                                  from rsc.io/goversion/version
         debug/pe                                                     from rsc.io/goversion/version
         embed                                                        from tailscale.com/cmd/tailscale/cli
-        encoding                                                     from encoding/json
+        encoding                                                     from encoding/json+
         encoding/asn1                                                from crypto/x509+
         encoding/base64                                              from encoding/json+
         encoding/binary                                              from compress/gzip+
         encoding/hex                                                 from crypto/x509+
         encoding/json                                                from expvar+
         encoding/pem                                                 from crypto/tls+
+        encoding/xml                                                 from tailscale.com/cmd/tailscale/cli+
         errors                                                       from bufio+
         expvar                                                       from tailscale.com/derp+
         flag                                                         from github.com/peterbourgon/ff/v2+
@@ -156,6 +168,7 @@ tailscale.com/cmd/tailscale dependencies: (generated by github.com/tailscale/dep
         os                                                           from crypto/rand+
         os/exec                                                      from github.com/toqueteos/webbrowser+
         os/signal                                                    from tailscale.com/cmd/tailscale/cli
+        os/user                                                      from tailscale.com/util/groupmember
         path                                                         from debug/dwarf+
         path/filepath                                                from crypto/x509+
         reflect                                                      from crypto/x509+

cmd/tailscaled/debug.go

@@ -11,19 +11,26 @@ import (
 	"errors"
 	"flag"
 	"fmt"
+	"io"
+	"io/ioutil"
 	"log"
+	"net"
 	"net/http"
 	"net/http/httptrace"
 	"net/url"
 	"os"
+	"strings"
 	"time"
 
+	"inet.af/netaddr"
 	"tailscale.com/derp/derphttp"
-	"tailscale.com/derp/derpmap"
+	"tailscale.com/ipn"
 	"tailscale.com/net/interfaces"
+	"tailscale.com/net/portmapper"
 	"tailscale.com/net/tshttpproxy"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/key"
+	"tailscale.com/types/logger"
 	"tailscale.com/wgengine/monitor"
 )
 
@@ -31,6 +38,7 @@ var debugArgs struct {
 	monitor   bool
 	getURL    string
 	derpCheck string
+	portmap   bool
 }
 
 var debugModeFunc = debugMode // so it can be addressable
@@ -38,6 +46,7 @@ var debugModeFunc = debugMode // so it can be addressable
 func debugMode(args []string) error {
 	fs := flag.NewFlagSet("debug", flag.ExitOnError)
 	fs.BoolVar(&debugArgs.monitor, "monitor", false, "If true, run link monitor forever. Precludes all other options.")
+	fs.BoolVar(&debugArgs.portmap, "portmap", false, "If true, run portmap debugging. Precludes all other options.")
 	fs.StringVar(&debugArgs.getURL, "get-url", "", "If non-empty, fetch provided URL.")
 	fs.StringVar(&debugArgs.derpCheck, "derp", "", "if non-empty, test a DERP ping via named region code")
 	if err := fs.Parse(args); err != nil {
@@ -53,6 +62,9 @@ func debugMode(args []string) error {
 	if debugArgs.monitor {
 		return runMonitor(ctx)
 	}
+	if debugArgs.portmap {
+		return debugPortmap(ctx)
+	}
 	if debugArgs.getURL != "" {
 		return getURL(ctx, debugArgs.getURL)
 	}
@@ -131,7 +143,26 @@ func getURL(ctx context.Context, urlStr string) error {
 }
 
 func checkDerp(ctx context.Context, derpRegion string) error {
-	dmap := derpmap.Prod()
+	req, err := http.NewRequestWithContext(ctx, "GET", ipn.DefaultControlURL+"/derpmap/default", nil)
+	if err != nil {
+		return fmt.Errorf("create derp map request: %w", err)
+	}
+	res, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return fmt.Errorf("fetch derp map failed: %w", err)
+	}
+	defer res.Body.Close()
+	b, err := ioutil.ReadAll(io.LimitReader(res.Body, 1<<20))
+	if err != nil {
+		return fmt.Errorf("fetch derp map failed: %w", err)
+	}
+	if res.StatusCode != 200 {
+		return fmt.Errorf("fetch derp map: %v: %s", res.Status, b)
+	}
+	var dmap tailcfg.DERPMap
+	if err = json.Unmarshal(b, &dmap); err != nil {
+		return fmt.Errorf("fetch DERP map: %w", err)
+	}
 	getRegion := func() *tailcfg.DERPRegion {
 		for _, r := range dmap.Regions {
 			if r.RegionCode == derpRegion {
@@ -170,3 +201,83 @@ func checkDerp(ctx context.Context, derpRegion string) error {
 	log.Printf("ok")
 	return err
 }
+
+func debugPortmap(ctx context.Context) error {
+	ctx, cancel := context.WithTimeout(ctx, 3*time.Second)
+	defer cancel()
+
+	portmapper.VerboseLogs = true
+
+	done := make(chan bool, 1)
+
+	var c *portmapper.Client
+	logf := log.Printf
+	c = portmapper.NewClient(logger.WithPrefix(logf, "portmapper: "), func() {
+		logf("portmapping changed.")
+		logf("have mapping: %v", c.HaveMapping())
+
+		if ext, ok := c.GetCachedMappingOrStartCreatingOne(); ok {
+			logf("cb: mapping: %v", ext)
+			select {
+			case done <- true:
+			default:
+			}
+			return
+		}
+		logf("cb: no mapping")
+	})
+	linkMon, err := monitor.New(logger.WithPrefix(logf, "monitor: "))
+	if err != nil {
+		return err
+	}
+
+	gatewayAndSelfIP := func() (gw, self netaddr.IP, ok bool) {
+		if v := os.Getenv("TS_DEBUG_GW_SELF"); strings.Contains(v, "/") {
+			i := strings.Index(v, "/")
+			gw = netaddr.MustParseIP(v[:i])
+			self = netaddr.MustParseIP(v[i+1:])
+			return gw, self, true
+		}
+		return linkMon.GatewayAndSelfIP()
+	}
+
+	c.SetGatewayLookupFunc(gatewayAndSelfIP)
+
+	gw, selfIP, ok := gatewayAndSelfIP()
+	if !ok {
+		logf("no gateway or self IP; %v", linkMon.InterfaceState())
+		return nil
+	}
+	logf("gw=%v; self=%v", gw, selfIP)
+
+	res, err := c.Probe(ctx)
+	if err != nil {
+		return fmt.Errorf("Probe: %v", err)
+	}
+	logf("Probe: %+v", res)
+
+	if !res.PCP && !res.PMP && !res.UPnP {
+		logf("no portmapping services available")
+		return nil
+	}
+
+	uc, err := net.ListenPacket("udp", "0.0.0.0:0")
+	if err != nil {
+		return err
+	}
+	defer uc.Close()
+	c.SetLocalPort(uint16(uc.LocalAddr().(*net.UDPAddr).Port))
+
+	if ext, ok := c.GetCachedMappingOrStartCreatingOne(); ok {
+		logf("mapping: %v", ext)
+	} else {
+		logf("no mapping")
+	}
+
+	select {
+	case <-done:
+		return nil
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+}

cmd/tailscaled/depaware.txt

@@ -1,43 +1,53 @@
 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/netlink                                  from tailscale.com/wgengine/monitor+
+   L 💣 github.com/mdlayher/netlink/nlenc                            from github.com/mdlayher/netlink+
    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+
+   L 💣 github.com/mdlayher/socket                                   from github.com/mdlayher/netlink
+   W    github.com/pkg/errors                                        from github.com/tailscale/certstore
+   W 💣 github.com/tailscale/certstore                               from tailscale.com/control/controlclient
+        github.com/tailscale/goupnp                                  from github.com/tailscale/goupnp/dcps/internetgateway2
+        github.com/tailscale/goupnp/dcps/internetgateway2            from tailscale.com/net/portmapper
+        github.com/tailscale/goupnp/httpu                            from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/scpd                             from github.com/tailscale/goupnp
+        github.com/tailscale/goupnp/soap                             from github.com/tailscale/goupnp+
+        github.com/tailscale/goupnp/ssdp                             from github.com/tailscale/goupnp
         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/mem                                                  from tailscale.com/derp+
         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/atomicbitops                                from inet.af/netstack/tcpip+
+     💣 inet.af/netstack/buffer                                      from inet.af/netstack/tcpip/stack
      💣 inet.af/netstack/gohacks                                     from inet.af/netstack/state/wire+
         inet.af/netstack/linewriter                                  from inet.af/netstack/log
         inet.af/netstack/log                                         from inet.af/netstack/state+
@@ -46,7 +56,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 inet.af/netstack/state                                       from inet.af/netstack/tcpip+
         inet.af/netstack/state/wire                                  from inet.af/netstack/state
      💣 inet.af/netstack/sync                                        from inet.af/netstack/linewriter+
-     💣 inet.af/netstack/tcpip                                       from inet.af/netstack/tcpip/adapters/gonet+
+        inet.af/netstack/tcpip                                       from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/tcpip/adapters/gonet                        from tailscale.com/wgengine/netstack
      💣 inet.af/netstack/tcpip/buffer                                from inet.af/netstack/tcpip/adapters/gonet+
         inet.af/netstack/tcpip/hash/jenkins                          from inet.af/netstack/tcpip/stack+
@@ -69,16 +79,18 @@ 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                               from tailscale.com/derp
         tailscale.com/client/tailscale/apitype                       from tailscale.com/ipn/ipnlocal+
         tailscale.com/control/controlclient                          from tailscale.com/ipn/ipnlocal+
+        tailscale.com/control/controlknobs                           from tailscale.com/control/controlclient+
         tailscale.com/derp                                           from tailscale.com/derp/derphttp+
         tailscale.com/derp/derphttp                                  from tailscale.com/net/netcheck+
-        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/hostinfo                                       from tailscale.com/control/controlclient+
         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
@@ -103,7 +115,8 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
      💣 tailscale.com/net/netstat                                    from tailscale.com/ipn/ipnserver
         tailscale.com/net/packet                                     from tailscale.com/wgengine+
         tailscale.com/net/portmapper                                 from tailscale.com/net/netcheck+
-        tailscale.com/net/socks5                                     from tailscale.com/cmd/tailscaled
+        tailscale.com/net/socks5                                     from tailscale.com/net/socks5/tssocks
+        tailscale.com/net/socks5/tssocks                             from tailscale.com/cmd/tailscaled
         tailscale.com/net/stun                                       from tailscale.com/net/netcheck+
         tailscale.com/net/tlsdial                                    from tailscale.com/control/controlclient+
         tailscale.com/net/tsaddr                                     from tailscale.com/ipn/ipnlocal+
@@ -111,13 +124,14 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         tailscale.com/net/tstun                                      from tailscale.com/cmd/tailscaled+
         tailscale.com/paths                                          from tailscale.com/cmd/tailscaled+
         tailscale.com/portlist                                       from tailscale.com/ipn/ipnlocal
-        tailscale.com/safesocket                                     from tailscale.com/ipn/ipnserver
+        tailscale.com/safesocket                                     from tailscale.com/ipn/ipnserver+
         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/tstime/mono                                    from tailscale.com/net/tstun+
+        tailscale.com/tstime/rate                                    from tailscale.com/wgengine/filter
         tailscale.com/types/empty                                    from tailscale.com/control/controlclient+
         tailscale.com/types/flagtype                                 from tailscale.com/cmd/tailscaled
         tailscale.com/types/ipproto                                  from tailscale.com/net/flowtrack+
@@ -128,13 +142,14 @@ 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/deephash                                  from tailscale.com/ipn/ipnlocal+
         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/groupmember                               from tailscale.com/ipn/ipnserver
+        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
@@ -143,18 +158,19 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         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/wgengine+
         tailscale.com/wgengine/monitor                               from tailscale.com/wgengine+
-        tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled
+        tailscale.com/wgengine/netstack                              from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/router                                from tailscale.com/cmd/tailscaled+
         tailscale.com/wgengine/wgcfg                                 from tailscale.com/ipn/ipnlocal+
         tailscale.com/wgengine/wgcfg/nmcfg                           from tailscale.com/ipn/ipnlocal
         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+
@@ -163,7 +179,7 @@ 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+
@@ -171,15 +187,15 @@ 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/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+
+  LD    golang.org/x/sys/unix                                        from github.com/mdlayher/netlink+
+   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/mgr                             from tailscale.com/cmd/tailscaled
@@ -221,7 +237,7 @@ 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
-   L    embed                                                        from tailscale.com/net/dns
+        embed                                                        from tailscale.com/net/dns+
         encoding                                                     from encoding/json+
         encoding/asn1                                                from crypto/x509+
         encoding/base64                                              from encoding/json+
@@ -229,6 +245,7 @@ tailscale.com/cmd/tailscaled dependencies: (generated by github.com/tailscale/de
         encoding/hex                                                 from crypto/x509+
         encoding/json                                                from expvar+
         encoding/pem                                                 from crypto/tls+
+        encoding/xml                                                 from github.com/tailscale/goupnp+
         errors                                                       from bufio+
         expvar                                                       from tailscale.com/derp+
         flag                                                         from tailscale.com/cmd/tailscaled+

cmd/tailscaled/tailscaled.go

@@ -24,22 +24,19 @@ import (
 	"runtime/debug"
 	"strconv"
 	"strings"
-	"sync"
 	"syscall"
 	"time"
 
 	"github.com/go-multierror/multierror"
-	"inet.af/netaddr"
+	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnserver"
 	"tailscale.com/logpolicy"
 	"tailscale.com/net/dns"
-	"tailscale.com/net/socks5"
-	"tailscale.com/net/tsaddr"
+	"tailscale.com/net/socks5/tssocks"
 	"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"
@@ -49,15 +46,6 @@ import (
 	"tailscale.com/wgengine/router"
 )
 
-// globalStateKey is the ipn.StateKey that tailscaled loads on
-// startup.
-//
-// We have to support multiple state keys for other OSes (Windows in
-// particular), but right now Unix daemons run with a single
-// node-global state. To keep open the option of having per-user state
-// later, the global state key doesn't look like a username.
-const globalStateKey = "_daemon"
-
 // defaultTunName returns the default tun device name for the platform.
 func defaultTunName() string {
 	switch runtime.GOOS {
@@ -171,6 +159,28 @@ func main() {
 	}
 }
 
+func ipnServerOpts() (o ipnserver.Options) {
+	// Allow changing the OS-specific IPN behavior for tests
+	// so we can e.g. test Windows-specific behaviors on Linux.
+	goos := os.Getenv("TS_DEBUG_TAILSCALED_IPN_GOOS")
+	if goos == "" {
+		goos = runtime.GOOS
+	}
+
+	o.Port = 41112
+	o.StatePath = args.statepath
+	o.SocketPath = args.socketpath // even for goos=="windows", for tests
+
+	switch goos {
+	default:
+		o.SurviveDisconnects = true
+		o.AutostartStateKey = ipn.GlobalDaemonStateKey
+	case "windows":
+		// Not those.
+	}
+	return o
+}
+
 func run() error {
 	var err error
 
@@ -200,6 +210,9 @@ func run() error {
 	logf = logger.RateLimitedFn(logf, 5*time.Second, 5, 100)
 
 	if args.cleanup {
+		if os.Getenv("TS_PLEASE_PANIC") != "" {
+			panic("TS_PLEASE_PANIC asked us to panic")
+		}
 		dns.Cleanup(logf, args.tunname)
 		router.Cleanup(logf, args.tunname)
 		return nil
@@ -228,6 +241,11 @@ func run() error {
 		if err != nil {
 			log.Fatalf("SOCKS5 listener: %v", err)
 		}
+		if strings.HasSuffix(args.socksAddr, ":0") {
+			// Log kernel-selected port number so integration tests
+			// can find it portably.
+			log.Printf("SOCKS5 listening on %v", socksListener.Addr())
+		}
 	}
 
 	e, useNetstack, err := createEngine(logf, linkMon)
@@ -243,35 +261,7 @@ func run() error {
 	}
 
 	if socksListener != nil {
-		srv := &socks5.Server{
-			Logf: logger.WithPrefix(logf, "socks5: "),
-		}
-		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)
-			}
-			var d net.Dialer
-			return d.DialContext(ctx, network, ipp.String())
-		}
+		srv := tssocks.NewServer(logger.WithPrefix(logf, "socks5: "), e, ns)
 		go func() {
 			log.Fatalf("SOCKS5 server exited: %v", srv.Serve(socksListener))
 		}()
@@ -298,14 +288,8 @@ func run() error {
 		}
 	}()
 
-	opts := ipnserver.Options{
-		SocketPath:         args.socketpath,
-		Port:               41112,
-		StatePath:          args.statepath,
-		AutostartStateKey:  globalStateKey,
-		SurviveDisconnects: runtime.GOOS != "windows",
-		DebugMux:           debugMux,
-	}
+	opts := ipnServerOpts()
+	opts.DebugMux = debugMux
 	err = ipnserver.Run(ctx, logf, pol.PublicID.String(), ipnserver.FixedEngine(e), opts)
 	// Cancelation is not an error: it is the only way to stop ipnserver.
 	if err != nil && err != context.Canceled {
@@ -368,7 +352,7 @@ func tryEngine(logf logger.Logf, linkMon *monitor.Mon, name string) (e wgengine.
 			return nil, false, err
 		}
 		conf.Tun = dev
-		r, err := router.New(logf, dev)
+		r, err := router.New(logf, dev, linkMon)
 		if err != nil {
 			dev.Close()
 			return nil, false, err

cmd/tailscaled/tailscaled.openrc

@@ -0,0 +1,25 @@
+#!/sbin/openrc-run
+
+set -a
+source /etc/default/tailscaled
+set +a
+
+command="/usr/sbin/tailscaled"
+command_args="--state=/var/lib/tailscale/tailscaled.state --port=$PORT --socket=/var/run/tailscale/tailscaled.sock $FLAGS"
+command_background=true
+pidfile="/run/tailscaled.pid"
+start_stop_daemon_args="-1 /var/log/tailscaled.log -2 /var/log/tailscaled.log"
+
+depend() {
+    need net
+}
+
+start_pre() {
+    mkdir -p /var/run/tailscale
+    mkdir -p /var/lib/tailscale
+    $command --cleanup
+}
+
+stop_post() {
+    $command --cleanup
+}

cmd/tailscaled/tailscaled.service

@@ -2,7 +2,7 @@
 Description=Tailscale node agent
 Documentation=https://tailscale.com/kb/
 Wants=network-pre.target
-After=network-pre.target
+After=network-pre.target NetworkManager.service systemd-resolved.service
 
 [Service]
 EnvironmentFile=/etc/default/tailscaled

cmd/tailscaled/tailscaled_windows.go

@@ -19,22 +19,23 @@ package main // import "tailscale.com/cmd/tailscaled"
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
 	"log"
-	"net"
 	"os"
 	"time"
 
 	"golang.org/x/sys/windows"
 	"golang.org/x/sys/windows/svc"
 	"golang.zx2c4.com/wireguard/windows/tunnel/winipcfg"
+	"inet.af/netaddr"
 	"tailscale.com/ipn/ipnserver"
 	"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"
@@ -127,16 +128,6 @@ func beFirewallKillswitch() bool {
 	log.SetFlags(0)
 	log.Printf("killswitch subprocess starting, tailscale GUID is %s", os.Args[2])
 
-	go func() {
-		b := make([]byte, 16)
-		for {
-			_, err := os.Stdin.Read(b)
-			if err != nil {
-				log.Fatalf("parent process died or requested exit, exiting (%v)", err)
-			}
-		}
-	}()
-
 	guid, err := windows.GUIDFromString(os.Args[2])
 	if err != nil {
 		log.Fatalf("invalid GUID %q: %v", os.Args[2], err)
@@ -144,17 +135,29 @@ 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)
+	fw, err := wf.New(uint64(luid))
+	if err != nil {
+		log.Fatalf("failed to enable firewall: %v", err)
+	}
 	log.Printf("killswitch enabled, took %s", time.Since(start))
 
-	// Block until the monitor goroutine shuts us down.
-	select {}
+	// Note(maisem): when local lan access toggled, tailscaled needs to
+	// inform the firewall to let local routes through. The set of routes
+	// is passed in via stdin encoded in json.
+	dcd := json.NewDecoder(os.Stdin)
+	for {
+		var routes []netaddr.IPPrefix
+		if err := dcd.Decode(&routes); err != nil {
+			log.Fatalf("parent process died or requested exit, exiting (%v)", err)
+		}
+		if err := fw.UpdatePermittedRoutes(routes); err != nil {
+			log.Fatalf("failed to update routes (%v)", err)
+		}
+	}
 }
 
 func startIPNServer(ctx context.Context, logid string) error {
@@ -165,7 +168,7 @@ func startIPNServer(ctx context.Context, logid string) error {
 		if err != nil {
 			return nil, fmt.Errorf("TUN: %w", err)
 		}
-		r, err := router.New(logf, dev)
+		r, err := router.New(logf, dev, nil)
 		if err != nil {
 			dev.Close()
 			return nil, fmt.Errorf("router: %w", err)
@@ -230,12 +233,6 @@ func startIPNServer(ctx context.Context, logid string) error {
 		}
 	}()
 
-	opts := ipnserver.Options{
-		Port:               41112,
-		SurviveDisconnects: false,
-		StatePath:          args.statepath,
-	}
-
 	// 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.
@@ -260,7 +257,7 @@ func startIPNServer(ctx context.Context, logid string) error {
 			return nil, fmt.Errorf("%w\n\nlogid: %v", res.Err, logid)
 		}
 	}
-	err := ipnserver.Run(ctx, logf, logid, getEngine, opts)
+	err := ipnserver.Run(ctx, logf, logid, getEngine, ipnServerOpts())
 	if err != nil {
 		logf("ipnserver.Run: %v", err)
 	}

cmd/tsshd/tsshd.go

@@ -29,8 +29,8 @@ import (
 	"time"
 	"unsafe"
 
+	"github.com/creack/pty"
 	"github.com/gliderlabs/ssh"
-	"github.com/kr/pty"
 	gossh "golang.org/x/crypto/ssh"
 	"inet.af/netaddr"
 	"tailscale.com/net/interfaces"

control/controlclient/auto.go

@@ -576,9 +576,12 @@ func (c *Auto) sendStatus(who string, err error, url string, nm *netmap.NetworkM
 	c.logf("[v1] sendStatus: %s: %v", who, state)
 
 	var p *persist.Persist
-	var fin *empty.Message
+	var loginFin, logoutFin *empty.Message
 	if state == StateAuthenticated {
-		fin = new(empty.Message)
+		loginFin = new(empty.Message)
+	}
+	if state == StateNotAuthenticated {
+		logoutFin = new(empty.Message)
 	}
 	if nm != nil && loggedIn && synced {
 		pp := c.direct.GetPersist()
@@ -589,12 +592,13 @@ func (c *Auto) sendStatus(who string, err error, url string, nm *netmap.NetworkM
 		nm = nil
 	}
 	new := Status{
-		LoginFinished: fin,
-		URL:           url,
-		Persist:       p,
-		NetMap:        nm,
-		Hostinfo:      hi,
-		State:         state,
+		LoginFinished:  loginFin,
+		LogoutFinished: logoutFin,
+		URL:            url,
+		Persist:        p,
+		NetMap:         nm,
+		Hostinfo:       hi,
+		State:          state,
 	}
 	if err != nil {
 		new.Err = err.Error()
@@ -712,3 +716,9 @@ func (c *Auto) TestOnlySetAuthKey(authkey string) {
 func (c *Auto) TestOnlyTimeNow() time.Time {
 	return c.timeNow()
 }
+
+// SetDNS sends the SetDNSRequest request to the control plane server,
+// requesting a DNS record be created or updated.
+func (c *Auto) SetDNS(ctx context.Context, req *tailcfg.SetDNSRequest) error {
+	return c.direct.SetDNS(ctx, req)
+}

control/controlclient/client.go

@@ -74,4 +74,7 @@ type Client interface {
 	// in a separate http request. It has nothing to do with the rest of
 	// the state machine.
 	UpdateEndpoints(localPort uint16, endpoints []tailcfg.Endpoint)
+	// SetDNS sends the SetDNSRequest request to the control plane server,
+	// requesting a DNS record be created or updated.
+	SetDNS(context.Context, *tailcfg.SetDNSRequest) error
 }

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", "NetMap", "State", "Persist", "Hostinfo"}
+	equalHandles := []string{"LoginFinished", "LogoutFinished", "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/debug.go

@@ -11,7 +11,6 @@ import (
 	"fmt"
 	"log"
 	"net/http"
-	"regexp"
 	"runtime"
 	"strconv"
 	"time"
@@ -42,28 +41,82 @@ func dumpGoroutinesToURL(c *http.Client, targetURL string) {
 	}
 }
 
-var reHexArgs = regexp.MustCompile(`\b0x[0-9a-f]+\b`)
-
 // scrubbedGoroutineDump returns the list of all current goroutines, but with the actual
 // values of arguments scrubbed out, lest it contain some private key material.
 func scrubbedGoroutineDump() []byte {
-	buf := make([]byte, 1<<20)
-	buf = buf[:runtime.Stack(buf, true)]
+	var buf []byte
+	// Grab stacks multiple times into increasingly larger buffer sizes
+	// to minimize the risk that we blow past our iOS memory limit.
+	for size := 1 << 10; size <= 1<<20; size += 1 << 10 {
+		buf = make([]byte, size)
+		buf = buf[:runtime.Stack(buf, true)]
+		if len(buf) < size {
+			// It fit.
+			break
+		}
+	}
+	return scrubHex(buf)
+}
 
+func scrubHex(buf []byte) []byte {
 	saw := map[string][]byte{} // "0x123" => "v1%3" (unique value 1 and its value mod 8)
-	return reHexArgs.ReplaceAllFunc(buf, func(in []byte) []byte {
+
+	foreachHexAddress(buf, func(in []byte) {
 		if string(in) == "0x0" {
-			return in
+			return
 		}
 		if v, ok := saw[string(in)]; ok {
-			return v
+			for i := range in {
+				in[i] = '_'
+			}
+			copy(in, v)
+			return
 		}
+		inStr := string(in)
 		u64, err := strconv.ParseUint(string(in[2:]), 16, 64)
+		for i := range in {
+			in[i] = '_'
+		}
 		if err != nil {
-			return []byte("??")
+			in[0] = '?'
+			return
 		}
 		v := []byte(fmt.Sprintf("v%d%%%d", len(saw)+1, u64%8))
-		saw[string(in)] = v
-		return v
+		saw[inStr] = v
+		copy(in, v)
 	})
+	return buf
+}
+
+var ohx = []byte("0x")
+
+// foreachHexAddress calls f with each subslice of b that matches
+// regexp `0x[0-9a-f]*`.
+func foreachHexAddress(b []byte, f func([]byte)) {
+	for len(b) > 0 {
+		i := bytes.Index(b, ohx)
+		if i == -1 {
+			return
+		}
+		b = b[i:]
+		hx := hexPrefix(b)
+		f(hx)
+		b = b[len(hx):]
+	}
+}
+
+func hexPrefix(b []byte) []byte {
+	for i, c := range b {
+		if i < 2 {
+			continue
+		}
+		if !isHexByte(c) {
+			return b[:i]
+		}
+	}
+	return b
+}
+
+func isHexByte(b byte) bool {
+	return '0' <= b && b <= '9' || 'a' <= b && b <= 'f' || 'A' <= b && b <= 'F'
 }

control/controlclient/debug_test.go

@@ -9,3 +9,22 @@ import "testing"
 func TestScrubbedGoroutineDump(t *testing.T) {
 	t.Logf("Got:\n%s\n", scrubbedGoroutineDump())
 }
+
+func TestScrubHex(t *testing.T) {
+	tests := []struct {
+		in, want string
+	}{
+		{"foo", "foo"},
+		{"", ""},
+		{"0x", "?_"},
+		{"0x001 and same 0x001", "v1%1_ and same v1%1_"},
+		{"0x008 and same 0x008", "v1%0_ and same v1%0_"},
+		{"0x001 and diff 0x002", "v1%1_ and diff v2%2_"},
+	}
+	for _, tt := range tests {
+		got := scrubHex([]byte(tt.in))
+		if string(got) != tt.want {
+			t.Errorf("for input:\n%s\n\ngot:\n%s\n\nwant:\n%s\n", tt.in, got, tt.want)
+		}
+	}
+}

control/controlclient/direct.go

@@ -31,7 +31,9 @@ import (
 
 	"golang.org/x/crypto/nacl/box"
 	"inet.af/netaddr"
+	"tailscale.com/control/controlknobs"
 	"tailscale.com/health"
+	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/log/logheap"
 	"tailscale.com/net/dnscache"
 	"tailscale.com/net/dnsfallback"
@@ -66,6 +68,7 @@ type Direct struct {
 	debugFlags             []string
 	keepSharerAndUserSplit bool
 	skipIPForwardingCheck  bool
+	pinger                 Pinger
 
 	mu           sync.Mutex // mutex guards the following fields
 	serverKey    wgkey.Key
@@ -78,6 +81,7 @@ type Direct struct {
 	endpoints     []tailcfg.Endpoint
 	everEndpoints bool   // whether we've ever had non-empty endpoints
 	localPort     uint16 // or zero to mean auto
+	lastPingURL   string // last PingRequest.URL received, for dup suppresion
 }
 
 type Options struct {
@@ -103,6 +107,18 @@ type Options struct {
 	// forwarding works and should not be double-checked by the
 	// controlclient package.
 	SkipIPForwardingCheck bool
+
+	// Pinger optionally specifies the Pinger to use to satisfy
+	// MapResponse.PingRequest queries from the control plane.
+	// If nil, PingRequest queries are not answered.
+	Pinger Pinger
+}
+
+// Pinger is a subset of the wgengine.Engine interface, containing just the Ping method.
+type Pinger interface {
+	// Ping is a request to start a discovery or TSMP 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 Decompressor interface {
@@ -165,6 +181,7 @@ 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())
@@ -203,6 +220,13 @@ func packageType() string {
 		// Using tailscaled or IPNExtension?
 		exe, _ := os.Executable()
 		return filepath.Base(exe)
+	case "linux":
+		// Report whether this is in a snap.
+		// See https://snapcraft.io/docs/environment-variables
+		// We just look at two somewhat arbitrarily.
+		if os.Getenv("SNAP_NAME") != "" && os.Getenv("SNAP") != "" {
+			return "snap"
+		}
 	}
 	return ""
 }
@@ -338,6 +362,7 @@ func (c *Direct) doLogin(ctx context.Context, opt loginOpt) (mustRegen bool, new
 		if err != nil {
 			return regen, opt.URL, err
 		}
+		c.logf("control server key %s from %s", serverKey.HexString(), c.serverURL)
 
 		c.mu.Lock()
 		c.serverKey = serverKey
@@ -460,10 +485,10 @@ 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
 	}
 
@@ -760,7 +785,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			health.GotStreamedMapResponse()
 		}
 
-		if pr := resp.PingRequest; pr != nil {
+		if pr := resp.PingRequest; pr != nil && c.isUniquePingRequest(pr) {
 			go answerPing(c.logf, c.httpc, pr)
 		}
 
@@ -780,7 +805,10 @@ func (c *Direct) sendMapRequest(ctx context.Context, maxPolls int, cb func(*netm
 			continue
 		}
 
-		if resp.Debug != nil {
+		hasDebug := resp.Debug != nil
+		// being conservative here, if Debug not present set to False
+		controlknobs.SetDisableUPnP(hasDebug && resp.Debug.DisableUPnP.EqualBool(true))
+		if hasDebug {
 			if resp.Debug.LogHeapPprof {
 				go logheap.LogHeap(resp.Debug.LogHeapURL)
 			}
@@ -1091,7 +1119,7 @@ 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
 		}
 	}
 
@@ -1100,10 +1128,10 @@ 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
@@ -1155,6 +1183,23 @@ func ipForwardingBroken(routes []netaddr.IPPrefix, state *interfaces.State) bool
 	return false
 }
 
+// isUniquePingRequest reports whether pr contains a new PingRequest.URL
+// not already handled, noting its value when returning true.
+func (c *Direct) isUniquePingRequest(pr *tailcfg.PingRequest) bool {
+	if pr == nil || pr.URL == "" {
+		// Bogus.
+		return false
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if pr.URL == c.lastPingURL {
+		return false
+	}
+	c.lastPingURL = pr.URL
+	return true
+}
+
 func answerPing(logf logger.Logf, c *http.Client, pr *tailcfg.PingRequest) {
 	if pr.URL == "" {
 		logf("invalid PingRequest with no URL")
@@ -1211,3 +1256,50 @@ func sleepAsRequested(ctx context.Context, logf logger.Logf, timeoutReset chan<-
 		}
 	}
 }
+
+// SetDNS sends the SetDNSRequest request to the control plane server,
+// requesting a DNS record be created or updated.
+func (c *Direct) SetDNS(ctx context.Context, req *tailcfg.SetDNSRequest) error {
+	c.mu.Lock()
+	serverKey := c.serverKey
+	c.mu.Unlock()
+
+	if serverKey.IsZero() {
+		return errors.New("zero serverKey")
+	}
+	machinePrivKey, err := c.getMachinePrivKey()
+	if err != nil {
+		return fmt.Errorf("getMachinePrivKey: %w", err)
+	}
+	if machinePrivKey.IsZero() {
+		return errors.New("getMachinePrivKey returned zero key")
+	}
+
+	bodyData, err := encode(req, &serverKey, &machinePrivKey)
+	if err != nil {
+		return err
+	}
+	body := bytes.NewReader(bodyData)
+
+	u := fmt.Sprintf("%s/machine/%s/set-dns", c.serverURL, machinePrivKey.Public().HexString())
+	hreq, err := http.NewRequestWithContext(ctx, "POST", u, body)
+	if err != nil {
+		return err
+	}
+	res, err := c.httpc.Do(hreq)
+	if err != nil {
+		return err
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		msg, _ := ioutil.ReadAll(res.Body)
+		return fmt.Errorf("set-dns response: %v, %.200s", res.Status, strings.TrimSpace(string(msg)))
+	}
+	var setDNSRes struct{} // no fields yet
+	if err := decode(res, &setDNSRes, &serverKey, &machinePrivKey); err != nil {
+		c.logf("error decoding SetDNSResponse with server key %s and machine key %s: %v", serverKey, machinePrivKey.Public(), err)
+		return fmt.Errorf("set-dns-response: %v", err)
+	}
+
+	return nil
+}

control/controlclient/direct_test.go

@@ -86,7 +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

control/controlclient/hostinfo_linux.go

@@ -9,13 +9,11 @@ package controlclient
 import (
 	"bytes"
 	"fmt"
-	"io"
 	"io/ioutil"
-	"os"
 	"strings"
 	"syscall"
 
-	"go4.org/mem"
+	"tailscale.com/hostinfo"
 	"tailscale.com/util/lineread"
 	"tailscale.com/version/distro"
 )
@@ -56,11 +54,11 @@ func osVersionLinux() string {
 		}
 		attrBuf.WriteByte(byte(b))
 	}
-	if inContainer() {
+	if hostinfo.InContainer() {
 		attrBuf.WriteString("; container")
 	}
-	if inKnative() {
-		attrBuf.WriteString("; env=kn")
+	if env := hostinfo.GetEnvType(); env != "" {
+		fmt.Fprintf(&attrBuf, "; env=%s", env)
 	}
 	attr := attrBuf.String()
 
@@ -93,31 +91,3 @@ func osVersionLinux() string {
 	}
 	return fmt.Sprintf("Other%s", attr)
 }
-
-func inContainer() (ret bool) {
-	lineread.File("/proc/1/cgroup", func(line []byte) error {
-		if mem.Contains(mem.B(line), mem.S("/docker/")) ||
-			mem.Contains(mem.B(line), mem.S("/lxc/")) {
-			ret = true
-			return io.EOF // arbitrary non-nil error to stop loop
-		}
-		return nil
-	})
-	lineread.File("/proc/mounts", func(line []byte) error {
-		if mem.Contains(mem.B(line), mem.S("fuse.lxcfs")) {
-			ret = true
-			return io.EOF
-		}
-		return nil
-	})
-	return
-}
-
-func inKnative() bool {
-	// https://cloud.google.com/run/docs/reference/container-contract#env-vars
-	if os.Getenv("K_REVISION") != "" && os.Getenv("K_CONFIGURATION") != "" &&
-		os.Getenv("K_SERVICE") != "" && os.Getenv("PORT") != "" {
-		return true
-	}
-	return false
-}

control/controlclient/map.go

@@ -6,8 +6,11 @@ package controlclient
 
 import (
 	"log"
+	"os"
 	"sort"
+	"strconv"
 
+	"inet.af/netaddr"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
@@ -124,7 +127,7 @@ func (ms *mapSession) netmapForResponse(resp *tailcfg.MapResponse) *netmap.Netwo
 		nm.SelfNode = node
 		nm.Expiry = node.KeyExpiry
 		nm.Name = node.Name
-		nm.Addresses = node.Addresses
+		nm.Addresses = filterSelfAddresses(node.Addresses)
 		nm.User = node.User
 		nm.Hostinfo = node.Hostinfo
 		if node.MachineAuthorized {
@@ -280,3 +283,19 @@ func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
 	}
 	return v2
 }
+
+var debugSelfIPv6Only, _ = strconv.ParseBool(os.Getenv("TS_DEBUG_SELF_V6_ONLY"))
+
+func filterSelfAddresses(in []netaddr.IPPrefix) (ret []netaddr.IPPrefix) {
+	switch {
+	default:
+		return in
+	case debugSelfIPv6Only:
+		for _, a := range in {
+			if a.IP().Is6() {
+				ret = append(ret, a)
+			}
+		}
+		return ret
+	}
+}

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

@@ -64,11 +64,12 @@ func (s State) String() string {
 }
 
 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
+	_              structs.Incomparable
+	LoginFinished  *empty.Message // nonempty when login finishes
+	LogoutFinished *empty.Message // nonempty when logout 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
@@ -86,6 +87,7 @@ func (s *Status) Equal(s2 *Status) bool {
 	}
 	return s != nil && s2 != nil &&
 		(s.LoginFinished == nil) == (s2.LoginFinished == nil) &&
+		(s.LogoutFinished == nil) == (s2.LogoutFinished == nil) &&
 		s.Err == s2.Err &&
 		s.URL == s2.URL &&
 		reflect.DeepEqual(s.Persist, s2.Persist) &&

control/controlknobs/controlknobs.go

@@ -0,0 +1,34 @@
+// 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 controlknobs contains client options configurable from control which can be turned on
+// or off. The ability to turn options on and off is for incrementally adding features in.
+package controlknobs
+
+import (
+	"os"
+	"strconv"
+
+	"tailscale.com/syncs"
+)
+
+// disableUPnP indicates whether to attempt UPnP mapping.
+var disableUPnP syncs.AtomicBool
+
+func init() {
+	v, _ := strconv.ParseBool(os.Getenv("TS_DISABLE_UPNP"))
+	SetDisableUPnP(v)
+}
+
+// DisableUPnP reports the last reported value from control
+// whether UPnP portmapping should be disabled.
+func DisableUPnP() bool {
+	return disableUPnP.Get()
+}
+
+// SetDisableUPnP sets whether control says that UPnP should be
+// disabled.
+func SetDisableUPnP(v bool) {
+	disableUPnP.Set(v)
+}

derp/derp_client.go

@@ -29,6 +29,7 @@ type Client struct {
 	br          *bufio.Reader
 	meshKey     string
 	canAckPings bool
+	isProber    bool
 
 	wmu sync.Mutex // hold while writing to bw
 	bw  *bufio.Writer
@@ -52,6 +53,7 @@ type clientOpt struct {
 	MeshKey     string
 	ServerPub   key.Public
 	CanAckPings bool
+	IsProber    bool
 }
 
 // MeshKey returns a ClientOpt to pass to the DERP server during connect to get
@@ -60,6 +62,10 @@ type clientOpt struct {
 // An empty key means to not use a mesh key.
 func MeshKey(key string) ClientOpt { return clientOptFunc(func(o *clientOpt) { o.MeshKey = key }) }
 
+// IsProber returns a ClientOpt to pass to the DERP server during connect to
+// declare that this client is a a prober.
+func IsProber(v bool) ClientOpt { return clientOptFunc(func(o *clientOpt) { o.IsProber = v }) }
+
 // ServerPublicKey returns a ClientOpt to declare that the server's DERP public key is known.
 // If key is the zero value, the returned ClientOpt is a no-op.
 func ServerPublicKey(key key.Public) ClientOpt {
@@ -93,6 +99,7 @@ func newClient(privateKey key.Private, nc Conn, brw *bufio.ReadWriter, logf logg
 		bw:          brw.Writer,
 		meshKey:     opt.MeshKey,
 		canAckPings: opt.CanAckPings,
+		isProber:    opt.IsProber,
 	}
 	if opt.ServerPub.IsZero() {
 		if err := c.recvServerKey(); err != nil {
@@ -160,6 +167,9 @@ type clientInfo struct {
 	// CanAckPings is whether the client declares it's able to ack
 	// pings.
 	CanAckPings bool
+
+	// IsProber is whether this client is a prober.
+	IsProber bool `json:",omitempty"`
 }
 
 func (c *Client) sendClientKey() error {
@@ -171,6 +181,7 @@ func (c *Client) sendClientKey() error {
 		Version:     ProtocolVersion,
 		MeshKey:     c.meshKey,
 		CanAckPings: c.canAckPings,
+		IsProber:    c.isProber,
 	})
 	if err != nil {
 		return err

derp/derp_server.go

@@ -20,18 +20,25 @@ import (
 	"io"
 	"io/ioutil"
 	"log"
+	"math"
 	"math/big"
 	"math/rand"
+	"net/http"
 	"os"
+	"os/exec"
 	"runtime"
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"go4.org/mem"
 	"golang.org/x/crypto/nacl/box"
 	"golang.org/x/sync/errgroup"
+	"golang.org/x/time/rate"
+	"inet.af/netaddr"
+	"tailscale.com/client/tailscale"
 	"tailscale.com/disco"
 	"tailscale.com/metrics"
 	"tailscale.com/types/key"
@@ -91,42 +98,46 @@ type Server struct {
 	metaCert    []byte // the encoded x509 cert to send after LetsEncrypt cert+intermediate
 
 	// Counters:
-	_                        [pad32bit]byte
-	packetsSent, bytesSent   expvar.Int
-	packetsRecv, bytesRecv   expvar.Int
-	packetsRecvByKind        metrics.LabelMap
-	packetsRecvDisco         *expvar.Int
-	packetsRecvOther         *expvar.Int
-	_                        [pad32bit]byte
-	packetsDropped           expvar.Int
-	packetsDroppedReason     metrics.LabelMap
-	packetsDroppedUnknown    *expvar.Int // unknown dst pubkey
-	packetsDroppedFwdUnknown *expvar.Int // unknown dst pubkey on forward
-	packetsDroppedGone       *expvar.Int // dst conn shutting down
-	packetsDroppedQueueHead  *expvar.Int // queue full, drop head packet
-	packetsDroppedQueueTail  *expvar.Int // queue full, drop tail packet
-	packetsDroppedWrite      *expvar.Int // error writing to dst conn
-	_                        [pad32bit]byte
-	packetsForwardedOut      expvar.Int
-	packetsForwardedIn       expvar.Int
-	peerGoneFrames           expvar.Int // number of peer gone frames sent
-	accepts                  expvar.Int
-	curClients               expvar.Int
-	curHomeClients           expvar.Int // ones with preferred
-	clientsReplaced          expvar.Int
-	unknownFrames            expvar.Int
-	homeMovesIn              expvar.Int // established clients announce home server moves in
-	homeMovesOut             expvar.Int // established clients announce home server moves out
-	multiForwarderCreated    expvar.Int
-	multiForwarderDeleted    expvar.Int
-	removePktForwardOther    expvar.Int
+	_                            [pad32bit]byte
+	packetsSent, bytesSent       expvar.Int
+	packetsRecv, bytesRecv       expvar.Int
+	packetsRecvByKind            metrics.LabelMap
+	packetsRecvDisco             *expvar.Int
+	packetsRecvOther             *expvar.Int
+	_                            [pad32bit]byte
+	packetsDropped               expvar.Int
+	packetsDroppedReason         metrics.LabelMap
+	packetsDroppedReasonCounters []*expvar.Int // indexed by dropReason
+	packetsDroppedType           metrics.LabelMap
+	packetsDroppedTypeDisco      *expvar.Int
+	packetsDroppedTypeOther      *expvar.Int
+	_                            [pad32bit]byte
+	packetsForwardedOut          expvar.Int
+	packetsForwardedIn           expvar.Int
+	peerGoneFrames               expvar.Int // number of peer gone frames sent
+	accepts                      expvar.Int
+	curClients                   expvar.Int
+	curHomeClients               expvar.Int // ones with preferred
+	clientsReplaced              expvar.Int
+	clientsReplaceLimited        expvar.Int
+	clientsReplaceSleeping       expvar.Int
+	unknownFrames                expvar.Int
+	homeMovesIn                  expvar.Int // established clients announce home server moves in
+	homeMovesOut                 expvar.Int // established clients announce home server moves out
+	multiForwarderCreated        expvar.Int
+	multiForwarderDeleted        expvar.Int
+	removePktForwardOther        expvar.Int
+	avgQueueDuration             *uint64 // In milliseconds; accessed atomically
 
-	mu          sync.Mutex
-	closed      bool
-	netConns    map[Conn]chan struct{} // chan is closed when conn closes
-	clients     map[key.Public]*sclient
-	clientsEver map[key.Public]bool // never deleted from, for stats; fine for now
-	watchers    map[*sclient]bool   // mesh peer -> true
+	// verifyClients only accepts client connections to the DERP server if the clientKey is a
+	// known peer in the network, as specified by a running tailscaled's client's local api.
+	verifyClients bool
+
+	mu       sync.Mutex
+	closed   bool
+	netConns map[Conn]chan struct{} // chan is closed when conn closes
+	clients  map[key.Public]*sclient
+	watchers map[*sclient]bool // mesh peer -> true
 	// clientsMesh tracks all clients in the cluster, both locally
 	// and to mesh peers.  If the value is nil, that means the
 	// peer is only local (and thus in the clients Map, but not
@@ -138,6 +149,9 @@ type Server struct {
 	// because it includes intra-region forwarded packets as the
 	// src.
 	sentTo map[key.Public]map[key.Public]int64 // src => dst => dst's latest sclient.connNum
+
+	// maps from netaddr.IPPort to a client's public key
+	keyOfAddr map[netaddr.IPPort]key.Public
 }
 
 // PacketForwarder is something that can forward packets.
@@ -153,7 +167,7 @@ type PacketForwarder interface {
 // Conn is the subset of the underlying net.Conn the DERP Server needs.
 // It is a defined type so that non-net connections can be used.
 type Conn interface {
-	io.Closer
+	io.WriteCloser
 
 	// The *Deadline methods follow the semantics of net.Conn.
 
@@ -175,23 +189,29 @@ func NewServer(privateKey key.Private, logf logger.Logf) *Server {
 		limitedLogf:          logger.RateLimitedFn(logf, 30*time.Second, 5, 100),
 		packetsRecvByKind:    metrics.LabelMap{Label: "kind"},
 		packetsDroppedReason: metrics.LabelMap{Label: "reason"},
+		packetsDroppedType:   metrics.LabelMap{Label: "type"},
 		clients:              map[key.Public]*sclient{},
-		clientsEver:          map[key.Public]bool{},
 		clientsMesh:          map[key.Public]PacketForwarder{},
 		netConns:             map[Conn]chan struct{}{},
 		memSys0:              ms.Sys,
 		watchers:             map[*sclient]bool{},
 		sentTo:               map[key.Public]map[key.Public]int64{},
+		avgQueueDuration:     new(uint64),
+		keyOfAddr:            map[netaddr.IPPort]key.Public{},
 	}
 	s.initMetacert()
 	s.packetsRecvDisco = s.packetsRecvByKind.Get("disco")
 	s.packetsRecvOther = s.packetsRecvByKind.Get("other")
-	s.packetsDroppedUnknown = s.packetsDroppedReason.Get("unknown_dest")
-	s.packetsDroppedFwdUnknown = s.packetsDroppedReason.Get("unknown_dest_on_fwd")
-	s.packetsDroppedGone = s.packetsDroppedReason.Get("gone")
-	s.packetsDroppedQueueHead = s.packetsDroppedReason.Get("queue_head")
-	s.packetsDroppedQueueTail = s.packetsDroppedReason.Get("queue_tail")
-	s.packetsDroppedWrite = s.packetsDroppedReason.Get("write_error")
+	s.packetsDroppedReasonCounters = []*expvar.Int{
+		s.packetsDroppedReason.Get("unknown_dest"),
+		s.packetsDroppedReason.Get("unknown_dest_on_fwd"),
+		s.packetsDroppedReason.Get("gone"),
+		s.packetsDroppedReason.Get("queue_head"),
+		s.packetsDroppedReason.Get("queue_tail"),
+		s.packetsDroppedReason.Get("write_error"),
+	}
+	s.packetsDroppedTypeDisco = s.packetsDroppedType.Get("disco")
+	s.packetsDroppedTypeOther = s.packetsDroppedType.Get("other")
 	return s
 }
 
@@ -203,6 +223,13 @@ func (s *Server) SetMeshKey(v string) {
 	s.meshKey = v
 }
 
+// SetVerifyClients sets whether this DERP server verifies clients through tailscaled.
+//
+// It must be called before serving begins.
+func (s *Server) SetVerifyClient(v bool) {
+	s.verifyClients = v
+}
+
 // HasMeshKey reports whether the server is configured with a mesh key.
 func (s *Server) HasMeshKey() bool { return s.meshKey != "" }
 
@@ -322,25 +349,40 @@ func (s *Server) initMetacert() {
 func (s *Server) MetaCert() []byte { return s.metaCert }
 
 // registerClient notes that client c is now authenticated and ready for packets.
-// If c's public key was already connected with a different connection, the prior one is closed.
-func (s *Server) registerClient(c *sclient) {
+//
+// If c's public key was already connected with a different
+// connection, the prior one is closed, unless it's fighting rapidly
+// with another client with the same key, in which case the returned
+// ok is false, and the caller should wait the provided duration
+// before trying again.
+func (s *Server) registerClient(c *sclient) (ok bool, d time.Duration) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 	old := s.clients[c.key]
 	if old == nil {
 		c.logf("adding connection")
 	} else {
+		// Take over the old rate limiter, discarding the one
+		// our caller just made.
+		c.replaceLimiter = old.replaceLimiter
+		if rr := c.replaceLimiter.ReserveN(timeNow(), 1); rr.OK() {
+			if d := rr.DelayFrom(timeNow()); d > 0 {
+				s.clientsReplaceLimited.Add(1)
+				return false, d
+			}
+		}
 		s.clientsReplaced.Add(1)
 		c.logf("adding connection, replacing %s", old.remoteAddr)
 		go old.nc.Close()
 	}
 	s.clients[c.key] = c
-	s.clientsEver[c.key] = true
 	if _, ok := s.clientsMesh[c.key]; !ok {
 		s.clientsMesh[c.key] = nil // just for varz of total users in cluster
 	}
+	s.keyOfAddr[c.remoteIPPort] = c.key
 	s.curClients.Add(1)
 	s.broadcastPeerStateChangeLocked(c.key, true)
+	return true, 0
 }
 
 // broadcastPeerStateChangeLocked enqueues a message to all watchers
@@ -373,6 +415,8 @@ func (s *Server) unregisterClient(c *sclient) {
 		delete(s.watchers, c)
 	}
 
+	delete(s.keyOfAddr, c.remoteIPPort)
+
 	s.curClients.Add(-1)
 	if c.preferred {
 		s.curHomeClients.Add(-1)
@@ -426,8 +470,9 @@ func (s *Server) addWatcher(c *sclient) {
 }
 
 func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connNum int64) error {
-	br, bw := brw.Reader, brw.Writer
+	br := brw.Reader
 	nc.SetDeadline(time.Now().Add(10 * time.Second))
+	bw := &lazyBufioWriter{w: nc, lbw: brw.Writer}
 	if err := s.sendServerKey(bw); err != nil {
 		return fmt.Errorf("send server key: %v", err)
 	}
@@ -446,21 +491,32 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 
+	remoteIPPort, _ := netaddr.ParseIPPort(remoteAddr)
+
 	c := &sclient{
-		connNum:     connNum,
-		s:           s,
-		key:         clientKey,
-		nc:          nc,
-		br:          br,
-		bw:          bw,
-		logf:        logger.WithPrefix(s.logf, fmt.Sprintf("derp client %v/%x: ", remoteAddr, clientKey)),
-		done:        ctx.Done(),
-		remoteAddr:  remoteAddr,
-		connectedAt: time.Now(),
-		sendQueue:   make(chan pkt, perClientSendQueueDepth),
-		peerGone:    make(chan key.Public),
-		canMesh:     clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
+		connNum:        connNum,
+		s:              s,
+		key:            clientKey,
+		nc:             nc,
+		br:             br,
+		bw:             bw,
+		logf:           logger.WithPrefix(s.logf, fmt.Sprintf("derp client %v/%x: ", remoteAddr, clientKey)),
+		done:           ctx.Done(),
+		remoteAddr:     remoteAddr,
+		remoteIPPort:   remoteIPPort,
+		connectedAt:    time.Now(),
+		sendQueue:      make(chan pkt, perClientSendQueueDepth),
+		discoSendQueue: make(chan pkt, perClientSendQueueDepth),
+		peerGone:       make(chan key.Public),
+		canMesh:        clientInfo.MeshKey != "" && clientInfo.MeshKey == s.meshKey,
+
+		// Allow kicking out previous connections once a
+		// minute, with a very high burst of 100. Once a
+		// minute is less than the client's 2 minute
+		// inactivity timeout.
+		replaceLimiter: rate.NewLimiter(rate.Every(time.Minute), 100),
 	}
+
 	if c.canMesh {
 		c.meshUpdate = make(chan struct{})
 	}
@@ -468,10 +524,18 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 		c.info = *clientInfo
 	}
 
-	s.registerClient(c)
+	for {
+		ok, d := s.registerClient(c)
+		if ok {
+			break
+		}
+		s.clientsReplaceSleeping.Add(1)
+		timeSleep(d)
+		s.clientsReplaceSleeping.Add(-1)
+	}
 	defer s.unregisterClient(c)
 
-	err = s.sendServerInfo(bw, clientKey)
+	err = s.sendServerInfo(c.bw, clientKey)
 	if err != nil {
 		return fmt.Errorf("send server info: %v", err)
 	}
@@ -479,6 +543,12 @@ func (s *Server) accept(nc Conn, brw *bufio.ReadWriter, remoteAddr string, connN
 	return c.run(ctx)
 }
 
+// for testing
+var (
+	timeSleep = time.Sleep
+	timeNow   = time.Now
+)
+
 // run serves the client until there's an error.
 // If the client hangs up or the server is closed, run returns nil, otherwise run returns an error.
 func (c *sclient) run(ctx context.Context) error {
@@ -602,17 +672,14 @@ func (c *sclient) handleFrameForwardPacket(ft frameType, fl uint32) error {
 	s.mu.Unlock()
 
 	if dst == nil {
-		s.packetsDropped.Add(1)
-		s.packetsDroppedFwdUnknown.Add(1)
-		if debug {
-			c.logf("dropping forwarded packet for unknown %x", dstKey)
-		}
+		s.recordDrop(contents, srcKey, dstKey, dropReasonUnknownDestOnFwd)
 		return nil
 	}
 
 	return c.sendPkt(dst, pkt{
-		bs:  contents,
-		src: srcKey,
+		bs:         contents,
+		enqueuedAt: time.Now(),
+		src:        srcKey,
 	})
 }
 
@@ -656,21 +723,53 @@ func (c *sclient) handleFrameSendPacket(ft frameType, fl uint32) error {
 			}
 			return nil
 		}
-		s.packetsDropped.Add(1)
-		s.packetsDroppedUnknown.Add(1)
-		if debug {
-			c.logf("dropping packet for unknown %x", dstKey)
-		}
+		s.recordDrop(contents, c.key, dstKey, dropReasonUnknownDest)
 		return nil
 	}
 
 	p := pkt{
-		bs:  contents,
-		src: c.key,
+		bs:         contents,
+		enqueuedAt: time.Now(),
+		src:        c.key,
 	}
 	return c.sendPkt(dst, p)
 }
 
+// dropReason is why we dropped a DERP frame.
+type dropReason int
+
+//go:generate go run tailscale.com/cmd/addlicense -year 2021 -file dropreason_string.go go run golang.org/x/tools/cmd/stringer -type=dropReason -trimprefix=dropReason
+
+const (
+	dropReasonUnknownDest      dropReason = iota // unknown destination pubkey
+	dropReasonUnknownDestOnFwd                   // unknown destination pubkey on a derp-forwarded packet
+	dropReasonGone                               // destination tailscaled disconnected before we could send
+	dropReasonQueueHead                          // destination queue is full, dropped packet at queue head
+	dropReasonQueueTail                          // destination queue is full, dropped packet at queue tail
+	dropReasonWriteError                         // OS write() failed
+)
+
+func (s *Server) recordDrop(packetBytes []byte, srcKey, dstKey key.Public, reason dropReason) {
+	s.packetsDropped.Add(1)
+	s.packetsDroppedReasonCounters[reason].Add(1)
+	if disco.LooksLikeDiscoWrapper(packetBytes) {
+		s.packetsDroppedTypeDisco.Add(1)
+	} else {
+		s.packetsDroppedTypeOther.Add(1)
+	}
+	if verboseDropKeys[dstKey] {
+		// Preformat the log string prior to calling limitedLogf. The
+		// limiter acts based on the format string, and we want to
+		// rate-limit per src/dst keys, not on the generic "dropped
+		// stuff" message.
+		msg := fmt.Sprintf("drop (%s) %s -> %s", srcKey.ShortString(), reason, dstKey.ShortString())
+		s.limitedLogf(msg)
+	}
+	if debug {
+		s.logf("dropping packet reason=%s dst=%s disco=%v", reason, dstKey, disco.LooksLikeDiscoWrapper(packetBytes))
+	}
+}
+
 func (c *sclient) sendPkt(dst *sclient, p pkt) error {
 	s := c.s
 	dstKey := dst.key
@@ -678,53 +777,34 @@ func (c *sclient) sendPkt(dst *sclient, p pkt) error {
 	// Attempt to queue for sending up to 3 times. On each attempt, if
 	// the queue is full, try to drop from queue head to prioritize
 	// fresher packets.
+	sendQueue := dst.sendQueue
+	if disco.LooksLikeDiscoWrapper(p.bs) {
+		sendQueue = dst.discoSendQueue
+	}
 	for attempt := 0; attempt < 3; attempt++ {
 		select {
 		case <-dst.done:
-			s.packetsDropped.Add(1)
-			s.packetsDroppedGone.Add(1)
-			if debug {
-				c.logf("dropping packet for shutdown client %x", dstKey)
-			}
+			s.recordDrop(p.bs, c.key, dstKey, dropReasonGone)
 			return nil
 		default:
 		}
 		select {
-		case dst.sendQueue <- p:
+		case sendQueue <- p:
 			return nil
 		default:
 		}
 
 		select {
-		case <-dst.sendQueue:
-			s.packetsDropped.Add(1)
-			s.packetsDroppedQueueHead.Add(1)
-			if verboseDropKeys[dstKey] {
-				// Generate a full string including src and dst, so
-				// the limiter kicks in once per src.
-				msg := fmt.Sprintf("tail drop %s -> %s", p.src.ShortString(), dstKey.ShortString())
-				c.s.limitedLogf(msg)
-			}
-			if debug {
-				c.logf("dropping packet from client %x queue head", dstKey)
-			}
+		case pkt := <-sendQueue:
+			s.recordDrop(pkt.bs, c.key, dstKey, dropReasonQueueHead)
+			c.recordQueueTime(pkt.enqueuedAt)
 		default:
 		}
 	}
 	// Failed to make room for packet. This can happen in a heavily
 	// contended queue with racing writers. Give up and tail-drop in
 	// this case to keep reader unblocked.
-	s.packetsDropped.Add(1)
-	s.packetsDroppedQueueTail.Add(1)
-	if verboseDropKeys[dstKey] {
-		// Generate a full string including src and dst, so
-		// the limiter kicks in once per src.
-		msg := fmt.Sprintf("head drop %s -> %s", p.src.ShortString(), dstKey.ShortString())
-		c.s.limitedLogf(msg)
-	}
-	if debug {
-		c.logf("dropping packet from client %x queue tail", dstKey)
-	}
+	s.recordDrop(p.bs, c.key, dstKey, dropReasonQueueTail)
 
 	return nil
 }
@@ -750,23 +830,37 @@ func (c *sclient) requestMeshUpdate() {
 }
 
 func (s *Server) verifyClient(clientKey key.Public, info *clientInfo) error {
-	// TODO(crawshaw): implement policy constraints on who can use the DERP server
-	// TODO(bradfitz): ... and at what rate.
+	if !s.verifyClients {
+		return nil
+	}
+	status, err := tailscale.Status(context.TODO())
+	if err != nil {
+		return fmt.Errorf("failed to query local tailscaled status: %w", err)
+	}
+	if clientKey == status.Self.PublicKey {
+		return nil
+	}
+	if _, exists := status.Peer[clientKey]; !exists {
+		return fmt.Errorf("client %v not in set of peers", clientKey)
+	}
+	// TODO(bradfitz): add policy for configurable bandwidth rate per client?
 	return nil
 }
 
-func (s *Server) sendServerKey(bw *bufio.Writer) error {
+func (s *Server) sendServerKey(lw *lazyBufioWriter) error {
 	buf := make([]byte, 0, len(magic)+len(s.publicKey))
 	buf = append(buf, magic...)
 	buf = append(buf, s.publicKey[:]...)
-	return writeFrame(bw, frameServerKey, buf)
+	err := writeFrame(lw.bw(), frameServerKey, buf)
+	lw.Flush() // redundant (no-op) flush to release bufio.Writer
+	return err
 }
 
 type serverInfo struct {
 	Version int `json:"version,omitempty"`
 }
 
-func (s *Server) sendServerInfo(bw *bufio.Writer, clientKey key.Public) error {
+func (s *Server) sendServerInfo(bw *lazyBufioWriter, clientKey key.Public) error {
 	var nonce [24]byte
 	if _, err := crand.Read(nonce[:]); err != nil {
 		return err
@@ -777,7 +871,7 @@ func (s *Server) sendServerInfo(bw *bufio.Writer, clientKey key.Public) error {
 	}
 
 	msgbox := box.Seal(nil, msg, &nonce, clientKey.B32(), s.privateKey.B32())
-	if err := writeFrameHeader(bw, frameServerInfo, nonceLen+uint32(len(msgbox))); err != nil {
+	if err := writeFrameHeader(bw.bw(), frameServerInfo, nonceLen+uint32(len(msgbox))); err != nil {
 		return err
 	}
 	if _, err := bw.Write(nonce[:]); err != nil {
@@ -885,18 +979,25 @@ func (s *Server) recvForwardPacket(br *bufio.Reader, frameLen uint32) (srcKey, d
 // (The "s" prefix is to more explicitly distinguish it from Client in derp_client.go)
 type sclient struct {
 	// Static after construction.
-	connNum    int64 // process-wide unique counter, incremented each Accept
-	s          *Server
-	nc         Conn
-	key        key.Public
-	info       clientInfo
-	logf       logger.Logf
-	done       <-chan struct{} // closed when connection closes
-	remoteAddr string          // usually ip:port from net.Conn.RemoteAddr().String()
-	sendQueue  chan pkt        // packets queued to this client; never closed
-	peerGone   chan key.Public // write request that a previous sender has disconnected (not used by mesh peers)
-	meshUpdate chan struct{}   // write request to write peerStateChange
-	canMesh    bool            // clientInfo had correct mesh token for inter-region routing
+	connNum        int64 // process-wide unique counter, incremented each Accept
+	s              *Server
+	nc             Conn
+	key            key.Public
+	info           clientInfo
+	logf           logger.Logf
+	done           <-chan struct{} // closed when connection closes
+	remoteAddr     string          // usually ip:port from net.Conn.RemoteAddr().String()
+	remoteIPPort   netaddr.IPPort  // zero if remoteAddr is not ip:port.
+	sendQueue      chan pkt        // packets queued to this client; never closed
+	discoSendQueue chan pkt        // important packets queued to this client; never closed
+	peerGone       chan key.Public // write request that a previous sender has disconnected (not used by mesh peers)
+	meshUpdate     chan struct{}   // write request to write peerStateChange
+	canMesh        bool            // clientInfo had correct mesh token for inter-region routing
+
+	// replaceLimiter controls how quickly two connections with
+	// the same client key can kick each other off the server by
+	// taking over ownership of a key.
+	replaceLimiter *rate.Limiter
 
 	// Owned by run, not thread-safe.
 	br          *bufio.Reader
@@ -904,7 +1005,7 @@ type sclient struct {
 	preferred   bool
 
 	// Owned by sender, not thread-safe.
-	bw *bufio.Writer
+	bw *lazyBufioWriter
 
 	// Guarded by s.mu
 	//
@@ -927,11 +1028,13 @@ type pkt struct {
 	// src is the who's the sender of the packet.
 	src key.Public
 
+	// enqueuedAt is when a packet was put onto a queue before it was sent,
+	// and is used for reporting metrics on the duration of packets in the queue.
+	enqueuedAt time.Time
+
 	// bs is the data packet bytes.
 	// The memory is owned by pkt.
 	bs []byte
-
-	// TODO(danderson): enqueue time, to measure queue latency?
 }
 
 func (c *sclient) setPreferred(v bool) {
@@ -959,6 +1062,25 @@ func (c *sclient) setPreferred(v bool) {
 	}
 }
 
+// expMovingAverage returns the new moving average given the previous average,
+// a new value, and an alpha decay factor.
+// https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average
+func expMovingAverage(prev, newValue, alpha float64) float64 {
+	return alpha*newValue + (1-alpha)*prev
+}
+
+// recordQueueTime updates the average queue duration metric after a packet has been sent.
+func (c *sclient) recordQueueTime(enqueuedAt time.Time) {
+	elapsed := float64(time.Since(enqueuedAt).Milliseconds())
+	for {
+		old := atomic.LoadUint64(c.s.avgQueueDuration)
+		newAvg := expMovingAverage(math.Float64frombits(old), elapsed, 0.1)
+		if atomic.CompareAndSwapUint64(c.s.avgQueueDuration, old, math.Float64bits(newAvg)) {
+			break
+		}
+	}
+}
+
 func (c *sclient) sendLoop(ctx context.Context) error {
 	defer func() {
 		// If the sender shuts down unilaterally due to an error, close so
@@ -968,12 +1090,10 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 		// Drain the send queue to count dropped packets
 		for {
 			select {
-			case <-c.sendQueue:
-				c.s.packetsDropped.Add(1)
-				c.s.packetsDroppedGone.Add(1)
-				if debug {
-					c.logf("dropping packet for shutdown %x", c.key)
-				}
+			case pkt := <-c.sendQueue:
+				c.s.recordDrop(pkt.bs, pkt.src, c.key, dropReasonGone)
+			case pkt := <-c.discoSendQueue:
+				c.s.recordDrop(pkt.bs, pkt.src, c.key, dropReasonGone)
 			default:
 				return
 			}
@@ -1002,6 +1122,11 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 			continue
 		case msg := <-c.sendQueue:
 			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
+			continue
+		case msg := <-c.discoSendQueue:
+			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
 			continue
 		case <-keepAliveTick.C:
 			werr = c.sendKeepAlive()
@@ -1025,6 +1150,10 @@ func (c *sclient) sendLoop(ctx context.Context) error {
 			continue
 		case msg := <-c.sendQueue:
 			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
+		case msg := <-c.discoSendQueue:
+			werr = c.sendPacket(msg.src, msg.bs)
+			c.recordQueueTime(msg.enqueuedAt)
 		case <-keepAliveTick.C:
 			werr = c.sendKeepAlive()
 		}
@@ -1038,14 +1167,14 @@ func (c *sclient) setWriteDeadline() {
 // sendKeepAlive sends a keep-alive frame, without flushing.
 func (c *sclient) sendKeepAlive() error {
 	c.setWriteDeadline()
-	return writeFrameHeader(c.bw, frameKeepAlive, 0)
+	return writeFrameHeader(c.bw.bw(), frameKeepAlive, 0)
 }
 
 // sendPeerGone sends a peerGone frame, without flushing.
 func (c *sclient) sendPeerGone(peer key.Public) error {
 	c.s.peerGoneFrames.Add(1)
 	c.setWriteDeadline()
-	if err := writeFrameHeader(c.bw, framePeerGone, keyLen); err != nil {
+	if err := writeFrameHeader(c.bw.bw(), framePeerGone, keyLen); err != nil {
 		return err
 	}
 	_, err := c.bw.Write(peer[:])
@@ -1055,7 +1184,7 @@ func (c *sclient) sendPeerGone(peer key.Public) error {
 // sendPeerPresent sends a peerPresent frame, without flushing.
 func (c *sclient) sendPeerPresent(peer key.Public) error {
 	c.setWriteDeadline()
-	if err := writeFrameHeader(c.bw, framePeerPresent, keyLen); err != nil {
+	if err := writeFrameHeader(c.bw.bw(), framePeerPresent, keyLen); err != nil {
 		return err
 	}
 	_, err := c.bw.Write(peer[:])
@@ -1114,11 +1243,7 @@ func (c *sclient) sendPacket(srcKey key.Public, contents []byte) (err error) {
 	defer func() {
 		// Stats update.
 		if err != nil {
-			c.s.packetsDropped.Add(1)
-			c.s.packetsDroppedWrite.Add(1)
-			if debug {
-				c.logf("dropping packet to %x: %v", c.key, err)
-			}
+			c.s.recordDrop(contents, srcKey, c.key, dropReasonWriteError)
 		} else {
 			c.s.packetsSent.Add(1)
 			c.s.bytesSent.Add(int64(len(contents)))
@@ -1132,11 +1257,11 @@ func (c *sclient) sendPacket(srcKey key.Public, contents []byte) (err error) {
 	if withKey {
 		pktLen += len(srcKey)
 	}
-	if err = writeFrameHeader(c.bw, frameRecvPacket, uint32(pktLen)); err != nil {
+	if err = writeFrameHeader(c.bw.bw(), frameRecvPacket, uint32(pktLen)); err != nil {
 		return err
 	}
 	if withKey {
-		err := writePublicKey(c.bw, &srcKey)
+		err := writePublicKey(c.bw.bw(), &srcKey)
 		if err != nil {
 			return err
 		}
@@ -1264,7 +1389,6 @@ func (s *Server) expVarFunc(f func() interface{}) expvar.Func {
 // ExpVar returns an expvar variable suitable for registering with expvar.Publish.
 func (s *Server) ExpVar() expvar.Var {
 	m := new(metrics.Set)
-	m.Set("counter_unique_clients_ever", s.expVarFunc(func() interface{} { return len(s.clientsEver) }))
 	m.Set("gauge_memstats_sys0", expvar.Func(func() interface{} { return int64(s.memSys0) }))
 	m.Set("gauge_watchers", s.expVarFunc(func() interface{} { return len(s.watchers) }))
 	m.Set("gauge_current_connections", &s.curClients)
@@ -1274,10 +1398,13 @@ func (s *Server) ExpVar() expvar.Var {
 	m.Set("gauge_clients_remote", expvar.Func(func() interface{} { return len(s.clientsMesh) - len(s.clients) }))
 	m.Set("accepts", &s.accepts)
 	m.Set("clients_replaced", &s.clientsReplaced)
+	m.Set("clients_replace_limited", &s.clientsReplaceLimited)
+	m.Set("gauge_clients_replace_sleeping", &s.clientsReplaceSleeping)
 	m.Set("bytes_received", &s.bytesRecv)
 	m.Set("bytes_sent", &s.bytesSent)
 	m.Set("packets_dropped", &s.packetsDropped)
 	m.Set("counter_packets_dropped_reason", &s.packetsDroppedReason)
+	m.Set("counter_packets_dropped_type", &s.packetsDroppedType)
 	m.Set("counter_packets_received_kind", &s.packetsRecvByKind)
 	m.Set("packets_sent", &s.packetsSent)
 	m.Set("packets_received", &s.packetsRecv)
@@ -1290,6 +1417,9 @@ func (s *Server) ExpVar() expvar.Var {
 	m.Set("multiforwarder_created", &s.multiForwarderCreated)
 	m.Set("multiforwarder_deleted", &s.multiForwarderDeleted)
 	m.Set("packet_forwarder_delete_other_value", &s.removePktForwardOther)
+	m.Set("average_queue_duration_ms", expvar.Func(func() interface{} {
+		return math.Float64frombits(atomic.LoadUint64(s.avgQueueDuration))
+	}))
 	var expvarVersion expvar.String
 	expvarVersion.Set(version.Long)
 	m.Set("version", &expvarVersion)
@@ -1365,3 +1495,126 @@ func writePublicKey(bw *bufio.Writer, key *key.Public) error {
 	}
 	return nil
 }
+
+const minTimeBetweenLogs = 2 * time.Second
+
+// BytesSentRecv records the number of bytes that have been sent since the last traffic check
+// for a given process, as well as the public key of the process sending those bytes.
+type BytesSentRecv struct {
+	Sent uint64
+	Recv uint64
+	// Key is the public key of the client which sent/received these bytes.
+	Key key.Public
+}
+
+// parseSSOutput parses the output from the specific call to ss in ServeDebugTraffic.
+// Separated out for ease of testing.
+func parseSSOutput(raw string) map[netaddr.IPPort]BytesSentRecv {
+	newState := map[netaddr.IPPort]BytesSentRecv{}
+	// parse every 2 lines and get src and dst ips, and kv pairs
+	lines := strings.Split(raw, "\n")
+	for i := 0; i < len(lines); i += 2 {
+		ipInfo := strings.Fields(strings.TrimSpace(lines[i]))
+		if len(ipInfo) < 5 {
+			continue
+		}
+		src, err := netaddr.ParseIPPort(ipInfo[4])
+		if err != nil {
+			continue
+		}
+		stats := strings.Fields(strings.TrimSpace(lines[i+1]))
+		stat := BytesSentRecv{}
+		for _, s := range stats {
+			if strings.Contains(s, "bytes_sent") {
+				sent, err := strconv.Atoi(s[strings.Index(s, ":")+1:])
+				if err == nil {
+					stat.Sent = uint64(sent)
+				}
+			} else if strings.Contains(s, "bytes_received") {
+				recv, err := strconv.Atoi(s[strings.Index(s, ":")+1:])
+				if err == nil {
+					stat.Recv = uint64(recv)
+				}
+			}
+		}
+		newState[src] = stat
+	}
+	return newState
+}
+
+func (s *Server) ServeDebugTraffic(w http.ResponseWriter, r *http.Request) {
+	prevState := map[netaddr.IPPort]BytesSentRecv{}
+	enc := json.NewEncoder(w)
+	for r.Context().Err() == nil {
+		output, err := exec.Command("ss", "-i", "-H", "-t").Output()
+		if err != nil {
+			fmt.Fprintf(w, "ss failed: %v", err)
+			return
+		}
+		newState := parseSSOutput(string(output))
+		s.mu.Lock()
+		for k, next := range newState {
+			prev := prevState[k]
+			if prev.Sent < next.Sent || prev.Recv < next.Recv {
+				if pkey, ok := s.keyOfAddr[k]; ok {
+					next.Key = pkey
+					if err := enc.Encode(next); err != nil {
+						s.mu.Unlock()
+						return
+					}
+				}
+			}
+		}
+		s.mu.Unlock()
+		prevState = newState
+		if _, err := fmt.Fprintln(w); err != nil {
+			return
+		}
+		if f, ok := w.(http.Flusher); ok {
+			f.Flush()
+		}
+		time.Sleep(minTimeBetweenLogs)
+	}
+}
+
+var bufioWriterPool = &sync.Pool{
+	New: func() interface{} {
+		return bufio.NewWriterSize(ioutil.Discard, 2<<10)
+	},
+}
+
+// lazyBufioWriter is a bufio.Writer-like wrapping writer that lazily
+// allocates its actual bufio.Writer from a sync.Pool, releasing it to
+// the pool upon flush.
+//
+// We do this to reduce memory overhead; most DERP connections are
+// idle and the idle bufio.Writers were 30% of overall memory usage.
+type lazyBufioWriter struct {
+	w   io.Writer     // underlying
+	lbw *bufio.Writer // lazy; nil means it needs an associated buffer
+}
+
+func (w *lazyBufioWriter) bw() *bufio.Writer {
+	if w.lbw == nil {
+		w.lbw = bufioWriterPool.Get().(*bufio.Writer)
+		w.lbw.Reset(w.w)
+	}
+	return w.lbw
+}
+
+func (w *lazyBufioWriter) Available() int { return w.bw().Available() }
+
+func (w *lazyBufioWriter) Write(p []byte) (int, error) { return w.bw().Write(p) }
+
+func (w *lazyBufioWriter) Flush() error {
+	if w.lbw == nil {
+		return nil
+	}
+	err := w.lbw.Flush()
+
+	w.lbw.Reset(ioutil.Discard)
+	bufioWriterPool.Put(w.lbw)
+	w.lbw = nil
+
+	return err
+}

derp/derp_test.go

@@ -23,6 +23,7 @@ import (
 	"testing"
 	"time"
 
+	"golang.org/x/time/rate"
 	"tailscale.com/net/nettest"
 	"tailscale.com/types/key"
 	"tailscale.com/types/logger"
@@ -849,6 +850,136 @@ func TestClientSendPong(t *testing.T) {
 
 }
 
+func TestServerReplaceClients(t *testing.T) {
+	defer func() {
+		timeSleep = time.Sleep
+		timeNow = time.Now
+	}()
+
+	var (
+		mu     sync.Mutex
+		now    = time.Unix(123, 0)
+		sleeps int
+		slept  time.Duration
+	)
+	timeSleep = func(d time.Duration) {
+		mu.Lock()
+		defer mu.Unlock()
+		sleeps++
+		slept += d
+		now = now.Add(d)
+	}
+	timeNow = func() time.Time {
+		mu.Lock()
+		defer mu.Unlock()
+		return now
+	}
+
+	serverPrivateKey := newPrivateKey(t)
+	var logger logger.Logf = logger.Discard
+	const debug = false
+	if debug {
+		logger = t.Logf
+	}
+
+	s := NewServer(serverPrivateKey, logger)
+	defer s.Close()
+
+	priv := newPrivateKey(t)
+
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	connNum := 0
+	connect := func() *Client {
+		connNum++
+		cout, err := net.Dial("tcp", ln.Addr().String())
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		cin, err := ln.Accept()
+		if err != nil {
+			t.Fatal(err)
+		}
+
+		brwServer := bufio.NewReadWriter(bufio.NewReader(cin), bufio.NewWriter(cin))
+		go s.Accept(cin, brwServer, fmt.Sprintf("test-client-%d", connNum))
+
+		brw := bufio.NewReadWriter(bufio.NewReader(cout), bufio.NewWriter(cout))
+		c, err := NewClient(priv, cout, brw, logger)
+		if err != nil {
+			t.Fatalf("client %d: %v", connNum, err)
+		}
+		return c
+	}
+
+	wantVar := func(v *expvar.Int, want int64) {
+		t.Helper()
+		if got := v.Value(); got != want {
+			t.Errorf("got %d; want %d", got, want)
+		}
+	}
+
+	wantClosed := func(c *Client) {
+		t.Helper()
+		for {
+			m, err := c.Recv()
+			if err != nil {
+				t.Logf("got expected error: %v", err)
+				return
+			}
+			switch m.(type) {
+			case ServerInfoMessage:
+				continue
+			default:
+				t.Fatalf("client got %T; wanted an error", m)
+			}
+		}
+	}
+
+	c1 := connect()
+	waitConnect(t, c1)
+	c2 := connect()
+	waitConnect(t, c2)
+	wantVar(&s.clientsReplaced, 1)
+	wantClosed(c1)
+
+	for i := 0; i < 100+5; i++ {
+		c := connect()
+		defer c.nc.Close()
+		if s.clientsReplaceLimited.Value() == 0 && i < 90 {
+			continue
+		}
+		t.Logf("for %d: replaced=%d, limited=%d, sleeping=%d", i,
+			s.clientsReplaced.Value(),
+			s.clientsReplaceLimited.Value(),
+			s.clientsReplaceSleeping.Value(),
+		)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	if sleeps == 0 {
+		t.Errorf("no sleeps")
+	}
+	if slept == 0 {
+		t.Errorf("total sleep duration was 0")
+	}
+}
+
+func TestLimiter(t *testing.T) {
+	rl := rate.NewLimiter(rate.Every(time.Minute), 100)
+	for i := 0; i < 200; i++ {
+		r := rl.Reserve()
+		d := r.Delay()
+		t.Logf("i=%d, allow=%v, d=%v", i, r.OK(), d)
+	}
+}
+
 func BenchmarkSendRecv(b *testing.B) {
 	for _, size := range []int{10, 100, 1000, 10000} {
 		b.Run(fmt.Sprintf("msgsize=%d", size), func(b *testing.B) { benchmarkSendRecvSize(b, size) })
@@ -948,3 +1079,14 @@ func waitConnect(t testing.TB, c *Client) {
 		t.Fatalf("client first Recv was unexpected type %T", v)
 	}
 }
+
+func TestParseSSOutput(t *testing.T) {
+	contents, err := ioutil.ReadFile("testdata/example_ss.txt")
+	if err != nil {
+		t.Errorf("ioutil.Readfile(example_ss.txt) failed: %v", err)
+	}
+	seen := parseSSOutput(string(contents))
+	if len(seen) == 0 {
+		t.Errorf("parseSSOutput expected non-empty map")
+	}
+}

derp/derphttp/derphttp_client.go

@@ -50,9 +50,11 @@ type Client struct {
 	TLSConfig *tls.Config        // optional; nil means default
 	DNSCache  *dnscache.Resolver // optional; nil means no caching
 	MeshKey   string             // optional; for trusted clients
+	IsProber  bool               // optional; for probers to optional declare themselves as such
 
 	privateKey key.Private
 	logf       logger.Logf
+	dialer     func(ctx context.Context, network, addr string) (net.Conn, error)
 
 	// Either url or getRegion is non-nil:
 	url       *url.URL
@@ -130,6 +132,11 @@ func (c *Client) ServerPublicKey() key.Public {
 	return c.serverPubKey
 }
 
+// SelfPublicKey returns our own public key.
+func (c *Client) SelfPublicKey() key.Public {
+	return c.privateKey.Public()
+}
+
 func urlPort(u *url.URL) string {
 	if p := u.Port(); p != "" {
 		return p
@@ -338,6 +345,7 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 		derp.MeshKey(c.MeshKey),
 		derp.ServerPublicKey(serverPub),
 		derp.CanAckPings(c.canAckPings),
+		derp.IsProber(c.IsProber),
 	)
 	if err != nil {
 		return nil, 0, err
@@ -356,14 +364,26 @@ func (c *Client) connect(ctx context.Context, caller string) (client *derp.Clien
 	return c.client, c.connGen, nil
 }
 
+// SetURLDialer sets the dialer to use for dialing URLs.
+// This dialer is only use for clients created with NewClient, not NewRegionClient.
+// If unset or nil, the default dialer is used.
+//
+// The primary use for this is the derper mesh mode to connect to each
+// other over a VPC network.
+func (c *Client) SetURLDialer(dialer func(ctx context.Context, network, addr string) (net.Conn, error)) {
+	c.dialer = dialer
+}
+
 func (c *Client) dialURL(ctx context.Context) (net.Conn, error) {
 	host := c.url.Hostname()
+	if c.dialer != nil {
+		return c.dialer(ctx, "tcp", net.JoinHostPort(host, urlPort(c.url)))
+	}
 	hostOrIP := host
-
 	dialer := netns.NewDialer()
 
 	if c.DNSCache != nil {
-		ip, _, err := c.DNSCache.LookupIP(ctx, host)
+		ip, _, _, err := c.DNSCache.LookupIP(ctx, host)
 		if err == nil {
 			hostOrIP = ip.String()
 		}
@@ -410,9 +430,7 @@ func (c *Client) dialRegion(ctx context.Context, reg *tailcfg.DERPRegion) (net.C
 func (c *Client) tlsClient(nc net.Conn, node *tailcfg.DERPNode) *tls.Conn {
 	tlsConf := tlsdial.Config(c.tlsServerName(node), c.TLSConfig)
 	if node != nil {
-		if node.DERPTestPort != 0 {
-			tlsConf.InsecureSkipVerify = true
-		}
+		tlsConf.InsecureSkipVerify = node.InsecureForTests
 		if node.CertName != "" {
 			tlsdial.SetConfigExpectedCert(tlsConf, node.CertName)
 		}
@@ -511,8 +529,8 @@ func (c *Client) dialNode(ctx context.Context, n *tailcfg.DERPNode) (net.Conn, e
 				dst = n.HostName
 			}
 			port := "443"
-			if n.DERPTestPort != 0 {
-				port = fmt.Sprint(n.DERPTestPort)
+			if n.DERPPort != 0 {
+				port = fmt.Sprint(n.DERPPort)
 			}
 			c, err := c.dialContext(ctx, proto, net.JoinHostPort(dst, port))
 			select {

derp/derpmap/derpmap.go

@@ -1,91 +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 derpmap contains information about Tailscale.com's production DERP nodes.
-//
-// This package is only used by the "tailscale netcheck" command for debugging.
-// In normal operation the Tailscale nodes get this sent to them from the control
-// server.
-//
-// TODO: remove this package and make "tailscale netcheck" get the
-// list from the control server too.
-package derpmap
-
-import (
-	"fmt"
-	"strings"
-
-	"tailscale.com/tailcfg"
-)
-
-func derpNode(suffix, v4, v6 string) *tailcfg.DERPNode {
-	return &tailcfg.DERPNode{
-		Name:     suffix, // updated later
-		RegionID: 0,      // updated later
-		IPv4:     v4,
-		IPv6:     v6,
-	}
-}
-
-func derpRegion(id int, code, name string, nodes ...*tailcfg.DERPNode) *tailcfg.DERPRegion {
-	region := &tailcfg.DERPRegion{
-		RegionID:   id,
-		RegionName: name,
-		RegionCode: code,
-		Nodes:      nodes,
-	}
-	for _, n := range nodes {
-		n.Name = fmt.Sprintf("%d%s", id, n.Name)
-		n.RegionID = id
-		n.HostName = fmt.Sprintf("derp%s.tailscale.com", strings.TrimSuffix(n.Name, "a"))
-	}
-	return region
-}
-
-// Prod returns Tailscale's map of relay servers.
-//
-// This list is only used by cmd/tailscale's netcheck subcommand. In
-// normal operation the Tailscale nodes get this sent to them from the
-// control server.
-//
-// This list is subject to change and should not be relied on.
-func Prod() *tailcfg.DERPMap {
-	return &tailcfg.DERPMap{
-		Regions: map[int]*tailcfg.DERPRegion{
-			1: derpRegion(1, "nyc", "New York City",
-				derpNode("a", "159.89.225.99", "2604:a880:400:d1::828:b001"),
-			),
-			2: derpRegion(2, "sfo", "San Francisco",
-				derpNode("a", "167.172.206.31", "2604:a880:2:d1::c5:7001"),
-			),
-			3: derpRegion(3, "sin", "Singapore",
-				derpNode("a", "68.183.179.66", "2400:6180:0:d1::67d:8001"),
-			),
-			4: derpRegion(4, "fra", "Frankfurt",
-				derpNode("a", "167.172.182.26", "2a03:b0c0:3:e0::36e:9001"),
-			),
-			5: derpRegion(5, "syd", "Sydney",
-				derpNode("a", "103.43.75.49", "2001:19f0:5801:10b7:5400:2ff:feaa:284c"),
-			),
-			6: derpRegion(6, "blr", "Bangalore",
-				derpNode("a", "68.183.90.120", "2400:6180:100:d0::982:d001"),
-			),
-			7: derpRegion(7, "tok", "Tokyo",
-				derpNode("a", "167.179.89.145", "2401:c080:1000:467f:5400:2ff:feee:22aa"),
-			),
-			8: derpRegion(8, "lhr", "London",
-				derpNode("a", "167.71.139.179", "2a03:b0c0:1:e0::3cc:e001"),
-			),
-			9: derpRegion(9, "dfw", "Dallas",
-				derpNode("a", "207.148.3.137", "2001:19f0:6401:1d9c:5400:2ff:feef:bb82"),
-			),
-			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"),
-			),
-		},
-	}
-}

derp/dropreason_string.go

@@ -0,0 +1,32 @@
+// 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.
+
+// Code generated by "stringer -type=dropReason -trimprefix=dropReason"; DO NOT EDIT.
+
+package derp
+
+import "strconv"
+
+func _() {
+	// An "invalid array index" compiler error signifies that the constant values have changed.
+	// Re-run the stringer command to generate them again.
+	var x [1]struct{}
+	_ = x[dropReasonUnknownDest-0]
+	_ = x[dropReasonUnknownDestOnFwd-1]
+	_ = x[dropReasonGone-2]
+	_ = x[dropReasonQueueHead-3]
+	_ = x[dropReasonQueueTail-4]
+	_ = x[dropReasonWriteError-5]
+}
+
+const _dropReason_name = "UnknownDestUnknownDestOnFwdGoneQueueHeadQueueTailWriteError"
+
+var _dropReason_index = [...]uint8{0, 11, 27, 31, 40, 49, 59}
+
+func (i dropReason) String() string {
+	if i < 0 || i >= dropReason(len(_dropReason_index)-1) {
+		return "dropReason(" + strconv.FormatInt(int64(i), 10) + ")"
+	}
+	return _dropReason_name[_dropReason_index[i]:_dropReason_index[i+1]]
+}

derp/testdata/example_ss.txt

@@ -0,0 +1,8 @@
+ESTAB 0      0       10.255.1.11:35238  34.210.105.16:https
+         cubic wscale:7,7 rto:236 rtt:34.14/3.432 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:8 ssthresh:6 bytes_sent:38056577 bytes_retrans:2918 bytes_acked:38053660 bytes_received:6973211 segs_out:165090 segs_in:124227 data_segs_out:78018 data_segs_in:71645 send 2.71Mbps lastsnd:1156 lastrcv:1120 lastack:1120 pacing_rate 3.26Mbps delivery_rate 2.35Mbps delivered:78017 app_limited busy:2586132ms retrans:0/6 dsack_dups:4 reordering:5 reord_seen:15 rcv_rtt:126355 rcv_space:65780 rcv_ssthresh:541928 minrtt:26.632
+ESTAB 0      80     100.79.58.14:ssh    100.95.73.104:58145
+         cubic wscale:6,7 rto:224 rtt:23.051/2.03 ato:172 mss:1228 pmtu:1280 rcvmss:1228 advmss:1228 cwnd:10 ssthresh:94 bytes_sent:1591815 bytes_retrans:944 bytes_acked:1590791 bytes_received:158925 segs_out:8070 segs_in:8858 data_segs_out:7452 data_segs_in:3789 send 4.26Mbps lastsnd:4 lastrcv:4 lastack:4 pacing_rate 8.52Mbps delivery_rate 10.9Mbps delivered:7451 app_limited busy:61656ms unacked:2 retrans:0/10 dsack_dups:10 rcv_rtt:174712 rcv_space:65025 rcv_ssthresh:64296 minrtt:16.186                                   
+ESTAB 0      374     10.255.1.11:43254 167.172.206.31:https
+         cubic wscale:7,7 rto:224 rtt:22.55/1.941 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:6 ssthresh:4 bytes_sent:14594668 bytes_retrans:173314 bytes_acked:14420981 bytes_received:4207111 segs_out:80566 segs_in:70310 data_segs_out:24317 data_segs_in:20365 send 3.08Mbps lastsnd:4 lastrcv:4 lastack:4 pacing_rate 3.7Mbps delivery_rate 3.05Mbps delivered:24111 app_limited busy:184820ms unacked:2 retrans:0/185 dsack_dups:1 reord_seen:3 rcv_rtt:651.262 rcv_space:226657 rcv_ssthresh:1557136 minrtt:10.18           
+ESTAB 0      0       10.255.1.11:33036    3.121.18.47:https
+         cubic wscale:7,7 rto:372 rtt:168.408/2.044 ato:40 mss:1448 pmtu:1500 rcvmss:1448 advmss:1448 cwnd:10 bytes_sent:27500 bytes_acked:27501 bytes_received:1386524 segs_out:10990 segs_in:11037 data_segs_out:303 data_segs_in:3414 send 688kbps lastsnd:125776 lastrcv:9640 lastack:22760 pacing_rate 1.38Mbps delivery_rate 482kbps delivered:304 app_limited busy:43024ms rcv_rtt:3345.12 rcv_space:62431 rcv_ssthresh:760472 minrtt:168.867

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

@@ -3,47 +3,52 @@ 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/frankban/quicktest v1.12.1
-	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/aws/aws-sdk-go v1.38.52
+	github.com/coreos/go-iptables v0.6.0
+	github.com/creack/pty v1.1.9
+	github.com/dave/jennifer v1.4.1
+	github.com/frankban/quicktest v1.13.0
+	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 v0.5.5
-	github.com/goreleaser/nfpm v1.1.10
-	github.com/jsimonetti/rtnetlink v0.0.0-20210212075122-66c871082f2b
+	github.com/go-ole/go-ole v1.2.5
+	github.com/godbus/dbus/v5 v5.0.4
+	github.com/google/go-cmp v0.5.6
+	github.com/google/goexpect v0.0.0-20210430020637-ab937bf7fd6f
+	github.com/google/uuid v1.1.2
+	github.com/goreleaser/nfpm v1.10.3
+	github.com/iancoleman/strcase v0.2.0
+	github.com/jsimonetti/rtnetlink v0.0.0-20210525051524-4cc836578190
 	github.com/kballard/go-shellquote v0.0.0-20180428030007-95032a82bc51
-	github.com/klauspost/compress v1.10.10
-	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/klauspost/compress v1.12.2
+	github.com/mdlayher/netlink v1.4.1
+	github.com/mdlayher/sdnotify v0.0.0-20210228150836-ea3ec207d697
+	github.com/miekg/dns v1.1.42
+	github.com/mitchellh/go-ps v1.0.0
+	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/pkg/sftp v1.13.0
+	github.com/tailscale/certstore v0.0.0-20210528134328-066c94b793d3
 	github.com/tailscale/depaware v0.0.0-20201214215404-77d1e9757027
-	github.com/tailscale/wireguard-go v0.0.0-20210429195722-6cd106ab1339
+	github.com/tailscale/goupnp v1.0.1-0.20210804011211-c64d0f06ea05
+	github.com/tailscale/hujson v0.0.0-20200924210142-dde312d0d6a2
 	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 v0.0.0-20210616213533-5ff15b29337e
+	golang.org/x/net v0.0.0-20210614182718-04defd469f4e
 	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/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
-	inet.af/netstack v0.0.0-20210317161235-a1bf4e56ef22
-	inet.af/peercred v0.0.0-20210302202138-56e694897155
+	golang.org/x/sys v0.0.0-20210616094352-59db8d763f22
+	golang.org/x/term v0.0.0-20210503060354-a79de5458b56
+	golang.org/x/time v0.0.0-20210611083556-38a9dc6acbc6
+	golang.org/x/tools v0.1.2
+	golang.zx2c4.com/wireguard v0.0.0-20210624150102-15b24b6179e0
+	golang.zx2c4.com/wireguard/windows v0.3.16
+	honnef.co/go/tools v0.1.4
+	inet.af/netaddr v0.0.0-20210721214506-ce7a8ad02cc1
+	inet.af/netstack v0.0.0-20210622165351-29b14ebc044e
+	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

hostinfo/hostinfo.go

@@ -0,0 +1,128 @@
+// 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 hostinfo answers questions about the host environment that Tailscale is
+// running on.
+//
+// TODO(bradfitz): move more of control/controlclient/hostinfo_* into this package.
+package hostinfo
+
+import (
+	"io"
+	"os"
+	"runtime"
+	"sync/atomic"
+
+	"go4.org/mem"
+	"tailscale.com/util/lineread"
+)
+
+// EnvType represents a known environment type.
+// The empty string, the default, means unknown.
+type EnvType string
+
+const (
+	KNative         = EnvType("kn")
+	AWSLambda       = EnvType("lm")
+	Heroku          = EnvType("hr")
+	AzureAppService = EnvType("az")
+	AWSFargate      = EnvType("fg")
+)
+
+var envType atomic.Value // of EnvType
+
+func GetEnvType() EnvType {
+	if e, ok := envType.Load().(EnvType); ok {
+		return e
+	}
+	e := getEnvType()
+	envType.Store(e)
+	return e
+}
+
+func getEnvType() EnvType {
+	if inKnative() {
+		return KNative
+	}
+	if inAWSLambda() {
+		return AWSLambda
+	}
+	if inHerokuDyno() {
+		return Heroku
+	}
+	if inAzureAppService() {
+		return AzureAppService
+	}
+	if inAWSFargate() {
+		return AWSFargate
+	}
+	return ""
+}
+
+// InContainer reports whether we're running in a container.
+func InContainer() bool {
+	if runtime.GOOS != "linux" {
+		return false
+	}
+	var ret bool
+	lineread.File("/proc/1/cgroup", func(line []byte) error {
+		if mem.Contains(mem.B(line), mem.S("/docker/")) ||
+			mem.Contains(mem.B(line), mem.S("/lxc/")) {
+			ret = true
+			return io.EOF // arbitrary non-nil error to stop loop
+		}
+		return nil
+	})
+	lineread.File("/proc/mounts", func(line []byte) error {
+		if mem.Contains(mem.B(line), mem.S("fuse.lxcfs")) {
+			ret = true
+			return io.EOF
+		}
+		return nil
+	})
+	return ret
+}
+
+func inKnative() bool {
+	// https://cloud.google.com/run/docs/reference/container-contract#env-vars
+	if os.Getenv("K_REVISION") != "" && os.Getenv("K_CONFIGURATION") != "" &&
+		os.Getenv("K_SERVICE") != "" && os.Getenv("PORT") != "" {
+		return true
+	}
+	return false
+}
+
+func inAWSLambda() bool {
+	// https://docs.aws.amazon.com/lambda/latest/dg/configuration-envvars.html
+	if os.Getenv("AWS_LAMBDA_FUNCTION_NAME") != "" &&
+		os.Getenv("AWS_LAMBDA_FUNCTION_VERSION") != "" &&
+		os.Getenv("AWS_LAMBDA_INITIALIZATION_TYPE") != "" &&
+		os.Getenv("AWS_LAMBDA_RUNTIME_API") != "" {
+		return true
+	}
+	return false
+}
+
+func inHerokuDyno() bool {
+	// https://devcenter.heroku.com/articles/dynos#local-environment-variables
+	if os.Getenv("PORT") != "" && os.Getenv("DYNO") != "" {
+		return true
+	}
+	return false
+}
+
+func inAzureAppService() bool {
+	if os.Getenv("APPSVC_RUN_ZIP") != "" && os.Getenv("WEBSITE_STACK") != "" &&
+		os.Getenv("WEBSITE_AUTH_AUTO_AAD") != "" {
+		return true
+	}
+	return false
+}
+
+func inAWSFargate() bool {
+	if os.Getenv("AWS_EXECUTION_ENV") == "AWS_ECS_FARGATE" {
+		return true
+	}
+	return false
+}

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",
-		},
-	}
-}

internal/deepprint/fmtsort.go

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

ipn/handle.go

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

ipn/ipnlocal/local.go

@@ -29,7 +29,6 @@ import (
 	"tailscale.com/client/tailscale/apitype"
 	"tailscale.com/control/controlclient"
 	"tailscale.com/health"
-	"tailscale.com/internal/deepprint"
 	"tailscale.com/ipn"
 	"tailscale.com/ipn/ipnstate"
 	"tailscale.com/ipn/policy"
@@ -44,11 +43,14 @@ import (
 	"tailscale.com/types/logger"
 	"tailscale.com/types/netmap"
 	"tailscale.com/types/persist"
+	"tailscale.com/types/preftype"
 	"tailscale.com/types/wgkey"
+	"tailscale.com/util/deephash"
 	"tailscale.com/util/dnsname"
 	"tailscale.com/util/osshare"
 	"tailscale.com/util/systemd"
 	"tailscale.com/version"
+	"tailscale.com/version/distro"
 	"tailscale.com/wgengine"
 	"tailscale.com/wgengine/filter"
 	"tailscale.com/wgengine/router"
@@ -93,7 +95,7 @@ type LocalBackend struct {
 	serverURL             string           // tailcontrol URL
 	newDecompressor       func() (controlclient.Decompressor, error)
 
-	filterHash string
+	filterHash deephash.Sum
 
 	// The mutex protects the following elements.
 	mu             sync.Mutex
@@ -177,6 +179,7 @@ func NewLocalBackend(logf logger.Logf, logid string, store ipn.StateStore, e wge
 		gotPortPollRes: make(chan struct{}),
 	}
 	b.statusChanged = sync.NewCond(&b.statusLock)
+	b.e.SetStatusCallback(b.setWgengineStatus)
 
 	linkMon := e.GetLinkMonitor()
 	b.prevIfState = linkMon.InterfaceState()
@@ -212,6 +215,15 @@ func (b *LocalBackend) SetDirectFileRoot(dir string) {
 	b.directFileRoot = dir
 }
 
+// b.mu must be held.
+func (b *LocalBackend) maybePauseControlClientLocked() {
+	if b.cc == nil {
+		return
+	}
+	networkUp := b.prevIfState.AnyInterfaceUp()
+	b.cc.SetPaused((b.state == ipn.Stopped && b.netMap != nil) || !networkUp)
+}
+
 // linkChange is our link monitor callback, called whenever the network changes.
 // major is whether ifst is different than earlier.
 func (b *LocalBackend) linkChange(major bool, ifst *interfaces.State) {
@@ -220,11 +232,7 @@ func (b *LocalBackend) linkChange(major bool, ifst *interfaces.State) {
 
 	hadPAC := b.prevIfState.HasPAC()
 	b.prevIfState = ifst
-
-	networkUp := ifst.AnyInterfaceUp()
-	if b.cc != nil {
-		go b.cc.SetPaused((b.state == ipn.Stopped && b.netMap != nil) || !networkUp)
-	}
+	b.maybePauseControlClientLocked()
 
 	// If the PAC-ness of the network changed, reconfig wireguard+route to
 	// add/remove subnets.
@@ -242,10 +250,10 @@ func (b *LocalBackend) linkChange(major bool, ifst *interfaces.State) {
 	// need updating to tweak default routes.
 	b.updateFilter(b.netMap, b.prefs)
 
-	if runtime.GOOS == "windows" && b.netMap != nil && b.state == ipn.Running {
+	if peerAPIListenAsync && b.netMap != nil && b.state == ipn.Running {
 		want := len(b.netMap.Addresses)
-		b.logf("linkChange: peerAPIListeners too low; trying again")
 		if len(b.peerAPIListeners) < want {
+			b.logf("linkChange: peerAPIListeners too low; trying again")
 			go b.initPeerAPIListener()
 		}
 	}
@@ -323,9 +331,13 @@ func (b *LocalBackend) updateStatus(sb *ipnstate.StatusBuilder, extraLocked func
 		s.AuthURL = b.authURLSticky
 		if b.netMap != nil {
 			s.MagicDNSSuffix = b.netMap.MagicDNSSuffix()
+			s.CertDomains = append([]string(nil), b.netMap.DNS.CertDomains...)
 		}
 	})
 	sb.MutateSelfStatus(func(ss *ipnstate.PeerStatus) {
+		if b.netMap != nil && b.netMap.SelfNode != nil {
+			ss.ID = b.netMap.SelfNode.StableID
+		}
 		for _, pln := range b.peerAPIListeners {
 			ss.PeerAPIURL = append(ss.PeerAPIURL, pln.urlStr)
 		}
@@ -353,18 +365,19 @@ func (b *LocalBackend) populatePeerStatusLocked(sb *ipnstate.StatusBuilder) {
 		var tailAddr4 string
 		var tailscaleIPs = make([]netaddr.IP, 0, len(p.Addresses))
 		for _, addr := range p.Addresses {
-			if addr.IsSingleIP() && tsaddr.IsTailscaleIP(addr.IP) {
-				if addr.IP.Is4() && tailAddr4 == "" {
+			if addr.IsSingleIP() && tsaddr.IsTailscaleIP(addr.IP()) {
+				if addr.IP().Is4() && tailAddr4 == "" {
 					// The peer struct previously only allowed a single
 					// Tailscale IP address. For compatibility for a few releases starting
 					// with 1.8, keep it pulled out as IPv4-only for a bit.
-					tailAddr4 = addr.IP.String()
+					tailAddr4 = addr.IP().String()
 				}
-				tailscaleIPs = append(tailscaleIPs, addr.IP)
+				tailscaleIPs = append(tailscaleIPs, addr.IP())
 			}
 		}
 		sb.AddPeer(key.Public(p.Key), &ipnstate.PeerStatus{
 			InNetworkMap:       true,
+			ID:                 p.StableID,
 			UserID:             p.User,
 			TailAddrDeprecated: tailAddr4,
 			TailscaleIPs:       tailscaleIPs,
@@ -386,10 +399,10 @@ func (b *LocalBackend) populatePeerStatusLocked(sb *ipnstate.StatusBuilder) {
 func (b *LocalBackend) WhoIs(ipp netaddr.IPPort) (n *tailcfg.Node, u tailcfg.UserProfile, ok bool) {
 	b.mu.Lock()
 	defer b.mu.Unlock()
-	n, ok = b.nodeByAddr[ipp.IP]
+	n, ok = b.nodeByAddr[ipp.IP()]
 	if !ok {
 		var ip netaddr.IP
-		if ipp.Port != 0 {
+		if ipp.Port() != 0 {
 			ip, ok = b.e.WhoIsIPPort(ipp)
 		}
 		if !ok {
@@ -430,14 +443,15 @@ func (b *LocalBackend) setClientStatus(st controlclient.Status) {
 		}
 		return
 	}
-	if st.LoginFinished != nil {
+
+	b.mu.Lock()
+	wasBlocked := b.blocked
+	b.mu.Unlock()
+
+	if st.LoginFinished != nil && wasBlocked {
 		// Auth completed, unblock the engine
 		b.blockEngineUpdates(false)
 		b.authReconfig()
-		b.EditPrefs(&ipn.MaskedPrefs{
-			LoggedOutSet: true,
-			Prefs:        ipn.Prefs{LoggedOut: false},
-		})
 		b.send(ipn.Notify{LoginFinished: &empty.Message{}})
 	}
 
@@ -446,6 +460,13 @@ func (b *LocalBackend) setClientStatus(st controlclient.Status) {
 	// Lock b once and do only the things that require locking.
 	b.mu.Lock()
 
+	if st.LogoutFinished != nil {
+		// Since we're logged out now, our netmap cache is invalid.
+		// Since st.NetMap==nil means "netmap is unchanged", there is
+		// no other way to represent this change.
+		b.setNetMapLocked(nil)
+	}
+
 	prefs := b.prefs
 	stateKey := b.stateKey
 	netMap := b.netMap
@@ -476,11 +497,15 @@ func (b *LocalBackend) setClientStatus(st controlclient.Status) {
 		b.authURL = st.URL
 		b.authURLSticky = st.URL
 	}
-	if b.state == ipn.NeedsLogin {
-		if !b.prefs.WantRunning {
+	if wasBlocked && st.LoginFinished != nil {
+		// Interactive login finished successfully (URL visited).
+		// After an interactive login, the user always wants
+		// WantRunning.
+		if !b.prefs.WantRunning || b.prefs.LoggedOut {
 			prefsChanged = true
 		}
 		b.prefs.WantRunning = true
+		b.prefs.LoggedOut = false
 	}
 	// Prefs will be written out; this is not safe unless locked or cloned.
 	if prefsChanged {
@@ -543,7 +568,7 @@ func (b *LocalBackend) findExitNodeIDLocked(nm *netmap.NetworkMap) (prefsChanged
 
 	for _, peer := range nm.Peers {
 		for _, addr := range peer.Addresses {
-			if !addr.IsSingleIP() || addr.IP != b.prefs.ExitNodeIP {
+			if !addr.IsSingleIP() || addr.IP() != b.prefs.ExitNodeIP {
 				continue
 			}
 			// Found the node being referenced, upgrade prefs to
@@ -562,10 +587,18 @@ func (b *LocalBackend) findExitNodeIDLocked(nm *netmap.NetworkMap) (prefsChanged
 func (b *LocalBackend) setWgengineStatus(s *wgengine.Status, err error) {
 	if err != nil {
 		b.logf("wgengine status error: %v", err)
+
+		b.statusLock.Lock()
+		b.statusChanged.Broadcast()
+		b.statusLock.Unlock()
 		return
 	}
 	if s == nil {
 		b.logf("[unexpected] non-error wgengine update with status=nil: %v", s)
+
+		b.statusLock.Lock()
+		b.statusChanged.Broadcast()
+		b.statusLock.Unlock()
 		return
 	}
 
@@ -578,8 +611,8 @@ func (b *LocalBackend) setWgengineStatus(s *wgengine.Status, err error) {
 
 	if cc != nil {
 		cc.UpdateEndpoints(0, s.LocalAddrs)
+		b.stateMachine()
 	}
-	b.stateMachine()
 
 	b.statusLock.Lock()
 	b.statusChanged.Broadcast()
@@ -631,6 +664,12 @@ func (b *LocalBackend) getNewControlClientFunc() clientGen {
 // startIsNoopLocked reports whether a Start call on this LocalBackend
 // with the provided Start Options would be a useless no-op.
 //
+// TODO(apenwarr): we shouldn't need this.
+//  The state machine is now nearly clean enough where it can accept a new
+//  connection while in any state, not just Running, and on any platform.
+//  We'd want to add a few more tests to state_test.go to ensure this continues
+//  to work as expected.
+//
 // b.mu must be held.
 func (b *LocalBackend) startIsNoopLocked(opts ipn.Options) bool {
 	// Options has 5 fields; check all of them:
@@ -684,6 +723,7 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 		b.send(ipn.Notify{
 			State:         &state,
 			NetMap:        nm,
+			Prefs:         b.prefs,
 			LoginFinished: new(empty.Message),
 		})
 		return nil
@@ -723,6 +763,12 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 		newPrefs := opts.UpdatePrefs
 		newPrefs.Persist = b.prefs.Persist
 		b.prefs = newPrefs
+
+		if opts.StateKey != "" {
+			if err := b.store.WriteState(opts.StateKey, b.prefs.ToBytes()); err != nil {
+				b.logf("failed to save UpdatePrefs state: %v", err)
+			}
+		}
 	}
 
 	wantRunning := b.prefs.WantRunning
@@ -823,7 +869,6 @@ func (b *LocalBackend) Start(opts ipn.Options) error {
 	}
 
 	cc.SetStatusFunc(b.setClientStatus)
-	b.e.SetStatusCallback(b.setWgengineStatus)
 	b.e.SetNetInfoCallback(b.setNetInfo)
 
 	b.mu.Lock()
@@ -874,7 +919,7 @@ func (b *LocalBackend) updateFilter(netMap *netmap.NetworkMap, prefs *ipn.Prefs)
 	}
 	if prefs != nil {
 		for _, r := range prefs.AdvertiseRoutes {
-			if r.Bits == 0 {
+			if r.Bits() == 0 {
 				// When offering a default route to the world, we
 				// filter out locally reachable LANs, so that the
 				// default route effectively appears to be a "guest
@@ -896,10 +941,10 @@ func (b *LocalBackend) updateFilter(netMap *netmap.NetworkMap, prefs *ipn.Prefs)
 			}
 		}
 	}
-	localNets := localNetsB.IPSet()
-	logNets := logNetsB.IPSet()
+	localNets, _ := localNetsB.IPSet()
+	logNets, _ := logNetsB.IPSet()
 
-	changed := deepprint.UpdateHash(&b.filterHash, haveNetmap, addrs, packetFilter, localNets.Ranges(), logNets.Ranges(), shieldsUp)
+	changed := deephash.Update(&b.filterHash, haveNetmap, addrs, packetFilter, localNets.Ranges(), logNets.Ranges(), shieldsUp)
 	if !changed {
 		return
 	}
@@ -939,22 +984,61 @@ var removeFromDefaultRoute = []netaddr.IPPrefix{
 	tsaddr.TailscaleULARange(),
 }
 
-func interfaceRoutes() (ips *netaddr.IPSet, hostIPs []netaddr.IP, err error) {
-	var b netaddr.IPSetBuilder
-	if err := interfaces.ForeachInterfaceAddress(func(_ interfaces.Interface, pfx netaddr.IPPrefix) {
-		if tsaddr.IsTailscaleIP(pfx.IP) {
+// internalAndExternalInterfaces splits interface routes into "internal"
+// and "external" sets. Internal routes are those of virtual ethernet
+// network interfaces used by guest VMs and containers, such as WSL and
+// Docker.
+//
+// Given that "internal" routes don't leave the device, we choose to
+// trust them more, allowing access to them when an Exit Node is enabled.
+func internalAndExternalInterfaces() (internal, external []netaddr.IPPrefix, err error) {
+	if err := interfaces.ForeachInterfaceAddress(func(iface interfaces.Interface, pfx netaddr.IPPrefix) {
+		if tsaddr.IsTailscaleIP(pfx.IP()) {
 			return
 		}
 		if pfx.IsSingleIP() {
 			return
 		}
-		hostIPs = append(hostIPs, pfx.IP)
+		if runtime.GOOS == "windows" {
+			// Windows Hyper-V prefixes all MAC addresses with 00:15:5d.
+			// https://docs.microsoft.com/en-us/troubleshoot/windows-server/virtualization/default-limit-256-dynamic-mac-addresses
+			//
+			// This includes WSL2 vEthernet.
+			// Importantly: by default WSL2 /etc/resolv.conf points to
+			// a stub resolver running on the host vEthernet IP.
+			// So enabling exit nodes with the default tailnet
+			// configuration breaks WSL2 DNS without this.
+			mac := iface.Interface.HardwareAddr
+			if len(mac) == 6 && mac[0] == 0x00 && mac[1] == 0x15 && mac[2] == 0x5d {
+				internal = append(internal, pfx)
+				return
+			}
+		}
+		external = append(external, pfx)
+	}); err != nil {
+		return nil, nil, err
+	}
+
+	return internal, external, nil
+}
+
+func interfaceRoutes() (ips *netaddr.IPSet, hostIPs []netaddr.IP, err error) {
+	var b netaddr.IPSetBuilder
+	if err := interfaces.ForeachInterfaceAddress(func(_ interfaces.Interface, pfx netaddr.IPPrefix) {
+		if tsaddr.IsTailscaleIP(pfx.IP()) {
+			return
+		}
+		if pfx.IsSingleIP() {
+			return
+		}
+		hostIPs = append(hostIPs, pfx.IP())
 		b.AddPrefix(pfx)
 	}); err != nil {
 		return nil, nil, err
 	}
 
-	return b.IPSet(), hostIPs, nil
+	ipSet, _ := b.IPSet()
+	return ipSet, hostIPs, nil
 }
 
 // shrinkDefaultRoute returns an IPSet representing the IPs in route,
@@ -985,7 +1069,7 @@ func shrinkDefaultRoute(route netaddr.IPPrefix) (*netaddr.IPSet, error) {
 	for _, pfx := range removeFromDefaultRoute {
 		b.RemovePrefix(pfx)
 	}
-	return b.IPSet(), nil
+	return b.IPSet()
 }
 
 // dnsCIDRsEqual determines whether two CIDR lists are equal
@@ -1677,9 +1761,64 @@ func (b *LocalBackend) authReconfig() {
 
 	rcfg := b.routerConfig(cfg, uc)
 
-	var dcfg dns.Config
+	dcfg := dns.Config{
+		Routes: map[dnsname.FQDN][]netaddr.IPPort{},
+		Hosts:  map[dnsname.FQDN][]netaddr.IP{},
+	}
+
+	// Populate MagicDNS records. We do this unconditionally so that
+	// quad-100 can always respond to MagicDNS queries, even if the OS
+	// isn't configured to make MagicDNS resolution truly
+	// magic. Details in
+	// https://github.com/tailscale/tailscale/issues/1886.
+	set := func(name string, addrs []netaddr.IPPrefix) {
+		if len(addrs) == 0 || name == "" {
+			return
+		}
+		fqdn, err := dnsname.ToFQDN(name)
+		if err != nil {
+			return // TODO: propagate error?
+		}
+		var ips []netaddr.IP
+		for _, addr := range addrs {
+			// Remove IPv6 addresses for now, as we don't
+			// guarantee that the peer node actually can speak
+			// IPv6 correctly.
+			//
+			// https://github.com/tailscale/tailscale/issues/1152
+			// tracks adding the right capability reporting to
+			// enable AAAA in MagicDNS.
+			if addr.IP().Is6() {
+				continue
+			}
+			ips = append(ips, addr.IP())
+		}
+		dcfg.Hosts[fqdn] = ips
+	}
+	set(nm.Name, nm.Addresses)
+	for _, peer := range nm.Peers {
+		set(peer.Name, peer.Addresses)
+	}
+	for _, rec := range nm.DNS.ExtraRecords {
+		switch rec.Type {
+		case "", "A", "AAAA":
+			// Treat these all the same for now: infer from the value
+		default:
+			// TODO: more
+			continue
+		}
+		ip, err := netaddr.ParseIP(rec.Value)
+		if err != nil {
+			// Ignore.
+			continue
+		}
+		fqdn, err := dnsname.ToFQDN(rec.Name)
+		if err != nil {
+			continue
+		}
+		dcfg.Hosts[fqdn] = append(dcfg.Hosts[fqdn], ip)
+	}
 
-	// If CorpDNS is false, dcfg remains the zero value.
 	if uc.CorpDNS {
 		addDefault := func(resolvers []tailcfg.DNSResolver) {
 			for _, resolver := range resolvers {
@@ -1693,9 +1832,6 @@ func (b *LocalBackend) authReconfig() {
 		}
 
 		addDefault(nm.DNS.Resolvers)
-		if len(nm.DNS.Routes) > 0 {
-			dcfg.Routes = map[dnsname.FQDN][]netaddr.IPPort{}
-		}
 		for suffix, resolvers := range nm.DNS.Routes {
 			fqdn, err := dnsname.ToFQDN(suffix)
 			if err != nil {
@@ -1717,36 +1853,9 @@ func (b *LocalBackend) authReconfig() {
 			}
 			dcfg.SearchDomains = append(dcfg.SearchDomains, fqdn)
 		}
-		set := func(name string, addrs []netaddr.IPPrefix) {
-			if len(addrs) == 0 || name == "" {
-				return
-			}
-			fqdn, err := dnsname.ToFQDN(name)
-			if err != nil {
-				return // TODO: propagate error?
-			}
-			var ips []netaddr.IP
-			for _, addr := range addrs {
-				// Remove IPv6 addresses for now, as we don't
-				// guarantee that the peer node actually can speak
-				// IPv6 correctly.
-				//
-				// https://github.com/tailscale/tailscale/issues/1152
-				// tracks adding the right capability reporting to
-				// enable AAAA in MagicDNS.
-				if addr.IP.Is6() {
-					continue
-				}
-				ips = append(ips, addr.IP)
-			}
-			dcfg.Hosts[fqdn] = ips
-		}
 		if nm.DNS.Proxied { // actually means "enable MagicDNS"
-			dcfg.AuthoritativeSuffixes = magicDNSRootDomains(nm)
-			dcfg.Hosts = map[dnsname.FQDN][]netaddr.IP{}
-			set(nm.Name, nm.Addresses)
-			for _, peer := range nm.Peers {
-				set(peer.Name, peer.Addresses)
+			for _, dom := range magicDNSRootDomains(nm) {
+				dcfg.Routes[dom] = nil // resolve internally with dcfg.Hosts
 			}
 		}
 
@@ -1770,7 +1879,7 @@ func (b *LocalBackend) authReconfig() {
 			//
 			// https://github.com/tailscale/tailscale/issues/1713
 			addDefault(nm.DNS.FallbackResolvers)
-		case len(dcfg.Routes) == 0 && len(dcfg.Hosts) == 0 && len(dcfg.AuthoritativeSuffixes) == 0:
+		case len(dcfg.Routes) == 0:
 			// No settings requiring split DNS, no problem.
 		case version.OS() == "android":
 			// We don't support split DNS at all on Android yet.
@@ -1778,7 +1887,7 @@ func (b *LocalBackend) authReconfig() {
 		}
 	}
 
-	err = b.e.Reconfig(cfg, rcfg, &dcfg)
+	err = b.e.Reconfig(cfg, rcfg, &dcfg, nm.Debug)
 	if err == wgengine.ErrNoChanges {
 		return
 	}
@@ -1792,17 +1901,15 @@ func parseResolver(cfg tailcfg.DNSResolver) (netaddr.IPPort, error) {
 	if err != nil {
 		return netaddr.IPPort{}, fmt.Errorf("[unexpected] non-IP resolver %q", cfg.Addr)
 	}
-	return netaddr.IPPort{
-		IP:   ip,
-		Port: 53,
-	}, nil
+	return netaddr.IPPortFrom(ip, 53), nil
 }
 
 // tailscaleVarRoot returns the root directory of Tailscale's writable
 // storage area. (e.g. "/var/lib/tailscale")
 func tailscaleVarRoot() string {
-	if runtime.GOOS == "ios" {
-		dir, _ := paths.IOSSharedDir.Load().(string)
+	switch runtime.GOOS {
+	case "ios", "android":
+		dir, _ := paths.AppSharedDir.Load().(string)
 		return dir
 	}
 	stateFile := paths.DefaultTailscaledStateFile()
@@ -1846,14 +1953,30 @@ func (b *LocalBackend) closePeerAPIListenersLocked() {
 	b.peerAPIListeners = nil
 }
 
+// peerAPIListenAsync is whether the operating system requires that we
+// retry listening on the peerAPI ip/port for whatever reason.
+//
+// On Windows, see Issue 1620.
+// On Android, see Issue 1960.
+const peerAPIListenAsync = runtime.GOOS == "windows" || runtime.GOOS == "android"
+
 func (b *LocalBackend) initPeerAPIListener() {
 	b.mu.Lock()
 	defer b.mu.Unlock()
 
+	if b.netMap == nil {
+		// We're called from authReconfig which checks that
+		// netMap is non-nil, but if a concurrent Logout,
+		// ResetForClientDisconnect, or Start happens when its
+		// mutex was released, the netMap could be
+		// nil'ed out (Issue 1996). Bail out early here if so.
+		return
+	}
+
 	if len(b.netMap.Addresses) == len(b.peerAPIListeners) {
 		allSame := true
 		for i, pln := range b.peerAPIListeners {
-			if pln.ip != b.netMap.Addresses[i].IP {
+			if pln.ip != b.netMap.Addresses[i].IP() {
 				allSame = false
 				break
 			}
@@ -1898,21 +2021,20 @@ func (b *LocalBackend) initPeerAPIListener() {
 		var err error
 		skipListen := i > 0 && isNetstack
 		if !skipListen {
-			ln, err = ps.listen(a.IP, b.prevIfState)
+			ln, err = ps.listen(a.IP(), b.prevIfState)
 			if err != nil {
-				if runtime.GOOS == "windows" {
-					// Expected for now. See Issue 1620.
-					// But we fix it later in linkChange
+				if peerAPIListenAsync {
+					// Expected. But we fix it later in linkChange
 					// ("peerAPIListeners too low").
 					continue
 				}
-				b.logf("[unexpected] peerapi listen(%q) error: %v", a.IP, err)
+				b.logf("[unexpected] peerapi listen(%q) error: %v", a.IP(), err)
 				continue
 			}
 		}
 		pln := &peerAPIListener{
 			ps: ps,
-			ip: a.IP,
+			ip: a.IP(),
 			ln: ln, // nil for 2nd+ on netstack
 			lb: b,
 		}
@@ -1921,7 +2043,7 @@ func (b *LocalBackend) initPeerAPIListener() {
 		} else {
 			pln.port = ln.Addr().(*net.TCPAddr).Port
 		}
-		pln.urlStr = "http://" + net.JoinHostPort(a.IP.String(), strconv.Itoa(pln.port))
+		pln.urlStr = "http://" + net.JoinHostPort(a.IP().String(), strconv.Itoa(pln.port))
 		b.logf("peerapi: serving on %s", pln.urlStr)
 		go pln.serve()
 		b.peerAPIListeners = append(b.peerAPIListeners, pln)
@@ -1972,14 +2094,14 @@ func peerRoutes(peers []wgcfg.Peer, cgnatThreshold int) (routes []netaddr.IPPref
 		for _, aip := range peer.AllowedIPs {
 			aip = unmapIPPrefix(aip)
 			// Only add the Tailscale IPv6 ULA once, if we see anybody using part of it.
-			if aip.IP.Is6() && aip.IsSingleIP() && tsULA.Contains(aip.IP) {
+			if aip.IP().Is6() && aip.IsSingleIP() && tsULA.Contains(aip.IP()) {
 				if !didULA {
 					didULA = true
 					routes = append(routes, tsULA)
 				}
 				continue
 			}
-			if aip.IsSingleIP() && cgNAT.Contains(aip.IP) {
+			if aip.IsSingleIP() && cgNAT.Contains(aip.IP()) {
 				cgNATIPs = append(cgNATIPs, aip)
 			} else {
 				routes = append(routes, aip)
@@ -2005,6 +2127,11 @@ func (b *LocalBackend) routerConfig(cfg *wgcfg.Config, prefs *ipn.Prefs) *router
 		Routes:           peerRoutes(cfg.Peers, 10_000),
 	}
 
+	if distro.Get() == distro.Synology {
+		// Issue 1995: we don't use iptables on Synology.
+		rs.NetfilterMode = preftype.NetfilterOff
+	}
+
 	// Sanity check: we expect the control server to program both a v4
 	// and a v6 default route, if default routing is on. Fill in
 	// blackhole routes appropriately if we're missing some. This is
@@ -2030,32 +2157,32 @@ func (b *LocalBackend) routerConfig(cfg *wgcfg.Config, prefs *ipn.Prefs) *router
 		if !default6 {
 			rs.Routes = append(rs.Routes, ipv6Default)
 		}
-		if runtime.GOOS == "linux" {
-			// Only allow local lan access on linux machines for now.
-			ips, _, err := interfaceRoutes()
-			if err != nil {
-				b.logf("failed to discover interface ips: %v", err)
-			}
+		internalIPs, externalIPs, err := internalAndExternalInterfaces()
+		if err != nil {
+			b.logf("failed to discover interface ips: %v", err)
+		}
+		if runtime.GOOS == "linux" || runtime.GOOS == "darwin" || runtime.GOOS == "windows" {
+			rs.LocalRoutes = internalIPs // unconditionally allow access to guest VM networks
 			if prefs.ExitNodeAllowLANAccess {
-				rs.LocalRoutes = ips.Prefixes()
+				rs.LocalRoutes = append(rs.LocalRoutes, externalIPs...)
+				if len(externalIPs) != 0 {
+					b.logf("allowing exit node access to internal IPs: %v", internalIPs)
+				}
 			} else {
 				// Explicitly add routes to the local network so that we do not
 				// leak any traffic.
-				rs.Routes = append(rs.Routes, ips.Prefixes()...)
+				rs.Routes = append(rs.Routes, externalIPs...)
 			}
 		}
 	}
 
-	rs.Routes = append(rs.Routes, netaddr.IPPrefix{
-		IP:   tsaddr.TailscaleServiceIP(),
-		Bits: 32,
-	})
+	rs.Routes = append(rs.Routes, netaddr.IPPrefixFrom(tsaddr.TailscaleServiceIP(), 32))
 
 	return rs
 }
 
 func unmapIPPrefix(ipp netaddr.IPPrefix) netaddr.IPPrefix {
-	return netaddr.IPPrefix{IP: ipp.IP.Unmap(), Bits: ipp.Bits}
+	return netaddr.IPPrefixFrom(ipp.IP().Unmap(), ipp.Bits())
 }
 
 func unmapIPPrefixes(ippsList ...[]netaddr.IPPrefix) (ret []netaddr.IPPrefix) {
@@ -2073,6 +2200,18 @@ func applyPrefsToHostinfo(hi *tailcfg.Hostinfo, prefs *ipn.Prefs) {
 	}
 	if v := prefs.OSVersion; v != "" {
 		hi.OSVersion = v
+
+		// The Android app annotates when Google Play Services
+		// aren't available by tacking on a string to the
+		// OSVersion. Promote that to the Hostinfo.Package
+		// field instead, rather than adding a new pref, as
+		// this applyPrefsToHostinfo mechanism is mostly
+		// abused currently. TODO(bradfitz): instead let
+		// frontends update Hostinfo, without using Prefs.
+		if runtime.GOOS == "android" && strings.HasSuffix(v, " [nogoogle]") {
+			hi.Package = "nogoogle"
+			hi.OSVersion = strings.TrimSuffix(v, " [nogoogle]")
+		}
 	}
 	if m := prefs.DeviceModel; m != "" {
 		hi.DeviceModel = m
@@ -2092,9 +2231,7 @@ func (b *LocalBackend) enterState(newState ipn.State) {
 	oldState := b.state
 	b.state = newState
 	prefs := b.prefs
-	cc := b.cc
 	netMap := b.netMap
-	networkUp := b.prevIfState.AnyInterfaceUp()
 	activeLogin := b.activeLogin
 	authURL := b.authURL
 	if newState == ipn.Running {
@@ -2104,27 +2241,24 @@ func (b *LocalBackend) enterState(newState ipn.State) {
 		// Transitioning away from running.
 		b.closePeerAPIListenersLocked()
 	}
+	b.maybePauseControlClientLocked()
 	b.mu.Unlock()
 
 	if oldState == newState {
 		return
 	}
-	b.logf("Switching ipn state %v -> %v (WantRunning=%v)",
-		oldState, newState, prefs.WantRunning)
+	b.logf("Switching ipn state %v -> %v (WantRunning=%v, nm=%v)",
+		oldState, newState, prefs.WantRunning, netMap != nil)
 	health.SetIPNState(newState.String(), prefs.WantRunning)
 	b.send(ipn.Notify{State: &newState})
 
-	if cc != nil {
-		cc.SetPaused((newState == ipn.Stopped && netMap != nil) || !networkUp)
-	}
-
 	switch newState {
 	case ipn.NeedsLogin:
 		systemd.Status("Needs login: %s", authURL)
 		b.blockEngineUpdates(true)
 		fallthrough
 	case ipn.Stopped:
-		err := b.e.Reconfig(&wgcfg.Config{}, &router.Config{}, &dns.Config{})
+		err := b.e.Reconfig(&wgcfg.Config{}, &router.Config{}, &dns.Config{}, nil)
 		if err != nil {
 			b.logf("Reconfig(down): %v", err)
 		}
@@ -2138,8 +2272,8 @@ func (b *LocalBackend) enterState(newState ipn.State) {
 		b.e.RequestStatus()
 	case ipn.Running:
 		var addrs []string
-		for _, addr := range b.netMap.Addresses {
-			addrs = append(addrs, addr.IP.String())
+		for _, addr := range netMap.Addresses {
+			addrs = append(addrs, addr.IP().String())
 		}
 		systemd.Status("Connected; %s; %s", activeLogin, strings.Join(addrs, " "))
 	default:
@@ -2166,13 +2300,14 @@ func (b *LocalBackend) nextState() ipn.State {
 		cc          = b.cc
 		netMap      = b.netMap
 		state       = b.state
+		blocked     = b.blocked
 		wantRunning = b.prefs.WantRunning
 		loggedOut   = b.prefs.LoggedOut
 	)
 	b.mu.Unlock()
 
 	switch {
-	case !wantRunning && !loggedOut && b.hasNodeKey():
+	case !wantRunning && !loggedOut && !blocked && b.hasNodeKey():
 		return ipn.Stopped
 	case netMap == nil:
 		if cc.AuthCantContinue() || loggedOut {
@@ -2243,7 +2378,7 @@ func (b *LocalBackend) stateMachine() {
 // a status update that predates the "I've shut down" update.
 func (b *LocalBackend) stopEngineAndWait() {
 	b.logf("stopEngineAndWait...")
-	b.e.Reconfig(&wgcfg.Config{}, &router.Config{}, &dns.Config{})
+	b.e.Reconfig(&wgcfg.Config{}, &router.Config{}, &dns.Config{}, nil)
 	b.requestEngineStatusAndWait()
 	b.logf("stopEngineAndWait: done.")
 }
@@ -2300,7 +2435,6 @@ func (b *LocalBackend) LogoutSync(ctx context.Context) error {
 func (b *LocalBackend) logout(ctx context.Context, sync bool) error {
 	b.mu.Lock()
 	cc := b.cc
-	b.setNetMapLocked(nil)
 	b.mu.Unlock()
 
 	b.EditPrefs(&ipn.MaskedPrefs{
@@ -2327,10 +2461,6 @@ func (b *LocalBackend) logout(ctx context.Context, sync bool) error {
 		cc.StartLogout()
 	}
 
-	b.mu.Lock()
-	b.setNetMapLocked(nil)
-	b.mu.Unlock()
-
 	b.stateMachine()
 	return err
 }
@@ -2382,6 +2512,7 @@ func (b *LocalBackend) setNetMapLocked(nm *netmap.NetworkMap) {
 		b.logf("active login: %v", login)
 		b.activeLogin = login
 	}
+	b.maybePauseControlClientLocked()
 
 	// Determine if file sharing is enabled
 	fs := hasCapability(nm, tailcfg.CapabilityFileSharing)
@@ -2406,7 +2537,7 @@ func (b *LocalBackend) setNetMapLocked(nm *netmap.NetworkMap) {
 	addNode := func(n *tailcfg.Node) {
 		for _, ipp := range n.Addresses {
 			if ipp.IsSingleIP() {
-				b.nodeByAddr[ipp.IP] = n
+				b.nodeByAddr[ipp.IP()] = n
 			}
 		}
 	}
@@ -2532,6 +2663,42 @@ func (b *LocalBackend) FileTargets() ([]*apitype.FileTarget, error) {
 	return ret, nil
 }
 
+// SetDNS adds a DNS record for the given domain name & TXT record
+// value.
+//
+// It's meant for use with dns-01 ACME (LetsEncrypt) challenges.
+//
+// This is the low-level interface. Other layers will provide more
+// friendly options to get HTTPS certs.
+func (b *LocalBackend) SetDNS(ctx context.Context, name, value string) error {
+	req := &tailcfg.SetDNSRequest{
+		Version: 1,
+		Type:    "TXT",
+		Name:    name,
+		Value:   value,
+	}
+
+	b.mu.Lock()
+	cc := b.cc
+	if prefs := b.prefs; prefs != nil {
+		req.NodeKey = tailcfg.NodeKey(prefs.Persist.PrivateNodeKey.Public())
+	}
+	b.mu.Unlock()
+	if cc == nil {
+		return errors.New("not connected")
+	}
+	if req.NodeKey.IsZero() {
+		return errors.New("no nodekey")
+	}
+	if name == "" {
+		return errors.New("missing 'name'")
+	}
+	if value == "" {
+		return errors.New("missing 'value'")
+	}
+	return cc.SetDNS(ctx, req)
+}
+
 func (b *LocalBackend) registerIncomingFile(inf *incomingFile, active bool) {
 	b.mu.Lock()
 	defer b.mu.Unlock()
@@ -2558,9 +2725,9 @@ func peerAPIBase(nm *netmap.NetworkMap, peer *tailcfg.Node) string {
 			continue
 		}
 		switch {
-		case a.IP.Is4():
+		case a.IP().Is4():
 			have4 = true
-		case a.IP.Is6():
+		case a.IP().Is6():
 			have6 = true
 		}
 	}
@@ -2576,11 +2743,11 @@ func peerAPIBase(nm *netmap.NetworkMap, peer *tailcfg.Node) string {
 	var ipp netaddr.IPPort
 	switch {
 	case have4 && p4 != 0:
-		ipp = netaddr.IPPort{IP: nodeIP(peer, netaddr.IP.Is4), Port: p4}
+		ipp = netaddr.IPPortFrom(nodeIP(peer, netaddr.IP.Is4), p4)
 	case have6 && p6 != 0:
-		ipp = netaddr.IPPort{IP: nodeIP(peer, netaddr.IP.Is6), Port: p6}
+		ipp = netaddr.IPPortFrom(nodeIP(peer, netaddr.IP.Is6), p6)
 	}
-	if ipp.IP.IsZero() {
+	if ipp.IP().IsZero() {
 		return ""
 	}
 	return fmt.Sprintf("http://%v", ipp)
@@ -2588,8 +2755,8 @@ func peerAPIBase(nm *netmap.NetworkMap, peer *tailcfg.Node) string {
 
 func nodeIP(n *tailcfg.Node, pred func(netaddr.IP) bool) netaddr.IP {
 	for _, a := range n.Addresses {
-		if a.IsSingleIP() && pred(a.IP) {
-			return a.IP
+		if a.IsSingleIP() && pred(a.IP()) {
+			return a.IP()
 		}
 	}
 	return netaddr.IP{}
@@ -2604,7 +2771,6 @@ func (b *LocalBackend) CheckIPForwarding() error {
 		return nil
 	}
 	if isBSD(runtime.GOOS) {
-		//lint:ignore ST1005 output to users as is
 		return fmt.Errorf("Subnet routing and exit nodes only work with additional manual configuration on %v, and is not currently officially supported.", runtime.GOOS)
 	}
 
@@ -2618,20 +2784,18 @@ func (b *LocalBackend) CheckIPForwarding() error {
 		return nil
 	}
 
+	const suffix = "\nSubnet routes won't work without IP forwarding.\nSee https://tailscale.com/kb/1104/enable-ip-forwarding/"
 	for _, key := range keys {
 		bs, err := exec.Command("sysctl", "-n", key).Output()
 		if err != nil {
-			//lint:ignore ST1005 output to users as is
-			return fmt.Errorf("couldn't check %s (%v).\nSubnet routes won't work without IP forwarding.", key, err)
+			return fmt.Errorf("couldn't check %s (%v)%s", key, err, suffix)
 		}
 		on, err := strconv.ParseBool(string(bytes.TrimSpace(bs)))
 		if err != nil {
-			//lint:ignore ST1005 output to users as is
-			return fmt.Errorf("couldn't parse %s (%v).\nSubnet routes won't work without IP forwarding.", key, err)
+			return fmt.Errorf("couldn't parse %s (%v)%s.", key, err, suffix)
 		}
 		if !on {
-			//lint:ignore ST1005 output to users as is
-			return fmt.Errorf("%s is disabled. Subnet routes won't work.", key)
+			return fmt.Errorf("%s is disabled.%s", key, suffix)
 		}
 	}
 	return nil
@@ -2649,3 +2813,13 @@ func (b *LocalBackend) PeerDialControlFunc() func(network, address string, c sys
 	}
 	return nil
 }
+
+// DERPMap returns the current DERPMap in use, or nil if not connected.
+func (b *LocalBackend) DERPMap() *tailcfg.DERPMap {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	if b.netMap == nil {
+		return nil
+	}
+	return b.netMap.DERPMap
+}

ipn/ipnlocal/local_test.go

@@ -171,7 +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() },
 		},

ipn/ipnlocal/peerapi.go

@@ -510,7 +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")

ipn/ipnlocal/peerapi_macios_ext.go

@@ -2,21 +2,19 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build darwin,redo ios,redo
+// +build darwin,ts_macext ios,ts_macext
 
 package ipnlocal
 
 import (
 	"errors"
 	"fmt"
-	"log"
 	"net"
-	"strings"
 	"syscall"
 
-	"golang.org/x/sys/unix"
 	"inet.af/netaddr"
 	"tailscale.com/net/interfaces"
+	"tailscale.com/net/netns"
 )
 
 func init() {
@@ -32,29 +30,7 @@ func initListenConfigNetworkExtension(nc *net.ListenConfig, ip netaddr.IP, st *i
 	if !ok {
 		return fmt.Errorf("no interface with name %q", tunIfName)
 	}
-	nc.Control = func(network, address string, c syscall.RawConn) error {
-		var sockErr error
-		err := c.Control(func(fd uintptr) {
-			sockErr = bindIf(fd, network, address, tunIf.Index)
-			log.Printf("peerapi: bind(%q, %q) on index %v = %v", network, address, tunIf.Index, sockErr)
-		})
-		if err != nil {
-			return err
-		}
-		return sockErr
-	}
-	return nil
-}
-
-func bindIf(fd uintptr, network, address string, ifIndex int) error {
-	v6 := strings.Contains(address, "]:") || strings.HasSuffix(network, "6") // hacky test for v6
-	proto := unix.IPPROTO_IP
-	opt := unix.IP_BOUND_IF
-	if v6 {
-		proto = unix.IPPROTO_IPV6
-		opt = unix.IPV6_BOUND_IF
-	}
-	return unix.SetsockoptInt(int(fd), proto, opt, ifIndex)
+	return netns.SetListenConfigInterfaceIndex(nc, tunIf.Index)
 }
 
 func peerDialControlFuncNetworkExtension(b *LocalBackend) func(network, address string, c syscall.RawConn) error {
@@ -68,17 +44,12 @@ func peerDialControlFuncNetworkExtension(b *LocalBackend) func(network, address
 			index = tunIf.Index
 		}
 	}
+	var lc net.ListenConfig
+	netns.SetListenConfigInterfaceIndex(&lc, index)
 	return func(network, address string, c syscall.RawConn) error {
 		if index == -1 {
 			return errors.New("failed to find TUN interface to bind to")
 		}
-		var sockErr error
-		err := c.Control(func(fd uintptr) {
-			sockErr = bindIf(fd, network, address, index)
-		})
-		if err != nil {
-			return err
-		}
-		return sockErr
+		return lc.Control(network, address, c)
 	}
 }

ipn/ipnlocal/state_test.go

@@ -14,6 +14,7 @@ import (
 
 	"tailscale.com/control/controlclient"
 	"tailscale.com/ipn"
+	"tailscale.com/syncs"
 	"tailscale.com/tailcfg"
 	"tailscale.com/types/empty"
 	"tailscale.com/types/logger"
@@ -140,6 +141,8 @@ func (cc *mockControl) send(err error, url string, loginFinished bool, nm *netma
 		}
 		if loginFinished {
 			s.LoginFinished = &empty.Message{}
+		} else if url == "" && err == nil && nm == nil {
+			s.LogoutFinished = &empty.Message{}
 		}
 		cc.statusFunc(s)
 	}
@@ -246,6 +249,10 @@ func (cc *mockControl) UpdateEndpoints(localPort uint16, endpoints []tailcfg.End
 	cc.called("UpdateEndpoints")
 }
 
+func (*mockControl) SetDNS(context.Context, *tailcfg.SetDNSRequest) error {
+	panic("unexpected SetDNS call")
+}
+
 // 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.
@@ -271,7 +278,7 @@ func TestStateMachine(t *testing.T) {
 	c := qt.New(t)
 
 	logf := t.Logf
-	store := new(ipn.MemoryStore)
+	store := new(testStateStorage)
 	e, err := wgengine.NewFakeUserspaceEngine(logf, 0)
 	if err != nil {
 		t.Fatalf("NewFakeUserspaceEngine: %v", err)
@@ -347,7 +354,7 @@ func TestStateMachine(t *testing.T) {
 	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())
+		c.Assert([]string{"Shutdown", "unpause", "New", "unpause"}, qt.DeepEquals, cc.getCalls())
 
 		nn := notifies.drain(2)
 		c.Assert(cc.getCalls(), qt.HasLen, 0)
@@ -368,13 +375,11 @@ func TestStateMachine(t *testing.T) {
 	{
 		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"Login"})
 		notifies.drain(0)
-		// BUG: this should immediately set WantRunning to true.
-		// Users don't log in if they don't want to also connect.
-		// (Generally, we're inconsistent about who is supposed to
-		// update Prefs at what time. But the overall philosophy is:
-		// update it when the user's intent changes. This is clearly
-		// at the time the user *requests* Login, not at the time
-		// the login finishes.)
+		// 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
@@ -384,18 +389,16 @@ func TestStateMachine(t *testing.T) {
 	url1 := "http://localhost:1/1"
 	cc.send(nil, url1, false, nil)
 	{
-		c.Assert(cc.getCalls(), qt.HasLen, 0)
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause"})
 
 		// ...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)
 
-		// Trying to log in automatically sets WantRunning.
-		// BUG: that should have happened right after Login().
 		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(nn[0].Prefs.WantRunning, qt.IsFalse)
 	}
 
 	// Now we'll try an interactive login.
@@ -411,7 +414,7 @@ func TestStateMachine(t *testing.T) {
 		// 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([]string{"UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
 		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
 		c.Assert(url1, qt.Equals, *nn[0].BrowseToURL)
 	}
@@ -437,7 +440,7 @@ func TestStateMachine(t *testing.T) {
 	cc.send(nil, url2, false, nil)
 	{
 		// BUG: UpdateEndpoints again, this is getting silly.
-		c.Assert([]string{"UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		c.Assert([]string{"UpdateEndpoints", "unpause", "unpause"}, 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.
@@ -451,11 +454,12 @@ func TestStateMachine(t *testing.T) {
 	// same time.
 	// The backend should propagate this upward for the UI.
 	t.Logf("\n\nLoginFinished")
-	notifies.expect(2)
+	notifies.expect(3)
 	cc.setAuthBlocked(false)
+	cc.persist.LoginName = "user1"
 	cc.send(nil, "", true, &netmap.NetworkMap{})
 	{
-		nn := notifies.drain(2)
+		nn := notifies.drain(3)
 		// BUG: still too soon for UpdateEndpoints.
 		//
 		// Arguably it makes sense to unpause now, since the machine
@@ -466,23 +470,20 @@ func TestStateMachine(t *testing.T) {
 		// 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([]string{"unpause", "unpause", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
 		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
-		c.Assert(nn[1].State, qt.Not(qt.IsNil))
-		c.Assert(ipn.NeedsMachineAuth, qt.Equals, *nn[1].State)
+		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)
 	}
 
-	// TODO: check that the logged-in username propagates from control
-	// through to the UI notifications. I think it's used as a hint
-	// for future logins, to pre-fill the username box? Not really sure
-	// how it works.
-
 	// 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,
+	// It 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.)
@@ -491,7 +492,7 @@ func TestStateMachine(t *testing.T) {
 	})
 	{
 		nn := notifies.drain(1)
-		c.Assert([]string{"unpause", "UpdateEndpoints"}, qt.DeepEquals, cc.getCalls())
+		c.Assert([]string{"unpause", "unpause", "UpdateEndpoints", "unpause"}, qt.DeepEquals, cc.getCalls())
 		c.Assert(nn[0].State, qt.Not(qt.IsNil))
 		c.Assert(ipn.Starting, qt.Equals, *nn[0].State)
 	}
@@ -523,6 +524,7 @@ func TestStateMachine(t *testing.T) {
 
 	// The user changes their preference to WantRunning after all.
 	t.Logf("\n\nWantRunning -> true")
+	store.awaitWrite()
 	notifies.expect(2)
 	b.EditPrefs(&ipn.MaskedPrefs{
 		WantRunningSet: true,
@@ -532,15 +534,17 @@ func TestStateMachine(t *testing.T) {
 		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())
+		c.Assert([]string{"Login", "unpause", "UpdateEndpoints", "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)
+		c.Assert(store.sawWrite(), qt.IsTrue)
 	}
 
 	// Test the fast-path frontend reconnection.
-	// This one is very finicky, so we have to force State==Running.
+	// This one is very finicky, so we have to force State==Running
+	// or it won't use the fast path.
 	// TODO: actually get to State==Running, rather than cheating.
 	//  That'll require spinning up a fake DERP server and putting it in
 	//  the netmap.
@@ -554,29 +558,27 @@ func TestStateMachine(t *testing.T) {
 		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?)
+		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
 	}
 
 	// undo the state hack above.
 	b.state = ipn.Starting
 
 	// User wants to logout.
+	store.awaitWrite()
 	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([]string{"pause", "StartLogout", "pause"}, 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(ipn.Stopped, 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())
+		c.Assert(ipn.Stopped, qt.Equals, b.State())
+		c.Assert(store.sawWrite(), qt.IsTrue)
 	}
 
 	// Let's make the logout succeed.
@@ -586,72 +588,67 @@ func TestStateMachine(t *testing.T) {
 	cc.send(nil, "", false, nil)
 	{
 		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.IsTrue)
-		// BUG: WantRunning should be false after manual logout.
-		c.Assert(nn[0].Prefs.WantRunning, qt.IsTrue)
+		c.Assert([]string{"unpause", "unpause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].State, qt.Not(qt.IsNil))
+		c.Assert(ipn.NeedsLogin, qt.Equals, *nn[0].State)
+		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(1)
+	notifies.expect(0)
 	b.Logout()
 	{
-		nn := notifies.drain(1)
+		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())
-		// BUG: Prefs should not change here. Already logged out.
-		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
-		c.Assert(nn[0].Prefs.LoggedOut, qt.IsTrue)
-		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		c.Assert([]string{"StartLogout", "unpause"}, 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(1)
+	notifies.expect(0)
 	cc.setAuthBlocked(true)
 	cc.send(nil, "", false, nil)
 	{
-		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.IsTrue)
-		// BUG: second logout shouldn't cause WantRunning->true !!
-		c.Assert(nn[0].Prefs.WantRunning, qt.IsTrue)
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause", "unpause"})
+		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(1)
+	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.
 	{
-		nn := notifies.drain(1)
-		c.Assert([]string{"Logout"}, qt.DeepEquals, cc.getCalls())
-		c.Assert(nn[0].Prefs, qt.Not(qt.IsNil))
-		c.Assert(nn[0].Prefs.LoggedOut, qt.IsTrue)
-		c.Assert(nn[0].Prefs.WantRunning, qt.IsFalse)
+		notifies.drain(0)
+		c.Assert([]string{"Logout", "unpause"}, 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(1)
+	notifies.expect(0)
 	cc.setAuthBlocked(true)
 	cc.send(nil, "", false, nil)
 	{
-		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.IsTrue)
-		// BUG: third logout shouldn't cause WantRunning->true !!
-		c.Assert(nn[0].Prefs.WantRunning, qt.IsTrue)
+		notifies.drain(0)
+		c.Assert(cc.getCalls(), qt.DeepEquals, []string{"unpause", "unpause"})
+		c.Assert(b.Prefs().LoggedOut, qt.IsTrue)
+		c.Assert(b.Prefs().WantRunning, qt.IsFalse)
 		c.Assert(ipn.NeedsLogin, qt.Equals, b.State())
 	}
 
@@ -669,10 +666,6 @@ func TestStateMachine(t *testing.T) {
 	// 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.
-	//
-	// BUG: WantRunning is true here (because of the bug above).
-	//  We'll have to adjust the following test's expectations if we
-	//  fix that.
 
 	// TODO: test user switching between statekeys.
 
@@ -684,14 +677,14 @@ func TestStateMachine(t *testing.T) {
 		// 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())
+		c.Assert([]string{"Shutdown", "unpause", "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.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())
 	}
@@ -703,16 +696,20 @@ func TestStateMachine(t *testing.T) {
 	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].Prefs, qt.Not(qt.IsNil))
-		c.Assert(nn[1].LoginFinished, qt.Not(qt.IsNil))
+		c.Assert([]string{"unpause", "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))
-		c.Assert(nn[0].Prefs.LoggedOut, qt.IsFalse)
+		// 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)
 	}
 
@@ -745,9 +742,8 @@ func TestStateMachine(t *testing.T) {
 		//  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([]string{"Shutdown", "unpause", "New", "UpdateEndpoints", "Login", "unpause"}, qt.DeepEquals, cc.getCalls())
 		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))
@@ -756,6 +752,25 @@ func TestStateMachine(t *testing.T) {
 		c.Assert(ipn.Stopped, qt.Equals, *nn[1].State)
 	}
 
+	// When logged in but !WantRunning, ipn leaves us unpaused to retrieve
+	// the first netmap. Simulate that netmap being received, after which
+	// it should pause us, to avoid wasting CPU retrieving unnecessarily
+	// additional netmap updates.
+	//
+	// TODO: really the various GUIs and prefs should be refactored to
+	//  not require the netmap structure at all when starting while
+	//  !WantRunning. That would remove the need for this (or contacting
+	//  the control server at all when stopped).
+	t.Logf("\n\nStart4 -> netmap")
+	notifies.expect(0)
+	cc.send(nil, "", true, &netmap.NetworkMap{
+		MachineStatus: tailcfg.MachineAuthorized,
+	})
+	{
+		notifies.drain(0)
+		c.Assert([]string{"pause", "pause"}, qt.DeepEquals, cc.getCalls())
+	}
+
 	// Request connection.
 	// The state machine didn't call Login() earlier, so now it needs to.
 	t.Logf("\n\nWantRunning4 -> true")
@@ -773,6 +788,71 @@ func TestStateMachine(t *testing.T) {
 		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{"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)
+	}
+
+	// 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(1)
+	b.StartLoginInteractive()
+	url3 := "http://localhost:1/3"
+	cc.send(nil, url3, false, nil)
+	{
+		nn := notifies.drain(1)
+		// 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.
+		//
+		// Because the login hasn't yet completed, the old login
+		// is still valid, so it's correct that we stay paused.
+		c.Assert([]string{"Login", "pause"}, qt.DeepEquals, cc.getCalls())
+		c.Assert(nn[0].BrowseToURL, qt.Not(qt.IsNil))
+		c.Assert(*nn[0].BrowseToURL, qt.Equals, url3)
+	}
+
+	// Now, let's complete the interactive login, using a different
+	// user account than before. WantRunning changes to true after an
+	// interactive login, so we end up unpaused.
+	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)
+		// BUG: pause() being called here is a bad sign.
+		//  It means that either the state machine ran at least once
+		//  with the old netmap, or it ran with the new login+netmap
+		//  and !WantRunning. But since it's a fresh and successful
+		//  new login, WantRunning is true, so there was never a
+		//  reason to pause().
+		c.Assert([]string{"pause", "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")
@@ -781,7 +861,7 @@ func TestStateMachine(t *testing.T) {
 	{
 		// 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())
+		c.Assert([]string{"Shutdown", "unpause", "New", "UpdateEndpoints", "Login", "unpause"}, qt.DeepEquals, cc.getCalls())
 
 		nn := notifies.drain(1)
 		c.Assert(cc.getCalls(), qt.HasLen, 0)
@@ -793,20 +873,47 @@ func TestStateMachine(t *testing.T) {
 
 	// Control server accepts our valid key from before.
 	t.Logf("\n\nLoginFinished5")
-	notifies.expect(2)
+	notifies.expect(1)
 	cc.setAuthBlocked(false)
 	cc.send(nil, "", true, &netmap.NetworkMap{
 		MachineStatus: tailcfg.MachineAuthorized,
 	})
 	{
-		nn := notifies.drain(2)
-		c.Assert([]string{"unpause"}, qt.DeepEquals, cc.getCalls())
-		c.Assert(nn[0].LoginFinished, qt.Not(qt.IsNil))
-		c.Assert(nn[1].State, qt.Not(qt.IsNil))
-		c.Assert(ipn.Starting, qt.Equals, *nn[1].State)
+		nn := notifies.drain(1)
+		c.Assert([]string{"unpause", "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())
 	}
 }
+
+type testStateStorage struct {
+	mem     ipn.MemoryStore
+	written syncs.AtomicBool
+}
+
+func (s *testStateStorage) ReadState(id ipn.StateKey) ([]byte, error) {
+	return s.mem.ReadState(id)
+}
+
+func (s *testStateStorage) WriteState(id ipn.StateKey, bs []byte) error {
+	s.written.Set(true)
+	return s.mem.WriteState(id, bs)
+}
+
+// awaitWrite clears the "I've seen writes" bit, in prep for a future
+// call to sawWrite to see if a write arrived.
+func (s *testStateStorage) awaitWrite() { s.written.Set(false) }
+
+// sawWrite reports whether there's been a WriteState call since the most
+// recent awaitWrite call.
+func (s *testStateStorage) sawWrite() bool {
+	v := s.written.Get()
+	s.awaitWrite()
+	return v
+}

ipn/ipnserver/server.go

@@ -6,7 +6,9 @@ package ipnserver
 
 import (
 	"bufio"
+	"bytes"
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
@@ -22,7 +24,6 @@ import (
 	"strconv"
 	"strings"
 	"sync"
-	"sync/atomic"
 	"syscall"
 	"time"
 
@@ -39,9 +40,11 @@ import (
 	"tailscale.com/safesocket"
 	"tailscale.com/smallzstd"
 	"tailscale.com/types/logger"
+	"tailscale.com/util/groupmember"
 	"tailscale.com/util/pidowner"
 	"tailscale.com/util/systemd"
 	"tailscale.com/version"
+	"tailscale.com/version/distro"
 	"tailscale.com/wgengine"
 )
 
@@ -145,7 +148,7 @@ 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 +254,7 @@ 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())
@@ -346,51 +348,32 @@ func isReadonlyConn(ci connIdentity, operatorUID string, logf logger.Logf) bool
 		logf("connection from userid %v; is configured operator", uid)
 		return rw
 	}
-	var adminGroupID string
-	switch runtime.GOOS {
-	case "darwin":
-		adminGroupID = darwinAdminGroupID()
-	default:
-		logf("connection from userid %v; read-only", uid)
+	if yes, err := isLocalAdmin(uid); err != nil {
+		logf("connection from userid %v; read-only; %v", uid, err)
 		return ro
-	}
-	if adminGroupID == "" {
-		logf("connection from userid %v; no system admin group found, read-only", uid)
-		return ro
-	}
-	u, err := user.LookupId(uid)
-	if err != nil {
-		logf("connection from userid %v; failed to look up user; read-only", uid)
-		return ro
-	}
-	gids, err := u.GroupIds()
-	if err != nil {
-		logf("connection from userid %v; failed to look up groups; read-only", uid)
-		return ro
-	}
-	for _, gid := range gids {
-		if gid == adminGroupID {
-			logf("connection from userid %v; is local admin, has access", uid)
-			return rw
-		}
+	} else if yes {
+		logf("connection from userid %v; is local admin, has access", uid)
+		return rw
 	}
 	logf("connection from userid %v; read-only", uid)
 	return ro
 }
 
-var darwinAdminGroupIDCache atomic.Value // of string
-
-func darwinAdminGroupID() string {
-	s, _ := darwinAdminGroupIDCache.Load().(string)
-	if s != "" {
-		return s
-	}
-	g, err := user.LookupGroup("admin")
+func isLocalAdmin(uid string) (bool, error) {
+	u, err := user.LookupId(uid)
 	if err != nil {
-		return ""
+		return false, err
 	}
-	darwinAdminGroupIDCache.Store(g.Gid)
-	return g.Gid
+	var adminGroup string
+	switch {
+	case runtime.GOOS == "darwin":
+		adminGroup = "admin"
+	case distro.Get() == distro.QNAP:
+		adminGroup = "administrators"
+	default:
+		return false, fmt.Errorf("no system admin group found")
+	}
+	return groupmember.IsMemberOfGroup(adminGroup, u.Username)
 }
 
 // inUseOtherUserError is the error type for when the server is in use
@@ -414,12 +397,10 @@ func (s *server) checkConnIdentityLocked(ci connIdentity) error {
 			break
 		}
 		if ci.UserID != active.UserID {
-			//lint:ignore ST1005 we want to capitalize Tailscale here
 			return inUseOtherUserError{fmt.Errorf("Tailscale already in use by %s, pid %d", active.User.Username, active.Pid)}
 		}
 	}
 	if su := s.serverModeUser; su != nil && ci.UserID != su.Uid {
-		//lint:ignore ST1005 we want to capitalize Tailscale here
 		return inUseOtherUserError{fmt.Errorf("Tailscale already in use by %s", su.Username)}
 	}
 	return nil
@@ -567,7 +548,9 @@ 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()
@@ -584,8 +567,17 @@ 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)
+		}
 	}
 }
 
@@ -671,8 +663,7 @@ 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)
 			}()
@@ -962,3 +953,25 @@ 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

@@ -20,6 +20,7 @@ import (
 
 	"inet.af/netaddr"
 	"tailscale.com/tailcfg"
+	"tailscale.com/tstime/mono"
 	"tailscale.com/types/key"
 	"tailscale.com/util/dnsname"
 )
@@ -45,6 +46,13 @@ type Status struct {
 	// has MagicDNS enabled.
 	MagicDNSSuffix string
 
+	// CertDomains are the set of DNS names for which the control
+	// plane server will assist with provisioning TLS
+	// certificates. See SetDNSRequest for dns-01 ACME challenges
+	// for e.g. LetsEncrypt. These names are FQDNs without
+	// trailing periods, and without any "_acme-challenge." prefix.
+	CertDomains []string
+
 	Peer map[key.Public]*PeerStatus
 	User map[tailcfg.UserID]tailcfg.UserProfile
 }
@@ -65,6 +73,7 @@ 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
@@ -82,7 +91,7 @@ type PeerStatus struct {
 	RxBytes       int64
 	TxBytes       int64
 	Created       time.Time // time registered with tailcontrol
-	LastWrite     time.Time // time last packet sent
+	LastWrite     mono.Time // time last packet sent
 	LastSeen      time.Time // last seen to tailcontrol
 	LastHandshake time.Time // with local wireguard
 	KeepAlive     bool
@@ -203,6 +212,9 @@ 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
 	}
@@ -309,7 +321,7 @@ table tbody tr:nth-child(even) td { background-color: #f5f5f5; }
 	f("<tr><th>Peer</th><th>OS</th><th>Node</th><th>Owner</th><th>Rx</th><th>Tx</th><th>Activity</th><th>Connection</th></tr>\n")
 	f("</thead>\n<tbody>\n")
 
-	now := time.Now()
+	now := mono.Now()
 
 	var peers []*PeerStatus
 	for _, peer := range st.Peers() {
@@ -367,7 +379,7 @@ table tbody tr:nth-child(even) td { background-color: #f5f5f5; }
 		f("<td>")
 
 		// TODO: let server report this active bool instead
-		active := !ps.LastWrite.IsZero() && time.Since(ps.LastWrite) < 2*time.Minute
+		active := !ps.LastWrite.IsZero() && mono.Since(ps.LastWrite) < 2*time.Minute
 		if active {
 			if ps.Relay != "" && ps.CurAddr == "" {
 				f("relay <b>%s</b>", html.EscapeString(ps.Relay))

ipn/localapi/localapi.go

@@ -100,6 +100,10 @@ func (h *Handler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		h.serveBugReport(w, r)
 	case "/localapi/v0/file-targets":
 		h.serveFileTargets(w, r)
+	case "/localapi/v0/set-dns":
+		h.serveSetDNS(w, r)
+	case "/localapi/v0/derpmap":
+		h.serveDERPMap(w, r)
 	case "/":
 		io.WriteString(w, "tailscaled\n")
 	default:
@@ -262,7 +266,7 @@ func (h *Handler) serveFiles(w http.ResponseWriter, r *http.Request) {
 		http.Error(w, "file access denied", http.StatusForbidden)
 		return
 	}
-	suffix := strings.TrimPrefix(r.URL.Path, "/localapi/v0/files/")
+	suffix := strings.TrimPrefix(r.URL.EscapedPath(), "/localapi/v0/files/")
 	if suffix == "" {
 		if r.Method != "GET" {
 			http.Error(w, "want GET to list files", 400)
@@ -382,6 +386,36 @@ func (h *Handler) serveFilePut(w http.ResponseWriter, r *http.Request) {
 	rp.ServeHTTP(w, outReq)
 }
 
+func (h *Handler) serveSetDNS(w http.ResponseWriter, r *http.Request) {
+	if !h.PermitWrite {
+		http.Error(w, "access denied", http.StatusForbidden)
+		return
+	}
+	if r.Method != "POST" {
+		http.Error(w, "want POST", 400)
+		return
+	}
+	ctx := r.Context()
+	err := h.b.SetDNS(ctx, r.FormValue("name"), r.FormValue("value"))
+	if err != nil {
+		writeErrorJSON(w, err)
+		return
+	}
+	w.Header().Set("Content-Type", "application/json")
+	json.NewEncoder(w).Encode(struct{}{})
+}
+
+func (h *Handler) serveDERPMap(w http.ResponseWriter, r *http.Request) {
+	if r.Method != "GET" {
+		http.Error(w, "want GET", 400)
+		return
+	}
+	w.Header().Set("Content-Type", "application/json")
+	e := json.NewEncoder(w)
+	e.SetIndent("", "\t")
+	e.Encode(h.b.DERPMap())
+}
+
 var dialPeerTransportOnce struct {
 	sync.Once
 	v *http.Transport
@@ -390,7 +424,7 @@ var dialPeerTransportOnce struct {
 func getDialPeerTransport(b *ipnlocal.LocalBackend) *http.Transport {
 	dialPeerTransportOnce.Do(func() {
 		t := http.DefaultTransport.(*http.Transport).Clone()
-		t.Dial = nil //lint:ignore SA1019 yes I know I'm setting it to nil defensively
+		t.Dial = nil
 		dialer := net.Dialer{
 			Timeout:   30 * time.Second,
 			KeepAlive: 30 * time.Second,

ipn/message.go

@@ -88,9 +88,9 @@ 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
 }
 
 // NewBackendServer creates a new BackendServer using b.
@@ -98,13 +98,15 @@ type BackendServer struct {
 // 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(b []byte)) *BackendServer {
+func NewBackendServer(logf logger.Logf, b Backend, sendNotifyMsg func(Notify)) *BackendServer {
 	bs := &BackendServer{
 		logf:          logf,
 		b:             b,
 		sendNotifyMsg: sendNotifyMsg,
 	}
-	if sendNotifyMsg != nil {
+	// 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
@@ -115,14 +117,7 @@ func (bs *BackendServer) send(n Notify) {
 		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) {

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) {
@@ -182,3 +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

@@ -25,10 +25,16 @@ import (
 
 //go:generate go run tailscale.com/cmd/cloner -type=Prefs -output=prefs_clone.go
 
-// DefaultControlURL returns the URL base of the control plane
+// DefaultControlURL is 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"
+const DefaultControlURL = "https://controlplane.tailscale.com"
+
+// IsLoginServerSynonym reports whether a URL is a drop-in replacement
+// for the primary Tailscale login server.
+func IsLoginServerSynonym(val interface{}) bool {
+	return val == "https://login.tailscale.com" || val == "https://controlplane.tailscale.com"
+}
 
 // Prefs are the user modifiable settings of the Tailscale node agent.
 type Prefs struct {
@@ -405,6 +411,16 @@ func (p *Prefs) ControlURLOrDefault() string {
 	return DefaultControlURL
 }
 
+// AdminPageURL returns the admin web site URL for the current ControlURL.
+func (p *Prefs) AdminPageURL() string {
+	url := p.ControlURLOrDefault()
+	if IsLoginServerSynonym(url) {
+		// TODO(crawshaw): In future release, make this https://console.tailscale.com
+		url = "https://login.tailscale.com"
+	}
+	return url + "/admin/machines"
+}
+
 // PrefsFromBytes deserializes Prefs from a JSON blob. If
 // enforceDefaults is true, Prefs.RouteAll and Prefs.AllowSingleHosts
 // are forced on.

ipn/prefs_test.go

@@ -92,13 +92,13 @@ func TestPrefsEqual(t *testing.T) {
 		},
 
 		{
-			&Prefs{ControlURL: "https://login.tailscale.com"},
+			&Prefs{ControlURL: "https://controlplane.tailscale.com"},
 			&Prefs{ControlURL: "https://login.private.co"},
 			false,
 		},
 		{
-			&Prefs{ControlURL: "https://login.tailscale.com"},
-			&Prefs{ControlURL: "https://login.tailscale.com"},
+			&Prefs{ControlURL: "https://controlplane.tailscale.com"},
+			&Prefs{ControlURL: "https://controlplane.tailscale.com"},
 			true,
 		},
 
@@ -324,7 +324,7 @@ func TestBasicPrefs(t *testing.T) {
 	tstest.PanicOnLog()
 
 	p := Prefs{
-		ControlURL: "https://login.tailscale.com",
+		ControlURL: "https://controlplane.tailscale.com",
 	}
 	checkPrefs(t, p)
 }
@@ -336,7 +336,7 @@ func TestPrefsPersist(t *testing.T) {
 		LoginName: "test@example.com",
 	}
 	p := Prefs{
-		ControlURL: "https://login.tailscale.com",
+		ControlURL: "https://controlplane.tailscale.com",
 		CorpDNS:    true,
 		Persist:    &c,
 	}

ipn/store.go

@@ -67,9 +67,6 @@ func (s *MemoryStore) String() string { return "MemoryStore" }
 func (s *MemoryStore) ReadState(id StateKey) ([]byte, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	if s.cache == nil {
-		s.cache = map[StateKey][]byte{}
-	}
 	bs, ok := s.cache[id]
 	if !ok {
 		return nil, ErrStateNotExist

logpolicy/logpolicy.go

@@ -128,6 +128,13 @@ func (l logWriter) Write(buf []byte) (int, error) {
 // logsDir returns the directory to use for log configuration and
 // buffer storage.
 func logsDir(logf logger.Logf) string {
+	if d := os.Getenv("TS_LOGS_DIR"); d != "" {
+		fi, err := os.Stat(d)
+		if err == nil && fi.IsDir() {
+			return d
+		}
+	}
+
 	// STATE_DIRECTORY is set by systemd 240+ but we support older
 	// systems-d. For example, Ubuntu 18.04 (Bionic Beaver) is 237.
 	systemdStateDir := os.Getenv("STATE_DIRECTORY")
@@ -180,6 +187,10 @@ func runningUnderSystemd() bool {
 	return false
 }
 
+func redirectStderrToLogPanics() bool {
+	return runningUnderSystemd() || os.Getenv("TS_PLEASE_PANIC") != ""
+}
+
 // tryFixLogStateLocation is a temporary fixup for
 // https://github.com/tailscale/tailscale/issues/247 . We accidentally
 // wrote logging state files to /, and then later to $CACHE_DIRECTORY
@@ -428,9 +439,14 @@ func New(collection string) *Policy {
 		c.HTTPC = &http.Client{Transport: newLogtailTransport(u.Host)}
 	}
 
-	filchBuf, filchErr := filch.New(filepath.Join(dir, cmdName), filch.Options{})
+	filchBuf, filchErr := filch.New(filepath.Join(dir, cmdName), filch.Options{
+		ReplaceStderr: redirectStderrToLogPanics(),
+	})
 	if filchBuf != nil {
 		c.Buffer = filchBuf
+		if filchBuf.OrigStderr != nil {
+			c.Stderr = filchBuf.OrigStderr
+		}
 	}
 	lw := logtail.NewLogger(c, log.Printf)
 	log.SetFlags(0) // other logflags are set on console, not here

logtail/filch/filch.go

@@ -15,7 +15,6 @@ 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
@@ -117,18 +118,17 @@ type Logger struct {
 	bo             *backoff.Backoff
 	zstdEncoder    Encoder
 	uploadCancel   func()
+	explainedRaw   bool
 
 	shutdownStart chan struct{} // closed when shutdown begins
-	shutdownDone  chan struct{} // closd when shutdown complete
+	shutdownDone  chan struct{} // closed when shutdown complete
 }
 
 // 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.
@@ -231,6 +231,14 @@ func (l *Logger) drainPending() (res []byte) {
 			// outside of the logtail logger. Encode it.
 			// Do not add a client time, as it could have been
 			// been written a long time ago.
+			if !l.explainedRaw {
+				fmt.Fprintf(l.stderr, "RAW-STDERR: ***\n")
+				fmt.Fprintf(l.stderr, "RAW-STDERR: *** Lines prefixed with RAW-STDERR below bypassed logtail and probably come from a previous run of the program\n")
+				fmt.Fprintf(l.stderr, "RAW-STDERR: ***\n")
+				fmt.Fprintf(l.stderr, "RAW-STDERR:\n")
+				l.explainedRaw = true
+			}
+			fmt.Fprintf(l.stderr, "RAW-STDERR: %s", b)
 			b = l.encodeText(b, true)
 		}
 
@@ -514,7 +522,7 @@ 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

@@ -5,9 +5,12 @@
 package dns
 
 import (
+	"bufio"
+	"fmt"
 	"sort"
 
 	"inet.af/netaddr"
+	"tailscale.com/net/dns/resolver"
 	"tailscale.com/util/dnsname"
 )
 
@@ -22,27 +25,40 @@ 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
+}
+
+// WriteToBufioWriter write a debug version of c for logs to w, omitting
+// spammy stuff like *.arpa entries and replacing it with a total count.
+func (c *Config) WriteToBufioWriter(w *bufio.Writer) {
+	w.WriteString("{DefaultResolvers:")
+	resolver.WriteIPPorts(w, c.DefaultResolvers)
+
+	w.WriteString(" Routes:")
+	resolver.WriteRoutes(w, c.Routes)
+
+	fmt.Fprintf(w, " SearchDomains:%v", c.SearchDomains)
+	fmt.Fprintf(w, " Hosts:%v", len(c.Hosts))
+	w.WriteString("}")
 }
 
 // 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 +68,7 @@ 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 +79,28 @@ 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/direct.go

@@ -2,23 +2,22 @@
 // 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 (
 	"bufio"
 	"bytes"
+	"crypto/rand"
 	"fmt"
 	"io"
 	"io/ioutil"
 	"os"
 	"os/exec"
+	"path/filepath"
 	"runtime"
 	"strings"
 
 	"inet.af/netaddr"
-	"tailscale.com/atomicfile"
 	"tailscale.com/util/dnsname"
 )
 
@@ -77,21 +76,17 @@ func readResolv(r io.Reader) (config OSConfig, err error) {
 	return config, nil
 }
 
-func readResolvFile(path string) (OSConfig, error) {
-	var config OSConfig
-
-	f, err := os.Open(path)
+func (m directManager) readResolvFile(path string) (OSConfig, error) {
+	b, err := m.fs.ReadFile(path)
 	if err != nil {
-		return config, err
+		return OSConfig{}, err
 	}
-	defer f.Close()
-
-	return readResolv(f)
+	return readResolv(bytes.NewReader(b))
 }
 
 // readResolvConf reads DNS configuration from /etc/resolv.conf.
-func readResolvConf() (OSConfig, error) {
-	return readResolvFile(resolvConf)
+func (m directManager) readResolvConf() (OSConfig, error) {
+	return m.readResolvFile(resolvConf)
 }
 
 // resolvOwner returns the apparent owner of the resolv.conf
@@ -143,33 +138,39 @@ func isResolvedRunning() bool {
 	return err == nil
 }
 
-// directManager is a managerImpl which replaces /etc/resolv.conf with a file
+// directManager is an OSConfigurator which replaces /etc/resolv.conf with a file
 // generated from the given configuration, creating a backup of its old state.
 //
 // This way of configuring DNS is precarious, since it does not react
 // to the disappearance of the Tailscale interface.
 // The caller must call Down before program shutdown
 // or as cleanup if the program terminates unexpectedly.
-type directManager struct{}
+type directManager struct {
+	fs wholeFileFS
+}
 
-func newDirectManager() (directManager, error) {
-	return directManager{}, nil
+func newDirectManager() directManager {
+	return directManager{fs: directFS{}}
+}
+
+func newDirectManagerOnFS(fs wholeFileFS) directManager {
+	return directManager{fs: fs}
 }
 
 // ownedByTailscale reports whether /etc/resolv.conf seems to be a
 // tailscale-managed file.
 func (m directManager) ownedByTailscale() (bool, error) {
-	st, err := os.Stat(resolvConf)
+	isRegular, err := m.fs.Stat(resolvConf)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return false, nil
 		}
 		return false, err
 	}
-	if !st.Mode().IsRegular() {
+	if !isRegular {
 		return false, nil
 	}
-	bs, err := ioutil.ReadFile(resolvConf)
+	bs, err := m.fs.ReadFile(resolvConf)
 	if err != nil {
 		return false, err
 	}
@@ -182,11 +183,11 @@ func (m directManager) ownedByTailscale() (bool, error) {
 // backupConfig creates or updates a backup of /etc/resolv.conf, if
 // resolv.conf does not currently contain a Tailscale-managed config.
 func (m directManager) backupConfig() error {
-	if _, err := os.Stat(resolvConf); err != nil {
+	if _, err := m.fs.Stat(resolvConf); err != nil {
 		if os.IsNotExist(err) {
 			// No resolv.conf, nothing to back up. Also get rid of any
 			// existing backup file, to avoid restoring something old.
-			os.Remove(backupConf)
+			m.fs.Remove(backupConf)
 			return nil
 		}
 		return err
@@ -200,11 +201,11 @@ func (m directManager) backupConfig() error {
 		return nil
 	}
 
-	return os.Rename(resolvConf, backupConf)
+	return m.fs.Rename(resolvConf, backupConf)
 }
 
 func (m directManager) restoreBackup() error {
-	if _, err := os.Stat(backupConf); err != nil {
+	if _, err := m.fs.Stat(backupConf); err != nil {
 		if os.IsNotExist(err) {
 			// No backup, nothing we can do.
 			return nil
@@ -215,7 +216,7 @@ func (m directManager) restoreBackup() error {
 	if err != nil {
 		return err
 	}
-	if _, err := os.Stat(resolvConf); err != nil && !os.IsNotExist(err) {
+	if _, err := m.fs.Stat(resolvConf); err != nil && !os.IsNotExist(err) {
 		return err
 	}
 	resolvConfExists := !os.IsNotExist(err)
@@ -223,12 +224,12 @@ func (m directManager) restoreBackup() error {
 	if resolvConfExists && !owned {
 		// There's already a non-tailscale config in place, get rid of
 		// our backup.
-		os.Remove(backupConf)
+		m.fs.Remove(backupConf)
 		return nil
 	}
 
 	// We own resolv.conf, and a backup exists.
-	if err := os.Rename(backupConf, resolvConf); err != nil {
+	if err := m.fs.Rename(backupConf, resolvConf); err != nil {
 		return err
 	}
 
@@ -247,7 +248,7 @@ func (m directManager) SetDNS(config OSConfig) error {
 
 		buf := new(bytes.Buffer)
 		writeResolvConf(buf, config.Nameservers, config.SearchDomains)
-		if err := atomicfile.WriteFile(resolvConf, buf.Bytes(), 0644); err != nil {
+		if err := atomicWriteFile(m.fs, resolvConf, buf.Bytes(), 0644); err != nil {
 			return err
 		}
 	}
@@ -279,7 +280,7 @@ func (m directManager) GetBaseConfig() (OSConfig, error) {
 		fileToRead = backupConf
 	}
 
-	return readResolvFile(fileToRead)
+	return m.readResolvFile(fileToRead)
 }
 
 func (m directManager) Close() error {
@@ -287,9 +288,9 @@ func (m directManager) Close() error {
 	// to it, but then we stopped because /etc/resolv.conf being a
 	// symlink to surprising places breaks snaps and other sandboxing
 	// things. Clean it up if it's still there.
-	os.Remove("/etc/resolv.tailscale.conf")
+	m.fs.Remove("/etc/resolv.tailscale.conf")
 
-	if _, err := os.Stat(backupConf); err != nil {
+	if _, err := m.fs.Stat(backupConf); err != nil {
 		if os.IsNotExist(err) {
 			// No backup, nothing we can do.
 			return nil
@@ -300,7 +301,7 @@ func (m directManager) Close() error {
 	if err != nil {
 		return err
 	}
-	_, err = os.Stat(resolvConf)
+	_, err = m.fs.Stat(resolvConf)
 	if err != nil && !os.IsNotExist(err) {
 		return err
 	}
@@ -309,12 +310,12 @@ func (m directManager) Close() error {
 	if resolvConfExists && !owned {
 		// There's already a non-tailscale config in place, get rid of
 		// our backup.
-		os.Remove(backupConf)
+		m.fs.Remove(backupConf)
 		return nil
 	}
 
 	// We own resolv.conf, and a backup exists.
-	if err := os.Rename(backupConf, resolvConf); err != nil {
+	if err := m.fs.Rename(backupConf, resolvConf); err != nil {
 		return err
 	}
 
@@ -324,3 +325,63 @@ func (m directManager) Close() error {
 
 	return nil
 }
+
+func atomicWriteFile(fs wholeFileFS, filename string, data []byte, perm os.FileMode) error {
+	var randBytes [12]byte
+	if _, err := rand.Read(randBytes[:]); err != nil {
+		return fmt.Errorf("atomicWriteFile: %w", err)
+	}
+
+	tmpName := fmt.Sprintf("%s.%x.tmp", filename, randBytes[:])
+	defer fs.Remove(tmpName)
+
+	if err := fs.WriteFile(tmpName, data, perm); err != nil {
+		return fmt.Errorf("atomicWriteFile: %w", err)
+	}
+	return fs.Rename(tmpName, filename)
+}
+
+// wholeFileFS is a high-level file system abstraction designed just for use
+// by directManager, with the goal that it is easy to implement over wsl.exe.
+//
+// All name parameters are absolute paths.
+type wholeFileFS interface {
+	Stat(name string) (isRegular bool, err error)
+	Rename(oldName, newName string) error
+	Remove(name string) error
+	ReadFile(name string) ([]byte, error)
+	WriteFile(name string, contents []byte, perm os.FileMode) error
+}
+
+// directFS is a wholeFileFS implemented directly on the OS.
+type directFS struct {
+	// prefix is file path prefix.
+	//
+	// All name parameters are absolute paths so this is typically a
+	// testing temporary directory like "/tmp".
+	prefix string
+}
+
+func (fs directFS) path(name string) string { return filepath.Join(fs.prefix, name) }
+
+func (fs directFS) Stat(name string) (isRegular bool, err error) {
+	fi, err := os.Stat(fs.path(name))
+	if err != nil {
+		return false, err
+	}
+	return fi.Mode().IsRegular(), nil
+}
+
+func (fs directFS) Rename(oldName, newName string) error {
+	return os.Rename(fs.path(oldName), fs.path(newName))
+}
+
+func (fs directFS) Remove(name string) error { return os.Remove(fs.path(name)) }
+
+func (fs directFS) ReadFile(name string) ([]byte, error) {
+	return ioutil.ReadFile(fs.path(name))
+}
+
+func (fs directFS) WriteFile(name string, contents []byte, perm os.FileMode) error {
+	return ioutil.WriteFile(fs.path(name), contents, perm)
+}

net/dns/direct_test.go

@@ -0,0 +1,83 @@
+// 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 dns
+
+import (
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"inet.af/netaddr"
+	"tailscale.com/util/dnsname"
+)
+
+func TestSetDNS(t *testing.T) {
+	const orig = "nameserver 9.9.9.9 # orig"
+	tmp := t.TempDir()
+	resolvPath := filepath.Join(tmp, "etc", "resolv.conf")
+	backupPath := filepath.Join(tmp, "etc", "resolv.pre-tailscale-backup.conf")
+
+	if err := os.MkdirAll(filepath.Dir(resolvPath), 0777); err != nil {
+		t.Fatal(err)
+	}
+	if err := ioutil.WriteFile(resolvPath, []byte(orig), 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	readFile := func(t *testing.T, path string) string {
+		t.Helper()
+		b, err := ioutil.ReadFile(path)
+		if err != nil {
+			t.Fatal(err)
+		}
+		return string(b)
+	}
+	assertBaseState := func(t *testing.T) {
+		if got := readFile(t, resolvPath); got != orig {
+			t.Fatalf("resolv.conf:\n%s, want:\n%s", got, orig)
+		}
+		if _, err := os.Stat(backupPath); !os.IsNotExist(err) {
+			t.Fatalf("resolv.conf backup: want it to be gone but: %v", err)
+		}
+	}
+
+	m := directManager{fs: directFS{prefix: tmp}}
+	if err := m.SetDNS(OSConfig{
+		Nameservers:   []netaddr.IP{netaddr.MustParseIP("8.8.8.8"), netaddr.MustParseIP("8.8.4.4")},
+		SearchDomains: []dnsname.FQDN{"ts.net.", "ts-dns.test."},
+		MatchDomains:  []dnsname.FQDN{"ignored."},
+	}); err != nil {
+		t.Fatal(err)
+	}
+	want := `# resolv.conf(5) file generated by tailscale
+# DO NOT EDIT THIS FILE BY HAND -- CHANGES WILL BE OVERWRITTEN
+
+nameserver 8.8.8.8
+nameserver 8.8.4.4
+search ts.net ts-dns.test
+`
+	if got := readFile(t, resolvPath); got != want {
+		t.Fatalf("resolv.conf:\n%s, want:\n%s", got, want)
+	}
+	if got := readFile(t, backupPath); got != orig {
+		t.Fatalf("resolv.conf backup:\n%s, want:\n%s", got, orig)
+	}
+
+	// Test that a nil OSConfig cleans up resolv.conf.
+	if err := m.SetDNS(OSConfig{}); err != nil {
+		t.Fatal(err)
+	}
+	assertBaseState(t)
+
+	// Test that Close cleans up resolv.conf.
+	if err := m.SetDNS(OSConfig{Nameservers: []netaddr.IP{netaddr.MustParseIP("8.8.8.8")}}); err != nil {
+		t.Fatal(err)
+	}
+	if err := m.Close(); err != nil {
+		t.Fatal(err)
+	}
+	assertBaseState(t)
+}

net/dns/ini.go

@@ -0,0 +1,31 @@
+// 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 windows
+
+package dns
+
+import (
+	"regexp"
+	"strings"
+)
+
+// parseIni parses a basic .ini file, used for wsl.conf.
+func parseIni(data string) map[string]map[string]string {
+	sectionRE := regexp.MustCompile(`^\[([^]]+)\]`)
+	kvRE := regexp.MustCompile(`^\s*(\w+)\s*=\s*([^#]*)`)
+
+	ini := map[string]map[string]string{}
+	var section string
+	for _, line := range strings.Split(data, "\n") {
+		if res := sectionRE.FindStringSubmatch(line); len(res) > 1 {
+			section = res[1]
+			ini[section] = map[string]string{}
+		} else if res := kvRE.FindStringSubmatch(line); len(res) > 2 {
+			k, v := strings.TrimSpace(res[1]), strings.TrimSpace(res[2])
+			ini[section][k] = v
+		}
+	}
+	return ini
+}

net/dns/ini_test.go

@@ -0,0 +1,39 @@
+// 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 windows
+
+package dns
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestParseIni(t *testing.T) {
+	var tests = []struct {
+		src  string
+		want map[string]map[string]string
+	}{
+		{
+			src: `# appended wsl.conf file
+[automount]
+	enabled = true
+	root=/mnt/
+# added by tailscale
+[network] # trailing comment
+generateResolvConf = false  # trailing comment`,
+			want: map[string]map[string]string{
+				"automount": map[string]string{"enabled": "true", "root": "/mnt/"},
+				"network":   map[string]string{"generateResolvConf": "false"},
+			},
+		},
+	}
+	for _, test := range tests {
+		got := parseIni(test.src)
+		if !reflect.DeepEqual(got, test.want) {
+			t.Errorf("for:\n%s\ngot:  %v\nwant: %v", test.src, got, test.want)
+		}
+	}
+}

net/dns/manager.go

@@ -5,8 +5,8 @@
 package dns
 
 import (
+	"bufio"
 	"runtime"
-	"strings"
 	"time"
 
 	"inet.af/netaddr"
@@ -21,8 +21,6 @@ import (
 // the lint exception is necessary and on others it is not,
 // and plain ignore complains if the exception is unnecessary.
 
-//lint:file-ignore U1000 reconfigTimeout is used on some platforms but not others
-
 // reconfigTimeout is the time interval within which Manager.{Up,Down} should complete.
 //
 // This is particularly useful because certain conditions can cause indefinite hangs
@@ -41,11 +39,11 @@ type Manager struct {
 }
 
 // NewManagers created a new manager from the given config.
-func NewManager(logf logger.Logf, oscfg OSConfigurator, linkMon *monitor.Mon) *Manager {
+func NewManager(logf logger.Logf, oscfg OSConfigurator, linkMon *monitor.Mon, linkSel resolver.ForwardLinkSelector) *Manager {
 	logf = logger.WithPrefix(logf, "dns: ")
 	m := &Manager{
 		logf:     logf,
-		resolver: resolver.New(logf, linkMon),
+		resolver: resolver.New(logf, linkMon, linkSel),
 		os:       oscfg,
 	}
 	m.logf("using %T", m.os)
@@ -53,14 +51,18 @@ func NewManager(logf logger.Logf, oscfg OSConfigurator, linkMon *monitor.Mon) *M
 }
 
 func (m *Manager) Set(cfg Config) error {
-	m.logf("Set: %+v", cfg)
+	m.logf("Set: %v", logger.ArgWriter(func(w *bufio.Writer) {
+		cfg.WriteToBufioWriter(w)
+	}))
 
 	rcfg, ocfg, err := m.compileConfig(cfg)
 	if err != nil {
 		return err
 	}
 
-	m.logf("Resolvercfg: %+v", rcfg)
+	m.logf("Resolvercfg: %v", logger.ArgWriter(func(w *bufio.Writer) {
+		rcfg.WriteToBufioWriter(w)
+	}))
 	m.logf("OScfg: %+v", ocfg)
 
 	if err := m.resolver.SetConfig(rcfg); err != nil {
@@ -75,40 +77,40 @@ 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
 	}
 
@@ -116,8 +118,6 @@ 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
 
 	// Workaround for
 	// https://github.com/tailscale/corp/issues/1662. Even though
@@ -135,35 +135,19 @@ func (m *Manager) compileConfig(cfg Config) (resolver.Config, OSConfig, error) {
 	// This bool is used in a couple of places below to implement this
 	// workaround.
 	isWindows := runtime.GOOS == "windows"
-
-	// The windows check is for
-	// . See also below
-	// for further routing workarounds there.
-	if !cfg.hasHosts() && cfg.singleResolverSet() != nil && m.os.SupportsSplitDNS() && !isWindows {
+	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.
@@ -173,28 +157,7 @@ func (m *Manager) compileConfig(cfg Config) (resolver.Config, OSConfig, error) {
 	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...)
@@ -211,7 +174,7 @@ 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
 }
@@ -219,7 +182,7 @@ 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
 }
@@ -249,7 +212,7 @@ func Cleanup(logf logger.Logf, interfaceName string) {
 		logf("creating dns cleanup: %v", err)
 		return
 	}
-	dns := NewManager(logf, oscfg, nil)
+	dns := NewManager(logf, oscfg, nil, nil)
 	if err := dns.Down(); err != nil {
 		logf("dns down: %v", err)
 	}

net/dns/manager_freebsd.go

@@ -15,7 +15,7 @@ import (
 func NewOSConfigurator(logf logger.Logf, _ string) (OSConfigurator, error) {
 	bs, err := ioutil.ReadFile("/etc/resolv.conf")
 	if os.IsNotExist(err) {
-		return newDirectManager()
+		return newDirectManager(), nil
 	}
 	if err != nil {
 		return nil, fmt.Errorf("reading /etc/resolv.conf: %w", err)
@@ -25,6 +25,6 @@ func NewOSConfigurator(logf logger.Logf, _ string) (OSConfigurator, error) {
 	case "resolvconf":
 		return newResolvconfManager(logf)
 	default:
-		return newDirectManager()
+		return newDirectManager(), nil
 	}
 }

net/dns/manager_linux.go

@@ -5,7 +5,6 @@
 package dns
 
 import (
-	"bytes"
 	"context"
 	"errors"
 	"fmt"
@@ -15,6 +14,7 @@ import (
 	"time"
 
 	"github.com/godbus/dbus/v5"
+	"inet.af/netaddr"
 	"tailscale.com/types/logger"
 	"tailscale.com/util/cmpver"
 )
@@ -42,7 +42,7 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (ret OSConfigurat
 	bs, err := ioutil.ReadFile("/etc/resolv.conf")
 	if os.IsNotExist(err) {
 		dbg("rc", "missing")
-		return newDirectManager()
+		return newDirectManager(), nil
 	}
 	if err != nil {
 		return nil, fmt.Errorf("reading /etc/resolv.conf: %w", err)
@@ -51,18 +51,27 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (ret OSConfigurat
 	switch resolvOwner(bs) {
 	case "systemd-resolved":
 		dbg("rc", "resolved")
+		// Some systems, for reasons known only to them, have a
+		// resolv.conf that has the word "systemd-resolved" in its
+		// header, but doesn't actually point to resolved. We mustn't
+		// try to program resolved in that case.
+		// https://github.com/tailscale/tailscale/issues/2136
+		if err := resolvedIsActuallyResolver(); err != nil {
+			dbg("resolved", "not-in-use")
+			return newDirectManager(), nil
+		}
 		if err := dbusPing("org.freedesktop.resolve1", "/org/freedesktop/resolve1"); err != nil {
 			dbg("resolved", "no")
-			return newDirectManager()
+			return newDirectManager(), nil
 		}
 		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")
 
@@ -79,109 +88,69 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (ret OSConfigurat
 		// "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")
+		//
+		// Even more fun, even-older versions of NM won't let us set
+		// DNS settings if the interface isn't managed by NM, with a
+		// hard failure on DBus requests. Empirically, NM 1.22 does
+		// this. Based on the versions popular distros shipped, we
+		// conservatively decree that only 1.26.0 through 1.26.5 are
+		// "safe" to use for our purposes. This roughly matches
+		// distros released in the latter half of 2020.
+		//
+		// In a perfect world, we'd avoid this by replacing
+		// configuration out from under NM entirely (e.g. using
+		// directManager to overwrite resolv.conf), but in a world
+		// where resolved runs, we need to get correct configuration
+		// into resolved regardless of what's in resolv.conf (because
+		// resolved can also be queried over dbus, or via an NSS
+		// module that bypasses /etc/resolv.conf). Given that we must
+		// get correct configuration into resolved, we have no choice
+		// but to use NM, and accept the loss of IPv6 configuration
+		// that comes with it (see
+		// https://github.com/tailscale/tailscale/issues/1699,
+		// https://github.com/tailscale/tailscale/pull/1945)
+		safe, err := nmVersionBetween("1.26.0", "1.26.5")
 		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")
+		if safe {
+			dbg("nm-safe", "yes")
 			return newNMManager(interfaceName)
 		}
-		dbg("nm-old", "no")
-		return newResolvedManager(logf)
+		dbg("nm-safe", "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")
-				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")
-			}
-		} else {
-			dbg("src-is-nm", "no")
-		}
 		if _, err := exec.LookPath("resolvconf"); err != nil {
 			dbg("resolvconf", "no")
-			return newDirectManager()
+			return newDirectManager(), nil
 		}
 		dbg("resolvconf", "yes")
 		return newResolvconfManager(logf)
 	case "NetworkManager":
+		// You'd think we would use newNMManager somewhere in
+		// here. However, as explained in
+		// https://github.com/tailscale/tailscale/issues/1699 , using
+		// NetworkManager for DNS configuration carries with it the
+		// cost of losing IPv6 configuration on the Tailscale network
+		// interface. So, when we can avoid it, we bypass
+		// NetworkManager by replacing resolv.conf directly.
+		//
+		// If you ever try to put NMManager back here, keep in mind
+		// that versions >=1.26.6 will ignore DNS configuration
+		// anyway, so you still need a fallback path that uses
+		// directManager.
 		dbg("rc", "nm")
-		if err := dbusPing("org.freedesktop.NetworkManager", "/org/freedesktop/NetworkManager/DnsManager"); err != nil {
-			dbg("nm", "no")
-			return newDirectManager()
-		}
-		dbg("nm", "yes")
-		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()
+		return newDirectManager(), nil
 	default:
 		dbg("rc", "unknown")
-		return newDirectManager()
+		return newDirectManager(), nil
 	}
 }
 
-func resolvconfSourceIsNM(resolvDotConf []byte) error {
-	b := bytes.NewBuffer(resolvDotConf)
-	cfg, err := readResolv(b)
-	if err != nil {
-		return fmt.Errorf("parsing /etc/resolv.conf: %w", err)
-	}
-
-	var (
-		paths = []string{
-			"/etc/resolvconf/run/interface/NetworkManager",
-			"/run/resolvconf/interface/NetworkManager",
-			"/var/run/resolvconf/interface/NetworkManager",
-			"/run/resolvconf/interfaces/NetworkManager",
-			"/var/run/resolvconf/interfaces/NetworkManager",
-		}
-		nmCfg OSConfig
-		found bool
-	)
-	for _, path := range paths {
-		nmCfg, err = readResolvFile(path)
-		if os.IsNotExist(err) {
-			continue
-		} else if err != nil {
-			return err
-		}
-		found = true
-		break
-	}
-	if !found {
-		return errors.New("NetworkManager resolvconf snippet not found")
-	}
-
-	if !nmCfg.Equal(cfg) {
-		return errors.New("NetworkManager config not applied by resolvconf")
-	}
-
-	return nil
-}
-
-func nmVersionOlderThan(want string) (bool, error) {
+func nmVersionBetween(first, last string) (bool, error) {
 	conn, err := dbus.SystemBus()
 	if err != nil {
 		// DBus probably not running.
@@ -199,7 +168,8 @@ func nmVersionOlderThan(want string) (bool, error) {
 		return false, fmt.Errorf("unexpected type %T for NM version", v.Value())
 	}
 
-	return cmpver.Compare(version, want) < 0, nil
+	outside := cmpver.Compare(version, first) < 0 || cmpver.Compare(version, last) > 0
+	return !outside, nil
 }
 
 func nmIsUsingResolved() error {
@@ -224,6 +194,17 @@ func nmIsUsingResolved() error {
 	return nil
 }
 
+func resolvedIsActuallyResolver() error {
+	cfg, err := newDirectManager().readResolvConf()
+	if err != nil {
+		return err
+	}
+	if len(cfg.Nameservers) != 1 || cfg.Nameservers[0] != netaddr.IPv4(127, 0, 0, 53) {
+		return errors.New("resolv.conf doesn't point to systemd-resolved")
+	}
+	return nil
+}
+
 func dbusPing(name, objectPath string) error {
 	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
 	defer cancel()

net/dns/manager_openbsd.go

@@ -7,5 +7,5 @@ package dns
 import "tailscale.com/types/logger"
 
 func NewOSConfigurator(logger.Logf, string) (OSConfigurator, error) {
-	return newDirectManager()
+	return newDirectManager(), nil
 }

net/dns/manager_test.go

@@ -76,6 +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{
@@ -104,10 +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"),
@@ -126,10 +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{
@@ -261,8 +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"),
@@ -286,8 +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{
@@ -305,12 +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"),
@@ -333,12 +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{
@@ -362,17 +376,18 @@ func TestManager(t *testing.T) {
 				SplitDNS:   test.split,
 				BaseConfig: test.bs,
 			}
-			m := NewManager(t.Logf, &f, nil)
+			m := NewManager(t.Logf, &f, nil, nil)
 			m.resolver.TestOnlySetHook(f.SetResolver)
 
 			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)
 			}
 		})
@@ -428,7 +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

@@ -35,16 +35,18 @@ const (
 )
 
 type windowsManager struct {
-	logf      logger.Logf
-	guid      string
-	nrptWorks bool
+	logf       logger.Logf
+	guid       string
+	nrptWorks  bool
+	wslManager *wslManager
 }
 
 func NewOSConfigurator(logf logger.Logf, interfaceName string) (OSConfigurator, error) {
 	ret := windowsManager{
-		logf:      logf,
-		guid:      interfaceName,
-		nrptWorks: !isWindows7(),
+		logf:       logf,
+		guid:       interfaceName,
+		nrptWorks:  isWindows10OrBetter(),
+		wslManager: newWSLManager(logf),
 	}
 
 	// Best-effort: if our NRPT rule exists, try to delete it. Unlike
@@ -57,6 +59,13 @@ func NewOSConfigurator(logf logger.Logf, interfaceName string) (OSConfigurator,
 		ret.delKey(nrptBase)
 	}
 
+	// Log WSL status once at startup.
+	if distros, err := wslDistros(); err != nil {
+		logf("WSL: could not list distributions: %v", err)
+	} else {
+		logf("WSL: found %d distributions", len(distros))
+	}
+
 	return ret, nil
 }
 
@@ -284,7 +293,7 @@ func (m windowsManager) SetDNS(cfg OSConfig) error {
 		}
 
 		t0 = time.Now()
-		m.logf("running ipconfig /registerdns ...")
+		m.logf("running ipconfig /flushdns ...")
 		cmd = exec.Command("ipconfig", "/flushdns")
 		cmd.SysProcAttr = &syscall.SysProcAttr{HideWindow: true}
 		err = cmd.Run()
@@ -296,6 +305,16 @@ func (m windowsManager) SetDNS(cfg OSConfig) error {
 		}
 	}()
 
+	// On initial setup of WSL, the restart caused by --shutdown is slow,
+	// so we do it out-of-line.
+	go func() {
+		if err := m.wslManager.SetDNS(cfg); err != nil {
+			m.logf("WSL SetDNS: %v", err) // continue
+		} else {
+			m.logf("WSL SetDNS: success")
+		}
+	}()
+
 	return nil
 }
 
@@ -407,22 +426,16 @@ var siteLocalResolvers = []netaddr.IP{
 	netaddr.MustParseIP("fec0:0:0:ffff::3"),
 }
 
-func isWindows7() bool {
+func isWindows10OrBetter() bool {
 	key, err := registry.OpenKey(registry.LOCAL_MACHINE, versionKey, registry.READ)
 	if err != nil {
-		// Fail safe, assume Windows 7.
-		return true
+		// Fail safe, assume old Windows.
+		return false
 	}
-	ver, _, err := key.GetStringValue("CurrentVersion")
-	if err != nil {
-		return true
+	// This key above only exists in Windows 10 and above. Its mere
+	// presence is good enough.
+	if _, _, err := key.GetIntegerValue("CurrentMajorVersionNumber"); err != nil {
+		return false
 	}
-	// Careful to not assume anything about version numbers beyond
-	// 6.3, Microsoft deprecated this registry key and locked its
-	// value to what it was in Windows 8.1. We can only use this to
-	// probe for versions before that. Good thing we only need Windows
-	// 7 (so far).
-	//
-	// And yes, Windows 7 is version 6.1. Don't ask.
-	return ver == "6.1"
+	return true
 }

net/dns/nm.go

@@ -4,8 +4,6 @@
 
 // +build linux
 
-//lint:file-ignore U1000 refactoring, temporarily unused code.
-
 package dns
 
 import (
@@ -175,9 +173,12 @@ func (m *nmManager) trySet(ctx context.Context, config OSConfig) error {
 		search = append(search, "~.")
 	}
 
-	general := settings["connection"]
-	general["llmnr"] = dbus.MakeVariant(0)
-	general["mdns"] = dbus.MakeVariant(0)
+	// 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)
@@ -247,7 +248,7 @@ 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/resolved.go

@@ -4,19 +4,17 @@
 
 // +build linux
 
-//lint:file-ignore U1000 refactoring, temporarily unused code.
-
 package dns
 
 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"
 )
@@ -69,7 +67,7 @@ func isResolvedActive() bool {
 		return false
 	}
 
-	config, err := readResolvConf()
+	config, err := newDirectManager().readResolvConf()
 	if err != nil {
 		return false
 	}
@@ -82,39 +80,35 @@ func isResolvedActive() bool {
 	return false
 }
 
-// resolvedManager uses the systemd-resolved DBus API.
+// resolvedManager is an OSConfigurator which 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
 }
 
-// Up implements managerImpl.
 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 +125,9 @@ 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)
@@ -174,13 +168,13 @@ 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 +183,22 @@ 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 +221,7 @@ 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/doh_test.go

@@ -0,0 +1,98 @@
+// 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 resolver
+
+import (
+	"context"
+	"flag"
+	"net/http"
+	"testing"
+
+	"golang.org/x/net/dns/dnsmessage"
+)
+
+var testDoH = flag.Bool("test-doh", false, "do real DoH tests against the network")
+
+const someDNSID = 123 // something non-zero as a test; in violation of spec's SHOULD of 0
+
+func someDNSQuestion(t testing.TB) []byte {
+	b := dnsmessage.NewBuilder(nil, dnsmessage.Header{
+		OpCode:           0, // query
+		RecursionDesired: true,
+		ID:               someDNSID,
+	})
+	b.StartQuestions() // err
+	b.Question(dnsmessage.Question{
+		Name:  dnsmessage.MustNewName("tailscale.com."),
+		Type:  dnsmessage.TypeA,
+		Class: dnsmessage.ClassINET,
+	})
+	msg, err := b.Finish()
+	if err != nil {
+		t.Fatal(err)
+	}
+	return msg
+}
+
+func TestDoH(t *testing.T) {
+	if !*testDoH {
+		t.Skip("skipping manual test without --test-doh flag")
+	}
+	if len(knownDoH) == 0 {
+		t.Fatal("no known DoH")
+	}
+
+	f := &forwarder{
+		dohSem: make(chan struct{}, 10),
+	}
+
+	for ip := range knownDoH {
+		t.Run(ip.String(), func(t *testing.T) {
+			urlBase, c, ok := f.getDoHClient(ip)
+			if !ok {
+				t.Fatal("expected DoH")
+			}
+			res, err := f.sendDoH(context.Background(), urlBase, c, someDNSQuestion(t))
+			if err != nil {
+				t.Fatal(err)
+			}
+			c.Transport.(*http.Transport).CloseIdleConnections()
+
+			var p dnsmessage.Parser
+			h, err := p.Start(res)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if h.ID != someDNSID {
+				t.Errorf("response DNS ID = %v; want %v", h.ID, someDNSID)
+			}
+
+			p.SkipAllQuestions()
+			aa, err := p.AllAnswers()
+			if err != nil {
+				t.Fatal(err)
+			}
+			if len(aa) == 0 {
+				t.Fatal("no answers")
+			}
+			for _, r := range aa {
+				t.Logf("got: %v", r.GoString())
+			}
+		})
+	}
+}
+
+func TestDoHV6Fallback(t *testing.T) {
+	for ip, base := range knownDoH {
+		if ip.Is4() {
+			ip6, ok := dohV6(base)
+			if !ok {
+				t.Errorf("no v6 DoH known for %v", ip)
+			} else if !ip6.Is6() {
+				t.Errorf("dohV6(%q) returned non-v6 address %v", base, ip6)
+			}
+		}
+	}
+}

net/dns/resolver/forwarder.go

@@ -11,38 +11,45 @@ import (
 	"errors"
 	"fmt"
 	"hash/crc32"
+	"io"
+	"io/ioutil"
 	"math/rand"
 	"net"
+	"net/http"
+	"runtime"
+	"sort"
+	"strings"
 	"sync"
 	"time"
 
 	dns "golang.org/x/net/dns/dnsmessage"
 	"inet.af/netaddr"
-	"tailscale.com/logtail/backoff"
+	"tailscale.com/net/netns"
 	"tailscale.com/types/logger"
 	"tailscale.com/util/dnsname"
+	"tailscale.com/wgengine/monitor"
 )
 
 // headerBytes is the number of bytes in a DNS message header.
 const headerBytes = 12
 
-// connCount is the number of UDP connections to use for forwarding.
-const connCount = 32
-
 const (
-	// cleanupInterval is the interval between purged of timed-out entries from txMap.
-	cleanupInterval = 30 * time.Second
 	// responseTimeout is the maximal amount of time to wait for a DNS response.
 	responseTimeout = 5 * time.Second
+
+	// dohTransportTimeout is how long to keep idle HTTP
+	// connections open to DNS-over-HTTPs servers. This is pretty
+	// arbitrary.
+	dohTransportTimeout = 30 * time.Second
+
+	// wellKnownHostBackupDelay is how long to artificially delay upstream
+	// DNS queries to the "fallback" DNS server IP for a known provider
+	// (e.g. how long to wait to query Google's 8.8.4.4 after 8.8.8.8).
+	wellKnownHostBackupDelay = 200 * time.Millisecond
 )
 
 var errNoUpstreams = errors.New("upstream nameservers not set")
 
-type forwardingRecord struct {
-	src       netaddr.IPPort
-	createdAt time.Time
-}
-
 // txid identifies a DNS transaction.
 //
 // As the standard DNS Request ID is only 16 bits, we extend it:
@@ -98,160 +105,426 @@ func getTxID(packet []byte) txid {
 	return (txid(hash) << 32) | txid(dnsid)
 }
 
+// clampEDNSSize attempts to limit the maximum EDNS response size. This is not
+// an exhaustive solution, instead only easy cases are currently handled in the
+// interest of speed and reduced complexity. Only OPT records at the very end of
+// the message with no option codes are addressed.
+// TODO: handle more situations if we discover that they happen often
+func clampEDNSSize(packet []byte, maxSize uint16) {
+	// optFixedBytes is the size of an OPT record with no option codes.
+	const optFixedBytes = 11
+	const edns0Version = 0
+
+	if len(packet) < headerBytes+optFixedBytes {
+		return
+	}
+
+	arCount := binary.BigEndian.Uint16(packet[10:12])
+	if arCount == 0 {
+		// OPT shows up in an AR, so there must be no OPT
+		return
+	}
+
+	// https://datatracker.ietf.org/doc/html/rfc6891#section-6.1.2
+	opt := packet[len(packet)-optFixedBytes:]
+
+	if opt[0] != 0 {
+		// OPT NAME must be 0 (root domain)
+		return
+	}
+	if dns.Type(binary.BigEndian.Uint16(opt[1:3])) != dns.TypeOPT {
+		// Not an OPT record
+		return
+	}
+	requestedSize := binary.BigEndian.Uint16(opt[3:5])
+	// Ignore extended RCODE in opt[5]
+	if opt[6] != edns0Version {
+		// Be conservative and don't touch unknown versions.
+		return
+	}
+	// Ignore flags in opt[6:9]
+	if binary.BigEndian.Uint16(opt[9:11]) != 0 {
+		// RDLEN must be 0 (no variable length data). We're at the end of the
+		// packet so this should be 0 anyway)..
+		return
+	}
+
+	if requestedSize <= maxSize {
+		return
+	}
+
+	// Clamp the maximum size
+	binary.BigEndian.PutUint16(opt[3:5], maxSize)
+}
+
 type route struct {
-	suffix    dnsname.FQDN
-	resolvers []netaddr.IPPort
+	Suffix    dnsname.FQDN
+	Resolvers []resolverAndDelay
+}
+
+// resolverAndDelay is an upstream DNS resolver and a delay for how
+// long to wait before querying it.
+type resolverAndDelay struct {
+	// ipp is the upstream resolver.
+	ipp netaddr.IPPort
+
+	// startDelay is an amount to delay this resolver at
+	// start. It's used when, say, there are four Google or
+	// Cloudflare DNS IPs (two IPv4 + two IPv6) and we don't want
+	// to race all four at once.
+	startDelay time.Duration
 }
 
 // forwarder forwards DNS packets to a number of upstream nameservers.
 type forwarder struct {
-	logf logger.Logf
+	logf    logger.Logf
+	linkMon *monitor.Mon
+	linkSel ForwardLinkSelector
+	dohSem  chan struct{}
+
+	ctx       context.Context    // good until Close
+	ctxCancel context.CancelFunc // closes ctx
 
 	// responses is a channel by which responses are returned.
 	responses chan packet
-	// closed signals all goroutines to stop.
-	closed chan struct{}
-	// wg signals when all goroutines have stopped.
-	wg sync.WaitGroup
 
-	// conns are the UDP connections used for forwarding.
-	// A random one is selected for each request, regardless of the target upstream.
-	conns []*fwdConn
+	mu sync.Mutex // guards following
 
-	mu sync.Mutex
-	// routes are per-suffix resolvers to use.
-	routes []route                   // most specific routes first
-	txMap  map[txid]forwardingRecord // txids to in-flight requests
+	dohClient map[netaddr.IP]*http.Client
+
+	// routes are per-suffix resolvers to use, with
+	// the most specific routes first.
+	routes []route
 }
 
 func init() {
 	rand.Seed(time.Now().UnixNano())
 }
 
-func newForwarder(logf logger.Logf, responses chan packet) *forwarder {
-	ret := &forwarder{
+func newForwarder(logf logger.Logf, responses chan packet, linkMon *monitor.Mon, linkSel ForwardLinkSelector) *forwarder {
+	maxDoHInFlight := 1000 // effectively unlimited
+	if runtime.GOOS == "ios" {
+		// No HTTP/2 on iOS yet (for size reasons), so DoH is
+		// pricier.
+		maxDoHInFlight = 10
+	}
+	f := &forwarder{
 		logf:      logger.WithPrefix(logf, "forward: "),
+		linkMon:   linkMon,
+		linkSel:   linkSel,
 		responses: responses,
-		closed:    make(chan struct{}),
-		conns:     make([]*fwdConn, connCount),
-		txMap:     make(map[txid]forwardingRecord),
+		dohSem:    make(chan struct{}, maxDoHInFlight),
 	}
-
-	ret.wg.Add(connCount + 1)
-	for idx := range ret.conns {
-		ret.conns[idx] = newFwdConn(ret.logf, idx)
-		go ret.recv(ret.conns[idx])
-	}
-	go ret.cleanMap()
-
-	return ret
+	f.ctx, f.ctxCancel = context.WithCancel(context.Background())
+	return f
 }
 
-func (f *forwarder) Close() {
-	select {
-	case <-f.closed:
-		return
-	default:
-		// continue
-	}
-	close(f.closed)
-
-	for _, conn := range f.conns {
-		conn.close()
-	}
-
-	f.wg.Wait()
+func (f *forwarder) Close() error {
+	f.ctxCancel()
+	return nil
 }
 
-func (f *forwarder) rebindFromNetworkChange() {
-	for _, c := range f.conns {
-		c.mu.Lock()
-		c.reconnectLocked()
-		c.mu.Unlock()
+// resolversWithDelays maps from a set of DNS server ip:ports (currently
+// the port is always 53) to a slice of a type that included a
+// startDelay. So if ipps contains e.g. four Google DNS IPs (two IPv4
+// + twoIPv6), this function partition adds delays to some.
+func resolversWithDelays(ipps []netaddr.IPPort) []resolverAndDelay {
+	type hostAndFam struct {
+		host string // some arbitrary string representing DNS host (currently the DoH base)
+		bits uint8  // either 32 or 128 for IPv4 vs IPv6s address family
 	}
+
+	// Track how many of each known resolver host are in the list,
+	// per address family.
+	total := map[hostAndFam]int{}
+
+	rr := make([]resolverAndDelay, len(ipps))
+	for _, ipp := range ipps {
+		ip := ipp.IP()
+		if host, ok := knownDoH[ip]; ok {
+			total[hostAndFam{host, ip.BitLen()}]++
+		}
+	}
+
+	done := map[hostAndFam]int{}
+	for i, ipp := range ipps {
+		ip := ipp.IP()
+		var startDelay time.Duration
+		if host, ok := knownDoH[ip]; ok {
+			key4 := hostAndFam{host, 32}
+			key6 := hostAndFam{host, 128}
+			switch {
+			case ip.Is4():
+				if done[key4] > 0 {
+					startDelay += wellKnownHostBackupDelay
+				}
+			case ip.Is6():
+				total4 := total[key4]
+				if total4 >= 2 {
+					// If we have two IPv4 IPs of the same provider
+					// already in the set, delay the IPv6 queries
+					// until halfway through the timeout (so wait
+					// 2.5 seconds). Even the network is IPv6-only,
+					// the DoH dialer will fallback to IPv6
+					// immediately anyway.
+					startDelay = responseTimeout / 2
+				} else if total4 == 1 {
+					startDelay += wellKnownHostBackupDelay
+				}
+				if done[key6] > 0 {
+					startDelay += wellKnownHostBackupDelay
+				}
+			}
+			done[hostAndFam{host, ip.BitLen()}]++
+		}
+		rr[i] = resolverAndDelay{
+			ipp:        ipp,
+			startDelay: startDelay,
+		}
+	}
+	return rr
 }
 
-func (f *forwarder) setRoutes(routes []route) {
+// setRoutes sets the routes to use for DNS forwarding. It's called by
+// Resolver.SetConfig on reconfig.
+//
+// The memory referenced by routesBySuffix should not be modified.
+func (f *forwarder) setRoutes(routesBySuffix map[dnsname.FQDN][]netaddr.IPPort) {
+	routes := make([]route, 0, len(routesBySuffix))
+	for suffix, ipps := range routesBySuffix {
+		routes = append(routes, route{
+			Suffix:    suffix,
+			Resolvers: resolversWithDelays(ipps),
+		})
+	}
+	// Sort from longest prefix to shortest.
+	sort.Slice(routes, func(i, j int) bool {
+		return routes[i].Suffix.NumLabels() > routes[j].Suffix.NumLabels()
+	})
+
 	f.mu.Lock()
+	defer f.mu.Unlock()
 	f.routes = routes
-	f.mu.Unlock()
+}
+
+var stdNetPacketListener packetListener = new(net.ListenConfig)
+
+type packetListener interface {
+	ListenPacket(ctx context.Context, network, address string) (net.PacketConn, error)
+}
+
+func (f *forwarder) packetListener(ip netaddr.IP) (packetListener, error) {
+	if f.linkSel == nil || initListenConfig == nil {
+		return stdNetPacketListener, nil
+	}
+	linkName := f.linkSel.PickLink(ip)
+	if linkName == "" {
+		return stdNetPacketListener, nil
+	}
+	lc := new(net.ListenConfig)
+	if err := initListenConfig(lc, f.linkMon, linkName); err != nil {
+		return nil, err
+	}
+	return lc, nil
+}
+
+func (f *forwarder) getDoHClient(ip netaddr.IP) (urlBase string, c *http.Client, ok bool) {
+	urlBase, ok = knownDoH[ip]
+	if !ok {
+		return
+	}
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if c, ok := f.dohClient[ip]; ok {
+		return urlBase, c, true
+	}
+	if f.dohClient == nil {
+		f.dohClient = map[netaddr.IP]*http.Client{}
+	}
+	nsDialer := netns.NewDialer()
+	c = &http.Client{
+		Transport: &http.Transport{
+			IdleConnTimeout: dohTransportTimeout,
+			DialContext: func(ctx context.Context, netw, addr string) (net.Conn, error) {
+				if !strings.HasPrefix(netw, "tcp") {
+					return nil, fmt.Errorf("unexpected network %q", netw)
+				}
+				c, err := nsDialer.DialContext(ctx, "tcp", net.JoinHostPort(ip.String(), "443"))
+				// If v4 failed, try an equivalent v6 also in the time remaining.
+				if err != nil && ctx.Err() == nil {
+					if ip6, ok := dohV6(urlBase); ok && ip.Is4() {
+						if c6, err := nsDialer.DialContext(ctx, "tcp", net.JoinHostPort(ip6.String(), "443")); err == nil {
+							return c6, nil
+						}
+					}
+				}
+				return c, err
+			},
+		},
+	}
+	f.dohClient[ip] = c
+	return urlBase, c, true
+}
+
+const dohType = "application/dns-message"
+
+func (f *forwarder) releaseDoHSem() { <-f.dohSem }
+
+func (f *forwarder) sendDoH(ctx context.Context, urlBase string, c *http.Client, packet []byte) ([]byte, error) {
+	// Bound the number of HTTP requests in flight. This primarily
+	// matters for iOS where we're very memory constrained and
+	// HTTP requests are heavier on iOS where we don't include
+	// HTTP/2 for binary size reasons (as binaries on iOS linked
+	// with Go code cost memory proportional to the binary size,
+	// for reasons not fully understood).
+	select {
+	case f.dohSem <- struct{}{}:
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+	defer f.releaseDoHSem()
+
+	req, err := http.NewRequestWithContext(ctx, "POST", urlBase, bytes.NewReader(packet))
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Content-Type", dohType)
+	// Note: we don't currently set the Accept header (which is
+	// only a SHOULD in the spec) as iOS doesn't use HTTP/2 and
+	// we'd rather save a few bytes on outgoing requests when
+	// empirically no provider cares about the Accept header's
+	// absence.
+
+	hres, err := c.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer hres.Body.Close()
+	if hres.StatusCode != 200 {
+		return nil, errors.New(hres.Status)
+	}
+	if ct := hres.Header.Get("Content-Type"); ct != dohType {
+		return nil, fmt.Errorf("unexpected response Content-Type %q", ct)
+	}
+	return ioutil.ReadAll(hres.Body)
 }
 
 // send sends packet to dst. It is best effort.
-func (f *forwarder) send(packet []byte, dst netaddr.IPPort) {
-	connIdx := rand.Intn(connCount)
-	conn := f.conns[connIdx]
-	conn.send(packet, dst)
-}
+//
+// send expects the reply to have the same txid as txidOut.
+//
+func (f *forwarder) send(ctx context.Context, fq *forwardQuery, dst netaddr.IPPort) ([]byte, error) {
+	ip := dst.IP()
 
-func (f *forwarder) recv(conn *fwdConn) {
-	defer f.wg.Done()
-
-	for {
-		select {
-		case <-f.closed:
-			return
-		default:
+	// Upgrade known DNS IPs to DoH (DNS-over-HTTPs).
+	if urlBase, dc, ok := f.getDoHClient(ip); ok {
+		res, err := f.sendDoH(ctx, urlBase, dc, fq.packet)
+		if err == nil || ctx.Err() != nil {
+			return res, err
 		}
-		out := make([]byte, maxResponseBytes)
-		n := conn.read(out)
-		if n == 0 {
-			continue
+		f.logf("DoH error from %v: %v", ip, err)
+	}
+
+	ln, err := f.packetListener(ip)
+	if err != nil {
+		return nil, err
+	}
+	conn, err := ln.ListenPacket(ctx, "udp", ":0")
+	if err != nil {
+		f.logf("ListenPacket failed: %v", err)
+		return nil, err
+	}
+	defer conn.Close()
+
+	fq.closeOnCtxDone.Add(conn)
+	defer fq.closeOnCtxDone.Remove(conn)
+
+	if _, err := conn.WriteTo(fq.packet, dst.UDPAddr()); err != nil {
+		if err := ctx.Err(); err != nil {
+			return nil, err
 		}
-		if n < headerBytes {
-			f.logf("recv: packet too small (%d bytes)", n)
+		return nil, err
+	}
+
+	// The 1 extra byte is to detect packet truncation.
+	out := make([]byte, maxResponseBytes+1)
+	n, _, err := conn.ReadFrom(out)
+	if err != nil {
+		if err := ctx.Err(); err != nil {
+			return nil, err
 		}
-
-		out = out[:n]
-		txid := getTxID(out)
-
-		f.mu.Lock()
-
-		record, found := f.txMap[txid]
-		// At most one nameserver will return a response:
-		// the first one to do so will delete txid from the map.
-		if !found {
-			f.mu.Unlock()
-			continue
-		}
-		delete(f.txMap, txid)
-
-		f.mu.Unlock()
-
-		pkt := packet{out, record.src}
-		select {
-		case <-f.closed:
-			return
-		case f.responses <- pkt:
-			// continue
+		if packetWasTruncated(err) {
+			err = nil
+		} else {
+			return nil, err
 		}
 	}
+	truncated := n > maxResponseBytes
+	if truncated {
+		n = maxResponseBytes
+	}
+	if n < headerBytes {
+		f.logf("recv: packet too small (%d bytes)", n)
+	}
+	out = out[:n]
+	txid := getTxID(out)
+	if txid != fq.txid {
+		return nil, errors.New("txid doesn't match")
+	}
+
+	if truncated {
+		const dnsFlagTruncated = 0x200
+		flags := binary.BigEndian.Uint16(out[2:4])
+		flags |= dnsFlagTruncated
+		binary.BigEndian.PutUint16(out[2:4], flags)
+
+		// TODO(#2067): Remove any incomplete records? RFC 1035 section 6.2
+		// states that truncation should head drop so that the authority
+		// section can be preserved if possible. However, the UDP read with
+		// a too-small buffer has already dropped the end, so that's the
+		// best we can do.
+	}
+
+	clampEDNSSize(out, maxResponseBytes)
+
+	return out, nil
 }
 
-// cleanMap periodically deletes timed-out forwarding records from f.txMap to bound growth.
-func (f *forwarder) cleanMap() {
-	defer f.wg.Done()
-
-	t := time.NewTicker(cleanupInterval)
-	defer t.Stop()
-
-	var now time.Time
-	for {
-		select {
-		case <-f.closed:
-			return
-		case now = <-t.C:
-			// continue
+// resolvers returns the resolvers to use for domain.
+func (f *forwarder) resolvers(domain dnsname.FQDN) []resolverAndDelay {
+	f.mu.Lock()
+	routes := f.routes
+	f.mu.Unlock()
+	for _, route := range routes {
+		if route.Suffix == "." || route.Suffix.Contains(domain) {
+			return route.Resolvers
 		}
-
-		f.mu.Lock()
-		for k, v := range f.txMap {
-			if now.Sub(v.createdAt) > responseTimeout {
-				delete(f.txMap, k)
-			}
-		}
-		f.mu.Unlock()
 	}
+	return nil
+}
+
+// forwardQuery is information and state about a forwarded DNS query that's
+// being sent to 1 or more upstreams.
+//
+// In the case of racing against multiple equivalent upstreams
+// (e.g. Google or CloudFlare's 4 DNS IPs: 2 IPv4 + 2 IPv6), this type
+// handles racing them more intelligently than just blasting away 4
+// queries at once.
+type forwardQuery struct {
+	txid   txid
+	packet []byte
+
+	// closeOnCtxDone lets send register values to Close if the
+	// caller's ctx expires. This avoids send from allocating its
+	// own waiting goroutine to interrupt the ReadFrom, as memory
+	// is tight on iOS and we want the number of pending DNS
+	// lookups to be bursty without too much associated
+	// goroutine/memory cost.
+	closeOnCtxDone *closePool
+
+	// TODO(bradfitz): add race delay state:
+	// mu sync.Mutex
+	// ...
 }
 
 // forward forwards the query to all upstream nameservers and returns the first response.
@@ -261,219 +534,75 @@ func (f *forwarder) forward(query packet) error {
 		return err
 	}
 
-	txid := getTxID(query.bs)
+	clampEDNSSize(query.bs, maxResponseBytes)
 
-	f.mu.Lock()
-	routes := f.routes
-	f.mu.Unlock()
-
-	var resolvers []netaddr.IPPort
-	for _, route := range routes {
-		if route.suffix != "." && !route.suffix.Contains(domain) {
-			continue
-		}
-		resolvers = route.resolvers
-		break
-	}
+	resolvers := f.resolvers(domain)
 	if len(resolvers) == 0 {
 		return errNoUpstreams
 	}
 
-	f.mu.Lock()
-	f.txMap[txid] = forwardingRecord{
-		src:       query.addr,
-		createdAt: time.Now(),
+	fq := &forwardQuery{
+		txid:           getTxID(query.bs),
+		packet:         query.bs,
+		closeOnCtxDone: new(closePool),
 	}
-	f.mu.Unlock()
+	defer fq.closeOnCtxDone.Close()
 
-	for _, resolver := range resolvers {
-		f.send(query.bs, resolver)
-	}
+	ctx, cancel := context.WithTimeout(f.ctx, responseTimeout)
+	defer cancel()
 
-	return nil
-}
+	resc := make(chan []byte, 1)
+	var (
+		mu       sync.Mutex
+		firstErr error
+	)
 
-// A fwdConn manages a single connection used to forward DNS requests.
-// Net link changes can cause a *net.UDPConn to become permanently unusable, particularly on macOS.
-// fwdConn detects such situations and transparently creates new connections.
-type fwdConn struct {
-	// logf allows a fwdConn to log.
-	logf logger.Logf
-
-	// change allows calls to read to block until a the network connection has been replaced.
-	change *sync.Cond
-
-	// mu protects fields that follow it; it is also change's Locker.
-	mu sync.Mutex
-	// closed tracks whether fwdConn has been permanently closed.
-	closed bool
-	// conn is the current active connection.
-	conn net.PacketConn
-}
-
-func newFwdConn(logf logger.Logf, idx int) *fwdConn {
-	c := new(fwdConn)
-	c.logf = logger.WithPrefix(logf, fmt.Sprintf("fwdConn %d: ", idx))
-	c.change = sync.NewCond(&c.mu)
-	// c.conn is created lazily in send
-	return c
-}
-
-// send sends packet to dst using c's connection.
-// It is best effort. It is UDP, after all. Failures are logged.
-func (c *fwdConn) send(packet []byte, dst netaddr.IPPort) {
-	var b *backoff.Backoff // lazily initialized, since it is not needed in the common case
-	backOff := func(err error) {
-		if b == nil {
-			b = backoff.NewBackoff("dns-fwdConn-send", c.logf, 30*time.Second)
-		}
-		b.BackOff(context.Background(), err)
-	}
-
-	for {
-		// Gather the current connection.
-		// We can't hold the lock while we call WriteTo.
-		c.mu.Lock()
-		conn := c.conn
-		closed := c.closed
-		if closed {
-			c.mu.Unlock()
-			return
-		}
-		if conn == nil {
-			c.reconnectLocked()
-			c.mu.Unlock()
-			continue
-		}
-		c.mu.Unlock()
-
-		_, err := conn.WriteTo(packet, dst.UDPAddr())
-		if err == nil {
-			// Success
-			return
-		}
-		if errors.Is(err, net.ErrClosed) {
-			// We intentionally closed this connection.
-			// It has been replaced by a new connection. Try again.
-			continue
-		}
-		// Something else went wrong.
-		// We have three choices here: try again, give up, or create a new connection.
-		var opErr *net.OpError
-		if !errors.As(err, &opErr) {
-			// Weird. All errors from the net package should be *net.OpError. Bail.
-			c.logf("send: non-*net.OpErr %v (%T)", err, err)
-			return
-		}
-		if opErr.Temporary() || opErr.Timeout() {
-			// I doubt that either of these can happen (this is UDP),
-			// but go ahead and try again.
-			backOff(err)
-			continue
-		}
-		if networkIsDown(err) {
-			// Fail.
-			c.logf("send: network is down")
-			return
-		}
-		if networkIsUnreachable(err) {
-			// This can be caused by a link change.
-			// Replace the existing connection with a new one.
-			c.mu.Lock()
-			// It's possible that multiple senders discovered simultaneously
-			// that the network is unreachable. Avoid reconnecting multiple times:
-			// Only reconnect if the current connection is the one that we
-			// discovered to be problematic.
-			if c.conn == conn {
-				backOff(err)
-				c.reconnectLocked()
+	for _, rr := range resolvers {
+		go func(rr resolverAndDelay) {
+			if rr.startDelay > 0 {
+				timer := time.NewTimer(rr.startDelay)
+				select {
+				case <-timer.C:
+				case <-ctx.Done():
+					timer.Stop()
+					return
+				}
 			}
-			c.mu.Unlock()
-			// Try again with our new network connection.
-			continue
+			resb, err := f.send(ctx, fq, rr.ipp)
+			if err != nil {
+				mu.Lock()
+				defer mu.Unlock()
+				if firstErr == nil {
+					firstErr = err
+				}
+				return
+			}
+			select {
+			case resc <- resb:
+			default:
+			}
+		}(rr)
+	}
+
+	select {
+	case v := <-resc:
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case f.responses <- packet{v, query.addr}:
+			return nil
 		}
-		// Unrecognized error. Fail.
-		c.logf("send: unrecognized error: %v", err)
-		return
+	case <-ctx.Done():
+		mu.Lock()
+		defer mu.Unlock()
+		if firstErr != nil {
+			return firstErr
+		}
+		return ctx.Err()
 	}
 }
 
-// read waits for a response from c's connection.
-// It returns the number of bytes read, which may be 0
-// in case of an error or a closed connection.
-func (c *fwdConn) read(out []byte) int {
-	for {
-		// Gather the current connection.
-		// We can't hold the lock while we call ReadFrom.
-		c.mu.Lock()
-		conn := c.conn
-		closed := c.closed
-		if closed {
-			c.mu.Unlock()
-			return 0
-		}
-		if conn == nil {
-			// There is no current connection.
-			// Wait for the connection to change, then try again.
-			c.change.Wait()
-			c.mu.Unlock()
-			continue
-		}
-		c.mu.Unlock()
-
-		n, _, err := conn.ReadFrom(out)
-		if err == nil {
-			// Success.
-			return n
-		}
-		if errors.Is(err, net.ErrClosed) {
-			// We intentionally closed this connection.
-			// It has been replaced by a new connection. Try again.
-			continue
-		}
-
-		c.logf("read: unrecognized error: %v", err)
-		return 0
-	}
-}
-
-// reconnectLocked replaces the current connection with a new one.
-// c.mu must be locked.
-func (c *fwdConn) reconnectLocked() {
-	c.closeConnLocked()
-	// Make a new connection.
-	conn, err := net.ListenPacket("udp", "")
-	if err != nil {
-		c.logf("ListenPacket failed: %v", err)
-	} else {
-		c.conn = conn
-	}
-	// Broadcast that a new connection is available.
-	c.change.Broadcast()
-}
-
-// closeCurrentConn closes the current connection.
-// c.mu must be locked.
-func (c *fwdConn) closeConnLocked() {
-	if c.conn == nil {
-		return
-	}
-	c.conn.Close() // unblocks all readers/writers, they'll pick up the next connection.
-	c.conn = nil
-}
-
-// close permanently closes c.
-func (c *fwdConn) close() {
-	c.mu.Lock()
-	defer c.mu.Unlock()
-	if c.closed {
-		return
-	}
-	c.closed = true
-	c.closeConnLocked()
-	// Unblock any remaining readers.
-	c.change.Broadcast()
-}
+var initListenConfig func(_ *net.ListenConfig, _ *monitor.Mon, tunName string) error
 
 // nameFromQuery extracts the normalized query name from bs.
 func nameFromQuery(bs []byte) (dnsname.FQDN, error) {
@@ -495,3 +624,108 @@ func nameFromQuery(bs []byte) (dnsname.FQDN, error) {
 	n := q.Name.Data[:q.Name.Length]
 	return dnsname.ToFQDN(rawNameToLower(n))
 }
+
+// closePool is a dynamic set of io.Closers to close as a group.
+// It's intended to be Closed at most once.
+//
+// The zero value is ready for use.
+type closePool struct {
+	mu     sync.Mutex
+	m      map[io.Closer]bool
+	closed bool
+}
+
+func (p *closePool) Add(c io.Closer) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.closed {
+		c.Close()
+		return
+	}
+	if p.m == nil {
+		p.m = map[io.Closer]bool{}
+	}
+	p.m[c] = true
+}
+
+func (p *closePool) Remove(c io.Closer) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.closed {
+		return
+	}
+	delete(p.m, c)
+}
+
+func (p *closePool) Close() error {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.closed {
+		return nil
+	}
+	p.closed = true
+	for c := range p.m {
+		c.Close()
+	}
+	return nil
+}
+
+var knownDoH = map[netaddr.IP]string{}
+
+var dohIPsOfBase = map[string][]netaddr.IP{}
+
+func addDoH(ipStr, base string) {
+	ip := netaddr.MustParseIP(ipStr)
+	knownDoH[ip] = base
+	dohIPsOfBase[base] = append(dohIPsOfBase[base], ip)
+}
+
+func dohV6(base string) (ip netaddr.IP, ok bool) {
+	for _, ip := range dohIPsOfBase[base] {
+		if ip.Is6() {
+			return ip, true
+		}
+	}
+	return ip, false
+}
+
+func init() {
+	// Cloudflare
+	addDoH("1.1.1.1", "https://cloudflare-dns.com/dns-query")
+	addDoH("1.0.0.1", "https://cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1111", "https://cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1001", "https://cloudflare-dns.com/dns-query")
+
+	// Cloudflare -Malware
+	addDoH("1.1.1.2", "https://security.cloudflare-dns.com/dns-query")
+	addDoH("1.0.0.2", "https://security.cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1112", "https://security.cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1002", "https://security.cloudflare-dns.com/dns-query")
+
+	// Cloudflare -Malware -Adult
+	addDoH("1.1.1.3", "https://family.cloudflare-dns.com/dns-query")
+	addDoH("1.0.0.3", "https://family.cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1113", "https://family.cloudflare-dns.com/dns-query")
+	addDoH("2606:4700:4700::1003", "https://family.cloudflare-dns.com/dns-query")
+
+	// Google
+	addDoH("8.8.8.8", "https://dns.google/dns-query")
+	addDoH("8.8.4.4", "https://dns.google/dns-query")
+	addDoH("2001:4860:4860::8888", "https://dns.google/dns-query")
+	addDoH("2001:4860:4860::8844", "https://dns.google/dns-query")
+
+	// OpenDNS
+	// TODO(bradfitz): OpenDNS is unique amongst this current set in that
+	// its DoH DNS names resolve to different IPs than its normal DNS
+	// IPs. Support that later. For now we assume that they're the same.
+	// addDoH("208.67.222.222", "https://doh.opendns.com/dns-query")
+	// addDoH("208.67.220.220", "https://doh.opendns.com/dns-query")
+	// addDoH("208.67.222.123", "https://doh.familyshield.opendns.com/dns-query")
+	// addDoH("208.67.220.123", "https://doh.familyshield.opendns.com/dns-query")
+
+	// Quad9
+	addDoH("9.9.9.9", "https://dns.quad9.net/dns-query")
+	addDoH("149.112.112.112", "https://dns.quad9.net/dns-query")
+	addDoH("2620:fe::fe", "https://dns.quad9.net/dns-query")
+	addDoH("2620:fe::fe:9", "https://dns.quad9.net/dns-query")
+}

net/dns/resolver/forwarder_test.go

@@ -0,0 +1,90 @@
+// 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 resolver
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	"inet.af/netaddr"
+)
+
+func (rr resolverAndDelay) String() string {
+	return fmt.Sprintf("%v+%v", rr.ipp, rr.startDelay)
+}
+
+func TestResolversWithDelays(t *testing.T) {
+	// query
+	q := func(ss ...string) (ipps []netaddr.IPPort) {
+		for _, s := range ss {
+			ipps = append(ipps, netaddr.MustParseIPPort(s))
+		}
+		return
+	}
+	// output
+	o := func(ss ...string) (rr []resolverAndDelay) {
+		for _, s := range ss {
+			var d time.Duration
+			if i := strings.Index(s, "+"); i != -1 {
+				var err error
+				d, err = time.ParseDuration(s[i+1:])
+				if err != nil {
+					panic(fmt.Sprintf("parsing duration in %q: %v", s, err))
+				}
+				s = s[:i]
+			}
+			rr = append(rr, resolverAndDelay{
+				ipp:        netaddr.MustParseIPPort(s),
+				startDelay: d,
+			})
+		}
+		return
+	}
+
+	tests := []struct {
+		name string
+		in   []netaddr.IPPort
+		want []resolverAndDelay
+	}{
+		{
+			name: "unknown-no-delays",
+			in:   q("1.2.3.4:53", "2.3.4.5:53"),
+			want: o("1.2.3.4:53", "2.3.4.5:53"),
+		},
+		{
+			name: "google-all-ipv4",
+			in:   q("8.8.8.8:53", "8.8.4.4:53"),
+			want: o("8.8.8.8:53", "8.8.4.4:53+200ms"),
+		},
+		{
+			name: "google-only-ipv6",
+			in:   q("[2001:4860:4860::8888]:53", "[2001:4860:4860::8844]:53"),
+			want: o("[2001:4860:4860::8888]:53", "[2001:4860:4860::8844]:53+200ms"),
+		},
+		{
+			name: "google-all-four",
+			in:   q("8.8.8.8:53", "8.8.4.4:53", "[2001:4860:4860::8888]:53", "[2001:4860:4860::8844]:53"),
+			want: o("8.8.8.8:53", "8.8.4.4:53+200ms", "[2001:4860:4860::8888]:53+2.5s", "[2001:4860:4860::8844]:53+2.7s"),
+		},
+		{
+			name: "quad9-one-v4-one-v6",
+			in:   q("9.9.9.9:53", "[2620:fe::fe]:53"),
+			want: o("9.9.9.9:53", "[2620:fe::fe]:53+200ms"),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := resolversWithDelays(tt.in)
+			if !reflect.DeepEqual(got, tt.want) {
+				t.Errorf("got %v; want %v", got, tt.want)
+			}
+		})
+	}
+
+}

net/dns/resolver/macios_ext.go

@@ -0,0 +1,27 @@
+// 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 darwin,ts_macext ios,ts_macext
+
+package resolver
+
+import (
+	"errors"
+	"net"
+
+	"tailscale.com/net/netns"
+	"tailscale.com/wgengine/monitor"
+)
+
+func init() {
+	initListenConfig = initListenConfigNetworkExtension
+}
+
+func initListenConfigNetworkExtension(nc *net.ListenConfig, mon *monitor.Mon, tunName string) error {
+	nif, ok := mon.InterfaceState().Interface[tunName]
+	if !ok {
+		return errors.New("utun not found")
+	}
+	return netns.SetListenConfigInterfaceIndex(nc, nif.Interface.Index)
+}

net/dns/resolver/neterr_darwin.go

@@ -23,3 +23,8 @@ func networkIsDown(err error) bool {
 func networkIsUnreachable(err error) bool {
 	return errors.Is(err, networkUnreachable)
 }
+
+// packetWasTruncated returns true if err indicates truncation but the RecvFrom
+// that generated err was otherwise successful. It always returns false on this
+// platform.
+func packetWasTruncated(err error) bool { return false }