tailcfg, control/controlclient: support delta-encoded netmaps

Should greatly reduce bandwidth for large networks (including our
hello.ipn.dev node).

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>

control/controlclient/direct.go

@@ -22,6 +22,7 @@ import (
 	"os"
 	"reflect"
 	"runtime"
+	"sort"
 	"strconv"
 	"strings"
 	"sync"
@@ -489,6 +490,7 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 	request := tailcfg.MapRequest{
 		Version:         4,
 		IncludeIPv6:     true,
+		DeltaPeers:      true,
 		KeepAlive:       c.keepAlive,
 		NodeKey:         tailcfg.NodeKey(persist.PrivateNodeKey.Public()),
 		DiscoKey:        c.discoPubKey,
@@ -573,6 +575,7 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 	// the same format before just closing the connection.
 	// We can use this same read loop either way.
 	var msg []byte
+	var previousPeers []*tailcfg.Node // for delta-purposes
 	for i := 0; i < maxPolls || maxPolls < 0; i++ {
 		vlogf("netmap: starting size read after %v (poll %v)", time.Since(t0).Round(time.Millisecond), i)
 		var siz [4]byte
@@ -594,6 +597,7 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 			vlogf("netmap: decode error: %v")
 			return err
 		}
+
 		if resp.KeepAlive {
 			vlogf("netmap: got keep-alive")
 		} else {
@@ -609,6 +613,10 @@ func (c *Direct) PollNetMap(ctx context.Context, maxPolls int, cb func(*NetworkM
 		if resp.KeepAlive {
 			continue
 		}
+
+		undeltaPeers(&resp, previousPeers)
+		previousPeers = cloneNodes(resp.Peers) // defensive/lazy clone, since this escapes to who knows where
+
 		if resp.DERPMap != nil {
 			vlogf("netmap: new map contains DERP map")
 			lastDERPMap = resp.DERPMap
@@ -851,3 +859,97 @@ func envBool(k string) bool {
 	}
 	return v
 }
+
+// undeltaPeers updates mapRes.Peers to be complete based on the provided previous peer list
+// and the PeersRemoved and PeersChanged fields in mapRes.
+// It then also nils out the delta fields.
+func undeltaPeers(mapRes *tailcfg.MapResponse, prev []*tailcfg.Node) {
+	if len(mapRes.Peers) > 0 {
+		// Not delta encoded.
+		if !nodesSorted(mapRes.Peers) {
+			log.Printf("netmap: undeltaPeers: MapResponse.Peers not sorted; sorting")
+			sortNodes(mapRes.Peers)
+		}
+		return
+	}
+
+	var removed map[tailcfg.NodeID]bool
+	if pr := mapRes.PeersRemoved; len(pr) > 0 {
+		removed = make(map[tailcfg.NodeID]bool, len(pr))
+		for _, id := range pr {
+			removed[id] = true
+		}
+	}
+	changed := mapRes.PeersChanged
+
+	if len(removed) == 0 && len(changed) == 0 {
+		// No changes fast path.
+		mapRes.Peers = prev
+		return
+	}
+
+	if !nodesSorted(changed) {
+		log.Printf("netmap: undeltaPeers: MapResponse.PeersChanged not sorted; sorting")
+		sortNodes(changed)
+	}
+	if !nodesSorted(prev) {
+		// Internal error (unrelated to the network) if we get here.
+		log.Printf("netmap: undeltaPeers: [unexpected] prev not sorted; sorting")
+		sortNodes(prev)
+	}
+
+	newFull := make([]*tailcfg.Node, 0, len(prev)-len(removed))
+	for len(prev) > 0 && len(changed) > 0 {
+		pID := prev[0].ID
+		cID := changed[0].ID
+		if removed[pID] {
+			prev = prev[1:]
+			continue
+		}
+		switch {
+		case pID < cID:
+			newFull = append(newFull, prev[0])
+			prev = prev[1:]
+		case pID == cID:
+			newFull = append(newFull, changed[0])
+			prev, changed = prev[1:], changed[1:]
+		case cID < pID:
+			newFull = append(newFull, changed[0])
+			changed = changed[1:]
+		}
+	}
+	newFull = append(newFull, changed...)
+	for _, n := range prev {
+		if !removed[n.ID] {
+			newFull = append(newFull, n)
+		}
+	}
+	sortNodes(newFull)
+	mapRes.Peers = newFull
+	mapRes.PeersChanged = nil
+	mapRes.PeersRemoved = nil
+}
+
+func nodesSorted(v []*tailcfg.Node) bool {
+	for i, n := range v {
+		if i > 0 && n.ID <= v[i-1].ID {
+			return false
+		}
+	}
+	return true
+}
+
+func sortNodes(v []*tailcfg.Node) {
+	sort.Slice(v, func(i, j int) bool { return v[i].ID < v[j].ID })
+}
+
+func cloneNodes(v1 []*tailcfg.Node) []*tailcfg.Node {
+	if v1 == nil {
+		return nil
+	}
+	v2 := make([]*tailcfg.Node, len(v1))
+	for i, n := range v1 {
+		v2[i] = n.Clone()
+	}
+	return v2
+}

control/controlclient/direct_test.go

@@ -0,0 +1,93 @@
+// Copyright (c) 2020 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package controlclient
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+	"testing"
+
+	"tailscale.com/tailcfg"
+)
+
+func TestUndeltaPeers(t *testing.T) {
+	n := func(id tailcfg.NodeID, name string) *tailcfg.Node {
+		return &tailcfg.Node{ID: id, Name: name}
+	}
+	peers := func(nv ...*tailcfg.Node) []*tailcfg.Node { return nv }
+	tests := []struct {
+		name   string
+		mapRes *tailcfg.MapResponse
+		prev   []*tailcfg.Node
+		want   []*tailcfg.Node
+	}{
+		{
+			name: "full_peers",
+			mapRes: &tailcfg.MapResponse{
+				Peers: peers(n(1, "foo"), n(2, "bar")),
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "full_peers_ignores_deltas",
+			mapRes: &tailcfg.MapResponse{
+				Peers:        peers(n(1, "foo"), n(2, "bar")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo"), n(2, "bar")),
+		},
+		{
+			name: "add_and_update",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(0, "zero"), n(2, "bar2"), n(3, "three")),
+			},
+			want: peers(n(0, "zero"), n(1, "foo"), n(2, "bar2"), n(3, "three")),
+		},
+		{
+			name: "remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersRemoved: []tailcfg.NodeID{1},
+			},
+			want: peers(n(2, "bar")),
+		},
+		{
+			name: "add_and_remove",
+			prev: peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{
+				PeersChanged: peers(n(1, "foo2")),
+				PeersRemoved: []tailcfg.NodeID{2},
+			},
+			want: peers(n(1, "foo2")),
+		},
+		{
+			name:   "unchanged",
+			prev:   peers(n(1, "foo"), n(2, "bar")),
+			mapRes: &tailcfg.MapResponse{},
+			want:   peers(n(1, "foo"), n(2, "bar")),
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			undeltaPeers(tt.mapRes, tt.prev)
+			if !reflect.DeepEqual(tt.mapRes.Peers, tt.want) {
+				t.Errorf("wrong results\n got: %s\nwant: %s", formatNodes(tt.mapRes.Peers), formatNodes(tt.want))
+			}
+		})
+	}
+}
+
+func formatNodes(nodes []*tailcfg.Node) string {
+	var sb strings.Builder
+	for i, n := range nodes {
+		if i > 0 {
+			sb.WriteString(", ")
+		}
+		fmt.Fprintf(&sb, "(%d, %q)", n.ID, n.Name)
+	}
+	return sb.String()
+}

tailcfg/tailcfg.go

@@ -443,11 +443,12 @@ type RegisterResponse struct {
 type MapRequest struct {
 	Version     int    // current version is 4
 	Compress    string // "zstd" or "" (no compression)
-	KeepAlive   bool   // server sends keep-alives
+	KeepAlive   bool   // whether server should send keep-alives back to us
 	NodeKey     NodeKey
 	DiscoKey    DiscoKey
 	Endpoints   []string // caller's endpoints (IPv4 or IPv6)
 	IncludeIPv6 bool     // include IPv6 endpoints in returned Node Endpoints
+	DeltaPeers  bool     // whether the 2nd+ network map in response should be deltas, using PeersChanged, PeersRemoved
 	Stream      bool     // if true, multiple MapResponse objects are returned
 	Hostinfo    *Hostinfo
 
@@ -502,22 +503,36 @@ type DNSConfig struct {
 }
 
 type MapResponse struct {
-	KeepAlive bool // if set, all other fields are ignored
+	KeepAlive bool `json:",omitempty"` // if set, all other fields are ignored
 
 	// Networking
 	Node    *Node
-	Peers   []*Node
-	DERPMap *DERPMap
+	DERPMap *DERPMap `json:",omitempty"` // if non-empty, a change in the DERP map.
+
+	// Peers, if non-empty, is the complete list of peers.
+	// It will be set in the first MapResponse for a long-polled request/response.
+	// Subsequent responses will be delta-encoded if DeltaPeers was set in the request.
+	// If Peers is non-empty, PeersChanged and PeersRemoved should
+	// be ignored (and should be empty).
+	// Peers is always returned sorted by Node.ID.
+	Peers []*Node `json:",omitempty"`
+	// PeersChanged are the Nodes (identified by their ID) that
+	// have changed or been added since the past update on the
+	// HTTP response. It's only set if MapRequest.DeltaPeers was true.
+	// PeersChanged is always returned sorted by Node.ID.
+	PeersChanged []*Node `json:",omitempty"`
+	// PeersRemoved are the NodeIDs that are no longer in the peer list.
+	PeersRemoved []NodeID `json:",omitempty"`
 
 	// DNS is the same as DNSConfig.Nameservers.
 	//
 	// TODO(dmytro): should be sent in DNSConfig.Nameservers once clients have updated.
-	DNS []wgcfg.IP
+	DNS []wgcfg.IP `json:",omitempty"`
 	// SearchPaths are the same as DNSConfig.Domains.
 	//
 	// TODO(dmytro): should be sent in DNSConfig.Domains once clients have updated.
-	SearchPaths []string
-	DNSConfig   DNSConfig
+	SearchPaths []string  `json:",omitempty"`
+	DNSConfig   DNSConfig `json:",omitempty"`
 
 	// ACLs
 	Domain       string