server.go 10.4 KB
Newer Older
Z
zelig 已提交
1 2 3 4
package p2p

import (
	"bytes"
F
Felix Lange 已提交
5
	"crypto/ecdsa"
6
	"errors"
Z
zelig 已提交
7 8 9 10 11
	"fmt"
	"net"
	"sync"
	"time"

12
	"github.com/ethereum/go-ethereum/logger"
F
Felix Lange 已提交
13
	"github.com/ethereum/go-ethereum/p2p/discover"
14
	"github.com/ethereum/go-ethereum/p2p/nat"
15
	"github.com/ethereum/go-ethereum/rlp"
Z
zelig 已提交
16 17 18
)

const (
19 20
	defaultDialTimeout   = 10 * time.Second
	refreshPeersInterval = 30 * time.Second
F
Felix Lange 已提交
21 22 23 24 25 26 27 28 29

	// total timeout for encryption handshake and protocol
	// handshake in both directions.
	handshakeTimeout = 5 * time.Second
	// maximum time allowed for reading a complete message.
	// this is effectively the amount of time a connection can be idle.
	frameReadTimeout = 1 * time.Minute
	// maximum amount of time allowed for writing a complete message.
	frameWriteTimeout = 5 * time.Second
Z
zelig 已提交
30 31
)

32
var srvlog = logger.NewLogger("P2P Server")
33
var srvjslog = logger.NewJsonLogger()
Z
zelig 已提交
34

35 36 37 38 39
// Server manages all peer connections.
//
// The fields of Server are used as configuration parameters.
// You should set them before starting the Server. Fields may not be
// modified while the server is running.
Z
zelig 已提交
40
type Server struct {
F
Felix Lange 已提交
41 42
	// This field must be set to a valid secp256k1 private key.
	PrivateKey *ecdsa.PrivateKey
43 44 45 46 47

	// MaxPeers is the maximum number of peers that can be
	// connected. It must be greater than zero.
	MaxPeers int

F
Felix Lange 已提交
48
	// Name sets the node name of this server.
O
obscuren 已提交
49
	// Use common.MakeName to create a name that follows existing conventions.
F
Felix Lange 已提交
50 51 52 53
	Name string

	// Bootstrap nodes are used to establish connectivity
	// with the rest of the network.
54
	BootstrapNodes []*discover.Node
F
Felix Lange 已提交
55

56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71
	// Protocols should contain the protocols supported
	// by the server. Matching protocols are launched for
	// each peer.
	Protocols []Protocol

	// If ListenAddr is set to a non-nil address, the server
	// will listen for incoming connections.
	//
	// If the port is zero, the operating system will pick a port. The
	// ListenAddr field will be updated with the actual address when
	// the server is started.
	ListenAddr string

	// If set to a non-nil value, the given NAT port mapper
	// is used to make the listening port available to the
	// Internet.
72
	NAT nat.Interface
73 74 75 76 77 78 79 80

	// If Dialer is set to a non-nil value, the given Dialer
	// is used to dial outbound peer connections.
	Dialer *net.Dialer

	// If NoDial is true, the server will not dial any peers.
	NoDial bool

F
Felix Lange 已提交
81
	// Hooks for testing. These are useful because we can inhibit
82
	// the whole protocol stack.
F
Felix Lange 已提交
83
	setupFunc
F
Felix Lange 已提交
84
	newPeerHook
Z
zelig 已提交
85

F
Felix Lange 已提交
86 87
	ourHandshake *protoHandshake

F
Felix Lange 已提交
88 89 90 91 92 93 94 95 96 97 98
	lock     sync.RWMutex
	running  bool
	listener net.Listener
	peers    map[discover.NodeID]*Peer

	ntab *discover.Table

	quit        chan struct{}
	loopWG      sync.WaitGroup // {dial,listen,nat}Loop
	peerWG      sync.WaitGroup // active peer goroutines
	peerConnect chan *discover.Node
Z
zelig 已提交
99 100
}

F
Felix Lange 已提交
101
type setupFunc func(net.Conn, *ecdsa.PrivateKey, *protoHandshake, *discover.Node) (*conn, error)
F
Felix Lange 已提交
102
type newPeerHook func(*Peer)
Z
zelig 已提交
103

104 105 106 107 108
// Peers returns all connected peers.
func (srv *Server) Peers() (peers []*Peer) {
	srv.lock.RLock()
	defer srv.lock.RUnlock()
	for _, peer := range srv.peers {
Z
zelig 已提交
109 110 111 112 113 114 115
		if peer != nil {
			peers = append(peers, peer)
		}
	}
	return
}

116 117 118
// PeerCount returns the number of connected peers.
func (srv *Server) PeerCount() int {
	srv.lock.RLock()
F
Felix Lange 已提交
119 120 121
	n := len(srv.peers)
	srv.lock.RUnlock()
	return n
Z
zelig 已提交
122 123
}

F
Felix Lange 已提交
124 125
// SuggestPeer creates a connection to the given Node if it
// is not already connected.
126 127
func (srv *Server) SuggestPeer(n *discover.Node) {
	srv.peerConnect <- n
Z
zelig 已提交
128 129
}

130 131
// Broadcast sends an RLP-encoded message to all connected peers.
// This method is deprecated and will be removed later.
132
func (srv *Server) Broadcast(protocol string, code uint64, data interface{}) error {
F
Felix Lange 已提交
133 134
	var payload []byte
	if data != nil {
135 136 137 138 139
		var err error
		payload, err = rlp.EncodeToBytes(data)
		if err != nil {
			return err
		}
F
Felix Lange 已提交
140
	}
141 142 143
	srv.lock.RLock()
	defer srv.lock.RUnlock()
	for _, peer := range srv.peers {
Z
zelig 已提交
144
		if peer != nil {
F
Felix Lange 已提交
145 146 147 148 149
			var msg = Msg{Code: code}
			if data != nil {
				msg.Payload = bytes.NewReader(payload)
				msg.Size = uint32(len(payload))
			}
150
			peer.writeProtoMsg(protocol, msg)
Z
zelig 已提交
151 152
		}
	}
153
	return nil
Z
zelig 已提交
154 155
}

156 157 158 159 160 161 162 163 164 165
// Start starts running the server.
// Servers can be re-used and started again after stopping.
func (srv *Server) Start() (err error) {
	srv.lock.Lock()
	defer srv.lock.Unlock()
	if srv.running {
		return errors.New("server already running")
	}
	srvlog.Infoln("Starting Server")

166
	// static fields
F
Felix Lange 已提交
167 168
	if srv.PrivateKey == nil {
		return fmt.Errorf("Server.PrivateKey must be set to a non-nil key")
Z
zelig 已提交
169
	}
170 171 172 173
	if srv.MaxPeers <= 0 {
		return fmt.Errorf("Server.MaxPeers must be > 0")
	}
	srv.quit = make(chan struct{})
F
Felix Lange 已提交
174 175
	srv.peers = make(map[discover.NodeID]*Peer)
	srv.peerConnect = make(chan *discover.Node)
F
Felix Lange 已提交
176 177
	if srv.setupFunc == nil {
		srv.setupFunc = setupConn
178
	}
F
Felix Lange 已提交
179

180
	// node table
F
Felix Lange 已提交
181
	ntab, err := discover.ListenUDP(srv.PrivateKey, srv.ListenAddr, srv.NAT)
F
Felix Lange 已提交
182 183 184
	if err != nil {
		return err
	}
F
Felix Lange 已提交
185 186
	srv.ntab = ntab

187
	// handshake
188
	srv.ourHandshake = &protoHandshake{Version: baseProtocolVersion, Name: srv.Name, ID: ntab.Self().ID}
F
Felix Lange 已提交
189 190 191 192
	for _, p := range srv.Protocols {
		srv.ourHandshake.Caps = append(srv.ourHandshake.Caps, p.cap())
	}

193 194 195 196 197 198
	// listen/dial
	if srv.ListenAddr != "" {
		if err := srv.startListening(); err != nil {
			return err
		}
	}
F
Felix Lange 已提交
199 200 201
	if srv.Dialer == nil {
		srv.Dialer = &net.Dialer{Timeout: defaultDialTimeout}
	}
202
	if !srv.NoDial {
F
Felix Lange 已提交
203
		srv.loopWG.Add(1)
204 205 206 207 208 209 210 211
		go srv.dialLoop()
	}
	if srv.NoDial && srv.ListenAddr == "" {
		srvlog.Warnln("I will be kind-of useless, neither dialing nor listening.")
	}

	srv.running = true
	return nil
Z
zelig 已提交
212 213
}

214 215 216 217 218
func (srv *Server) startListening() error {
	listener, err := net.Listen("tcp", srv.ListenAddr)
	if err != nil {
		return err
	}
219 220
	laddr := listener.Addr().(*net.TCPAddr)
	srv.ListenAddr = laddr.String()
221
	srv.listener = listener
F
Felix Lange 已提交
222
	srv.loopWG.Add(1)
223
	go srv.listenLoop()
224
	if !laddr.IP.IsLoopback() && srv.NAT != nil {
F
Felix Lange 已提交
225
		srv.loopWG.Add(1)
226 227 228 229
		go func() {
			nat.Map(srv.NAT, srv.quit, "tcp", laddr.Port, laddr.Port, "ethereum p2p")
			srv.loopWG.Done()
		}()
230 231 232 233 234 235 236 237 238 239 240
	}
	return nil
}

// Stop terminates the server and all active peer connections.
// It blocks until all active connections have been closed.
func (srv *Server) Stop() {
	srv.lock.Lock()
	if !srv.running {
		srv.lock.Unlock()
		return
Z
zelig 已提交
241
	}
242 243 244
	srv.running = false
	srv.lock.Unlock()

F
Felix Lange 已提交
245 246
	srvlog.Infoln("Stopping Server")
	srv.ntab.Close()
247 248 249 250 251
	if srv.listener != nil {
		// this unblocks listener Accept
		srv.listener.Close()
	}
	close(srv.quit)
F
Felix Lange 已提交
252
	srv.loopWG.Wait()
253

F
Felix Lange 已提交
254 255 256 257 258
	// No new peers can be added at this point because dialLoop and
	// listenLoop are down. It is safe to call peerWG.Wait because
	// peerWG.Add is not called outside of those loops.
	for _, peer := range srv.peers {
		peer.Disconnect(DiscQuitting)
259
	}
F
Felix Lange 已提交
260
	srv.peerWG.Wait()
261
}
Z
zelig 已提交
262

263 264
// main loop for adding connections via listening
func (srv *Server) listenLoop() {
F
Felix Lange 已提交
265
	defer srv.loopWG.Done()
266
	srvlog.Infoln("Listening on", srv.listener.Addr())
Z
zelig 已提交
267
	for {
F
Felix Lange 已提交
268 269
		conn, err := srv.listener.Accept()
		if err != nil {
270
			return
Z
zelig 已提交
271
		}
F
Felix Lange 已提交
272 273 274
		srvlog.Debugf("Accepted conn %v\n", conn.RemoteAddr())
		srv.peerWG.Add(1)
		go srv.startPeer(conn, nil)
Z
zelig 已提交
275 276 277
	}
}

278
func (srv *Server) dialLoop() {
F
Felix Lange 已提交
279 280 281 282 283 284 285 286 287 288 289 290 291 292
	defer srv.loopWG.Done()
	refresh := time.NewTicker(refreshPeersInterval)
	defer refresh.Stop()

	srv.ntab.Bootstrap(srv.BootstrapNodes)
	go srv.findPeers()

	dialed := make(chan *discover.Node)
	dialing := make(map[discover.NodeID]bool)

	// TODO: limit number of active dials
	// TODO: ensure only one findPeers goroutine is running
	// TODO: pause findPeers when we're at capacity

Z
zelig 已提交
293 294
	for {
		select {
F
Felix Lange 已提交
295
		case <-refresh.C:
296

F
Felix Lange 已提交
297 298 299
			go srv.findPeers()

		case dest := <-srv.peerConnect:
F
Felix Lange 已提交
300 301 302
			// avoid dialing nodes that are already connected.
			// there is another check for this in addPeer,
			// which runs after the handshake.
F
Felix Lange 已提交
303 304 305
			srv.lock.Lock()
			_, isconnected := srv.peers[dest.ID]
			srv.lock.Unlock()
306
			if isconnected || dialing[dest.ID] || dest.ID == srv.Self().ID {
F
Felix Lange 已提交
307
				continue
Z
zelig 已提交
308
			}
F
Felix Lange 已提交
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323

			dialing[dest.ID] = true
			srv.peerWG.Add(1)
			go func() {
				srv.dialNode(dest)
				// at this point, the peer has been added
				// or discarded. either way, we're not dialing it anymore.
				dialed <- dest
			}()

		case dest := <-dialed:
			delete(dialing, dest.ID)

		case <-srv.quit:
			// TODO: maybe wait for active dials
Z
zelig 已提交
324 325 326 327 328
			return
		}
	}
}

F
Felix Lange 已提交
329
func (srv *Server) dialNode(dest *discover.Node) {
330 331 332
	addr := &net.TCPAddr{IP: dest.IP, Port: dest.TCPPort}
	srvlog.Debugf("Dialing %v\n", dest)
	conn, err := srv.Dialer.Dial("tcp", addr.String())
Z
zelig 已提交
333
	if err != nil {
334
		srvlog.DebugDetailf("dial error: %v", err)
F
Felix Lange 已提交
335
		return
Z
zelig 已提交
336
	}
F
Felix Lange 已提交
337
	srv.startPeer(conn, dest)
Z
zelig 已提交
338 339
}

340 341 342 343
func (srv *Server) Self() *discover.Node {
	return srv.ntab.Self()
}

F
Felix Lange 已提交
344
func (srv *Server) findPeers() {
345
	far := srv.Self().ID
F
Felix Lange 已提交
346 347 348
	for i := range far {
		far[i] = ^far[i]
	}
349
	closeToSelf := srv.ntab.Lookup(srv.Self().ID)
F
Felix Lange 已提交
350 351 352 353 354 355 356 357 358 359 360 361
	farFromSelf := srv.ntab.Lookup(far)

	for i := 0; i < len(closeToSelf) || i < len(farFromSelf); i++ {
		if i < len(closeToSelf) {
			srv.peerConnect <- closeToSelf[i]
		}
		if i < len(farFromSelf) {
			srv.peerConnect <- farFromSelf[i]
		}
	}
}

F
Felix Lange 已提交
362
func (srv *Server) startPeer(fd net.Conn, dest *discover.Node) {
363
	// TODO: handle/store session token
F
Felix Lange 已提交
364
	fd.SetDeadline(time.Now().Add(handshakeTimeout))
F
Felix Lange 已提交
365
	conn, err := srv.setupFunc(fd, srv.PrivateKey, srv.ourHandshake, dest)
F
Felix Lange 已提交
366
	if err != nil {
F
Felix Lange 已提交
367 368
		fd.Close()
		srvlog.Debugf("Handshake with %v failed: %v", fd.RemoteAddr(), err)
F
Felix Lange 已提交
369 370
		return
	}
F
Felix Lange 已提交
371 372 373 374 375

	conn.MsgReadWriter = &netWrapper{
		wrapped: conn.MsgReadWriter,
		conn:    fd, rtimeout: frameReadTimeout, wtimeout: frameWriteTimeout,
	}
F
Felix Lange 已提交
376
	p := newPeer(fd, conn, srv.Protocols)
F
Felix Lange 已提交
377
	if ok, reason := srv.addPeer(conn.ID, p); !ok {
378 379
		srvlog.DebugDetailf("Not adding %v (%v)\n", p, reason)
		p.politeDisconnect(reason)
F
Felix Lange 已提交
380
		return
Z
zelig 已提交
381
	}
382

383
	srvlog.Debugf("Added %v\n", p)
384 385
	srvjslog.LogJson(&logger.P2PConnected{
		RemoteId:            fmt.Sprintf("%x", conn.ID[:]),
F
Felix Lange 已提交
386
		RemoteAddress:       fd.RemoteAddr().String(),
387 388 389
		RemoteVersionString: conn.Name,
		NumConnections:      srv.PeerCount(),
	})
F
Felix Lange 已提交
390

F
Felix Lange 已提交
391 392 393
	if srv.newPeerHook != nil {
		srv.newPeerHook(p)
	}
394
	discreason := p.run()
F
Felix Lange 已提交
395
	srv.removePeer(p)
396

397
	srvlog.Debugf("Removed %v (%v)\n", p, discreason)
398 399 400 401
	srvjslog.LogJson(&logger.P2PDisconnected{
		RemoteId:       fmt.Sprintf("%x", conn.ID[:]),
		NumConnections: srv.PeerCount(),
	})
Z
zelig 已提交
402 403
}

F
Felix Lange 已提交
404
func (srv *Server) addPeer(id discover.NodeID, p *Peer) (bool, DiscReason) {
405 406
	srv.lock.Lock()
	defer srv.lock.Unlock()
F
Felix Lange 已提交
407 408 409 410 411 412 413
	switch {
	case !srv.running:
		return false, DiscQuitting
	case len(srv.peers) >= srv.MaxPeers:
		return false, DiscTooManyPeers
	case srv.peers[id] != nil:
		return false, DiscAlreadyConnected
414
	case id == srv.Self().ID:
F
Felix Lange 已提交
415 416 417 418
		return false, DiscSelf
	}
	srv.peers[id] = p
	return true, 0
Z
zelig 已提交
419 420
}

F
Felix Lange 已提交
421 422
func (srv *Server) removePeer(p *Peer) {
	srv.lock.Lock()
F
Felix Lange 已提交
423
	delete(srv.peers, p.ID())
F
Felix Lange 已提交
424 425
	srv.lock.Unlock()
	srv.peerWG.Done()
F
Felix Lange 已提交
426
}