summaryrefslogtreecommitdiff
path: root/node/Cluster.cpp
blob: 8afd3debb04fc285ba970f4e6e1ea25906ba67c9 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
/*
 * ZeroTier One - Network Virtualization Everywhere
 * Copyright (C) 2011-2015  ZeroTier, Inc.
 *
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 * --
 *
 * ZeroTier may be used and distributed under the terms of the GPLv3, which
 * are available at: http://www.gnu.org/licenses/gpl-3.0.html
 *
 * If you would like to embed ZeroTier into a commercial application or
 * redistribute it in a modified binary form, please contact ZeroTier Networks
 * LLC. Start here: http://www.zerotier.com/
 */

#ifdef ZT_ENABLE_CLUSTER

#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>

#include <algorithm>
#include <utility>

#include "../version.h"

#include "Cluster.hpp"
#include "RuntimeEnvironment.hpp"
#include "MulticastGroup.hpp"
#include "CertificateOfMembership.hpp"
#include "Salsa20.hpp"
#include "Poly1305.hpp"
#include "Packet.hpp"
#include "Identity.hpp"
#include "Peer.hpp"
#include "Switch.hpp"
#include "Node.hpp"

namespace ZeroTier {

static inline double _dist3d(int x1,int y1,int z1,int x2,int y2,int z2)
	throw()
{
	double dx = ((double)x2 - (double)x1);
	double dy = ((double)y2 - (double)y1);
	double dz = ((double)z2 - (double)z1);
	return sqrt((dx * dx) + (dy * dy) + (dz * dz));
}

Cluster::Cluster(
	const RuntimeEnvironment *renv,
	uint16_t id,
	const std::vector<InetAddress> &zeroTierPhysicalEndpoints,
	int32_t x,
	int32_t y,
	int32_t z,
	void (*sendFunction)(void *,unsigned int,const void *,unsigned int),
	void *sendFunctionArg,
	int (*addressToLocationFunction)(void *,const struct sockaddr_storage *,int *,int *,int *),
	void *addressToLocationFunctionArg) :
	RR(renv),
	_sendFunction(sendFunction),
	_sendFunctionArg(sendFunctionArg),
	_addressToLocationFunction(addressToLocationFunction),
	_addressToLocationFunctionArg(addressToLocationFunctionArg),
	_x(x),
	_y(y),
	_z(z),
	_id(id),
	_zeroTierPhysicalEndpoints(zeroTierPhysicalEndpoints),
	_members(new _Member[ZT_CLUSTER_MAX_MEMBERS])
{
	uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];

	// Generate master secret by hashing the secret from our Identity key pair
	RR->identity.sha512PrivateKey(_masterSecret);

	// Generate our inbound message key, which is the master secret XORed with our ID and hashed twice
	memcpy(stmp,_masterSecret,sizeof(stmp));
	stmp[0] ^= Utils::hton(id);
	SHA512::hash(stmp,stmp,sizeof(stmp));
	SHA512::hash(stmp,stmp,sizeof(stmp));
	memcpy(_key,stmp,sizeof(_key));
	Utils::burn(stmp,sizeof(stmp));
}

Cluster::~Cluster()
{
	Utils::burn(_masterSecret,sizeof(_masterSecret));
	Utils::burn(_key,sizeof(_key));
	delete [] _members;
}

void Cluster::handleIncomingStateMessage(const void *msg,unsigned int len)
{
	Buffer<ZT_CLUSTER_MAX_MESSAGE_LENGTH> dmsg;
	{
		// FORMAT: <[16] iv><[8] MAC><... data>
		if ((len < 24)||(len > ZT_CLUSTER_MAX_MESSAGE_LENGTH))
			return;

		// 16-byte IV: first 8 bytes XORed with key, last 8 bytes used as Salsa20 64-bit IV
		char keytmp[32];
		memcpy(keytmp,_key,32);
		for(int i=0;i<8;++i)
			keytmp[i] ^= reinterpret_cast<const char *>(msg)[i];
		Salsa20 s20(keytmp,256,reinterpret_cast<const char *>(msg) + 8);
		Utils::burn(keytmp,sizeof(keytmp));

		// One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
		char polykey[ZT_POLY1305_KEY_LEN];
		memset(polykey,0,sizeof(polykey));
		s20.encrypt12(polykey,polykey,sizeof(polykey));

		// Compute 16-byte MAC
		char mac[ZT_POLY1305_MAC_LEN];
		Poly1305::compute(mac,reinterpret_cast<const char *>(msg) + 24,len - 24,polykey);

		// Check first 8 bytes of MAC against 64-bit MAC in stream
		if (!Utils::secureEq(mac,reinterpret_cast<const char *>(msg) + 16,8))
			return;

		// Decrypt!
		dmsg.setSize(len - 24);
		s20.decrypt12(reinterpret_cast<const char *>(msg) + 24,const_cast<void *>(dmsg.data()),dmsg.size());
	}

	if (dmsg.size() < 4)
		return;
	const uint16_t fromMemberId = dmsg.at<uint16_t>(0);
	unsigned int ptr = 2;
	if (fromMemberId == _id)
		return;
	const uint16_t toMemberId = dmsg.at<uint16_t>(ptr);
	ptr += 2;
	if (toMemberId != _id)
		return;

	_Member &m = _members[fromMemberId];
	Mutex::Lock mlck(m.lock);

	try {
		while (ptr < dmsg.size()) {
			const unsigned int mlen = dmsg.at<uint16_t>(ptr); ptr += 2;
			const unsigned int nextPtr = ptr + mlen;

			int mtype = -1;
			try {
				switch((StateMessageType)(mtype = (int)dmsg[ptr++])) {
					default:
						break;

					case STATE_MESSAGE_ALIVE: {
						ptr += 7; // skip version stuff, not used yet
						m.x = dmsg.at<int32_t>(ptr); ptr += 4;
						m.y = dmsg.at<int32_t>(ptr); ptr += 4;
						m.z = dmsg.at<int32_t>(ptr); ptr += 4;
						ptr += 8; // skip local clock, not used
						m.load = dmsg.at<uint64_t>(ptr); ptr += 8;
						ptr += 8; // skip flags, unused
						unsigned int physicalAddressCount = dmsg[ptr++];
						for(unsigned int i=0;i<physicalAddressCount;++i) {
							m.zeroTierPhysicalEndpoints.push_back(InetAddress());
							ptr += m.zeroTierPhysicalEndpoints.back().deserialize(dmsg,ptr);
							if (!(m.zeroTierPhysicalEndpoints.back()))
								m.zeroTierPhysicalEndpoints.pop_back();
						}
						m.lastReceivedAliveAnnouncement = RR->node->now();
					}	break;

					case STATE_MESSAGE_HAVE_PEER: {
						try {
							Identity id;
							ptr += id.deserialize(dmsg,ptr);
							if (id) {
								RR->topology->saveIdentity(id);

								{	// Add or update peer affinity entry
									_PeerAffinity pa(id.address(),fromMemberId,RR->node->now());
									Mutex::Lock _l2(_peerAffinities_m);
									std::vector<_PeerAffinity>::iterator i(std::lower_bound(_peerAffinities.begin(),_peerAffinities.end(),pa)); // O(log(n))
									if ((i != _peerAffinities.end())&&(i->key == pa.key)) {
										i->timestamp = pa.timestamp;
									} else {
										_peerAffinities.push_back(pa);
										std::sort(_peerAffinities.begin(),_peerAffinities.end()); // probably a more efficient way to insert but okay for now
									}
		 						}
		 					}
						} catch ( ... ) {
							// ignore invalid identities
						}
					}	break;

					case STATE_MESSAGE_MULTICAST_LIKE: {
						const uint64_t nwid = dmsg.at<uint64_t>(ptr); ptr += 8;
						const Address address(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
						const MAC mac(dmsg.field(ptr,6),6); ptr += 6;
						const uint32_t adi = dmsg.at<uint32_t>(ptr); ptr += 4;
						RR->mc->add(RR->node->now(),nwid,MulticastGroup(mac,adi),address);
					}	break;

					case STATE_MESSAGE_COM: {
						// TODO: not used yet
					}	break;

					case STATE_MESSAGE_RELAY: {
						const unsigned int numRemotePeerPaths = dmsg[ptr++];
						InetAddress remotePeerPaths[256]; // size is 8-bit, so 256 is max
						for(unsigned int i=0;i<numRemotePeerPaths;++i)
							ptr += remotePeerPaths[i].deserialize(dmsg,ptr);
						const unsigned int packetLen = dmsg.at<uint16_t>(ptr); ptr += 2;
						const void *packet = (const void *)dmsg.field(ptr,packetLen); ptr += packetLen;

						if (packetLen >= ZT_PROTO_MIN_FRAGMENT_LENGTH) { // ignore anything too short to contain a dest address
							const Address destinationAddress(reinterpret_cast<const char *>(packet) + 8,ZT_ADDRESS_LENGTH);
							SharedPtr<Peer> destinationPeer(RR->topology->getPeer(destinationAddress));
							if (destinationPeer) {
								if (
								    (destinationPeer->send(RR,packet,packetLen,RR->node->now()))&&
								    (numRemotePeerPaths > 0)&&
								    (packetLen >= 18)&&
								    (reinterpret_cast<const unsigned char *>(packet)[ZT_PACKET_FRAGMENT_IDX_FRAGMENT_INDICATOR] == ZT_PACKET_FRAGMENT_INDICATOR)
								   ) {
									// If remote peer paths were sent with this relayed packet, we do
									// RENDEZVOUS. It's handled here for cluster-relayed packets since
									// we don't have both Peer records so this is a different path.

									const Address remotePeerAddress(reinterpret_cast<const char *>(packet) + 13,ZT_ADDRESS_LENGTH);

									InetAddress bestDestV4,bestDestV6;
									destinationPeer->getBestActiveAddresses(RR->node->now(),bestDestV4,bestDestV6);
									InetAddress bestRemoteV4,bestRemoteV6;
									for(unsigned int i=0;i<numRemotePeerPaths;++i) {
										if ((bestRemoteV4)&&(bestRemoteV6))
											break;
										switch(remotePeerPaths[i].ss_family) {
											case AF_INET:
												if (!bestRemoteV4)
													bestRemoteV4 = remotePeerPaths[i];
												break;
											case AF_INET6:
												if (!bestRemoteV6)
													bestRemoteV6 = remotePeerPaths[i];
												break;
										}
									}

									Packet rendezvousForDest(destinationAddress,RR->identity.address(),Packet::VERB_RENDEZVOUS);
									rendezvousForDest.append((uint8_t)0);
									remotePeerAddress.appendTo(rendezvousForDest);

									Buffer<2048> rendezvousForOtherEnd;
									remotePeerAddress.appendTo(rendezvousForOtherEnd);
									rendezvousForOtherEnd.append((uint8_t)Packet::VERB_RENDEZVOUS);
									const unsigned int rendezvousForOtherEndPayloadSizePtr = rendezvousForOtherEnd.size();
									rendezvousForOtherEnd.addSize(2); // space for actual packet payload length
									rendezvousForOtherEnd.append((uint8_t)0); // flags == 0
									destinationAddress.appendTo(rendezvousForOtherEnd);

									bool haveMatch = false;
									if ((bestDestV6)&&(bestRemoteV6)) {
										haveMatch = true;

										rendezvousForDest.append((uint16_t)bestRemoteV6.port());
										rendezvousForDest.append((uint8_t)16);
										rendezvousForDest.append(bestRemoteV6.rawIpData(),16);

										rendezvousForOtherEnd.append((uint16_t)bestDestV6.port());
										rendezvousForOtherEnd.append((uint8_t)16);
										rendezvousForOtherEnd.append(bestDestV6.rawIpData(),16);
										rendezvousForOtherEnd.setAt<uint16_t>(rendezvousForOtherEndPayloadSizePtr,(uint16_t)(9 + 16));
									} else if ((bestDestV4)&&(bestRemoteV4)) {
										haveMatch = true;

										rendezvousForDest.append((uint16_t)bestRemoteV4.port());
										rendezvousForDest.append((uint8_t)4);
										rendezvousForDest.append(bestRemoteV4.rawIpData(),4);

										rendezvousForOtherEnd.append((uint16_t)bestDestV4.port());
										rendezvousForOtherEnd.append((uint8_t)4);
										rendezvousForOtherEnd.append(bestDestV4.rawIpData(),4);
										rendezvousForOtherEnd.setAt<uint16_t>(rendezvousForOtherEndPayloadSizePtr,(uint16_t)(9 + 4));
									}

									if (haveMatch) {
										_send(fromMemberId,STATE_MESSAGE_PROXY_SEND,rendezvousForOtherEnd.data(),rendezvousForOtherEnd.size());
										RR->sw->send(rendezvousForDest,true,0);
									}
								}
							}
						}
					}	break;

					case STATE_MESSAGE_PROXY_SEND: {
						const Address rcpt(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH);
						const Packet::Verb verb = (Packet::Verb)dmsg[ptr++];
						const unsigned int len = dmsg.at<uint16_t>(ptr); ptr += 2;
						Packet outp(rcpt,RR->identity.address(),verb);
						outp.append(dmsg.field(ptr,len),len);
						RR->sw->send(outp,true,0);
					}	break;
				}
			} catch ( ... ) {
				TRACE("invalid message of size %u type %d (inner decode), discarding",mlen,mtype);
				// drop invalids
			}

			ptr = nextPtr;
		}
	} catch ( ... ) {
		TRACE("invalid message (outer loop), discarding");
		// drop invalids
	}
}

void Cluster::replicateHavePeer(const Identity &peerId)
{
	{	// Use peer affinity table to track our own last announce time for peers
		_PeerAffinity pa(peerId.address(),_id,RR->node->now());
		Mutex::Lock _l2(_peerAffinities_m);
		std::vector<_PeerAffinity>::iterator i(std::lower_bound(_peerAffinities.begin(),_peerAffinities.end(),pa)); // O(log(n))
		if ((i != _peerAffinities.end())&&(i->key == pa.key)) {
			if ((pa.timestamp - i->timestamp) >= ZT_CLUSTER_HAVE_PEER_ANNOUNCE_PERIOD) {
				i->timestamp = pa.timestamp;
				// continue to announcement
			} else {
				// we've already announced this peer recently, so skip
				return;
			}
		} else {
			_peerAffinities.push_back(pa);
			std::sort(_peerAffinities.begin(),_peerAffinities.end()); // probably a more efficient way to insert but okay for now
			// continue to announcement
		}
	}

	// announcement
	Buffer<4096> buf;
	peerId.serialize(buf,false);
	{
		Mutex::Lock _l(_memberIds_m);
		for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
			Mutex::Lock _l2(_members[*mid].lock);
			_send(*mid,STATE_MESSAGE_HAVE_PEER,buf.data(),buf.size());
		}
	}
}

void Cluster::replicateMulticastLike(uint64_t nwid,const Address &peerAddress,const MulticastGroup &group)
{
	Buffer<4096> buf;
	buf.append((uint64_t)nwid);
	peerAddress.appendTo(buf);
	group.mac().appendTo(buf);
	buf.append((uint32_t)group.adi());
	{
		Mutex::Lock _l(_memberIds_m);
		for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
			Mutex::Lock _l2(_members[*mid].lock);
			_send(*mid,STATE_MESSAGE_MULTICAST_LIKE,buf.data(),buf.size());
		}
	}
}

void Cluster::replicateCertificateOfNetworkMembership(const CertificateOfMembership &com)
{
	Buffer<4096> buf;
	com.serialize(buf);
	{
		Mutex::Lock _l(_memberIds_m);
		for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
			Mutex::Lock _l2(_members[*mid].lock);
			_send(*mid,STATE_MESSAGE_COM,buf.data(),buf.size());
		}
	}
}

void Cluster::doPeriodicTasks()
{
	const uint64_t now = RR->node->now();

	{
		Mutex::Lock _l(_memberIds_m);
		for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
			Mutex::Lock _l2(_members[*mid].lock);

			if ((now - _members[*mid].lastAnnouncedAliveTo) >= ((ZT_CLUSTER_TIMEOUT / 2) - 1000)) {
				Buffer<2048> alive;
				alive.append((uint16_t)ZEROTIER_ONE_VERSION_MAJOR);
				alive.append((uint16_t)ZEROTIER_ONE_VERSION_MINOR);
				alive.append((uint16_t)ZEROTIER_ONE_VERSION_REVISION);
				alive.append((uint8_t)ZT_PROTO_VERSION);
				if (_addressToLocationFunction) {
					alive.append((int32_t)_x);
					alive.append((int32_t)_y);
					alive.append((int32_t)_z);
				} else {
					alive.append((int32_t)0);
					alive.append((int32_t)0);
					alive.append((int32_t)0);
				}
				alive.append((uint64_t)now);
				alive.append((uint64_t)0); // TODO: compute and send load average
				alive.append((uint64_t)0); // unused/reserved flags
				alive.append((uint8_t)_zeroTierPhysicalEndpoints.size());
				for(std::vector<InetAddress>::const_iterator pe(_zeroTierPhysicalEndpoints.begin());pe!=_zeroTierPhysicalEndpoints.end();++pe)
					pe->serialize(alive);
				_send(*mid,STATE_MESSAGE_ALIVE,alive.data(),alive.size());
				_members[*mid].lastAnnouncedAliveTo = now;
			}

			_flush(*mid); // does nothing if nothing to flush
		}
	}
}

void Cluster::addMember(uint16_t memberId)
{
	if (memberId >= ZT_CLUSTER_MAX_MEMBERS)
		return;

	Mutex::Lock _l2(_members[memberId].lock);

	{
		Mutex::Lock _l(_memberIds_m);
		if (std::find(_memberIds.begin(),_memberIds.end(),memberId) != _memberIds.end())
			return;
		_memberIds.push_back(memberId);
		std::sort(_memberIds.begin(),_memberIds.end());
	}

	_members[memberId].clear();

	// Generate this member's message key from the master and its ID
	uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];
	memcpy(stmp,_masterSecret,sizeof(stmp));
	stmp[0] ^= Utils::hton(memberId);
	SHA512::hash(stmp,stmp,sizeof(stmp));
	SHA512::hash(stmp,stmp,sizeof(stmp));
	memcpy(_members[memberId].key,stmp,sizeof(_members[memberId].key));
	Utils::burn(stmp,sizeof(stmp));

	// Prepare q
	_members[memberId].q.clear();
	char iv[16];
	Utils::getSecureRandom(iv,16);
	_members[memberId].q.append(iv,16);
	_members[memberId].q.addSize(8); // room for MAC
	_members[memberId].q.append((uint16_t)_id);
	_members[memberId].q.append((uint16_t)memberId);
}

void Cluster::removeMember(uint16_t memberId)
{
	Mutex::Lock _l(_memberIds_m);
	std::vector<uint16_t> newMemberIds;
	for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
		if (*mid != memberId)
			newMemberIds.push_back(*mid);
	}
	_memberIds = newMemberIds;
}

bool Cluster::redirectPeer(const SharedPtr<Peer> &peer,const InetAddress &peerPhysicalAddress,bool offload)
{
	if (!peerPhysicalAddress) // sanity check
		return false;
	if (_addressToLocationFunction) {
		// Pick based on location if it can be determined
		int px = 0,py = 0,pz = 0;
		if (_addressToLocationFunction(_addressToLocationFunctionArg,reinterpret_cast<const struct sockaddr_storage *>(&peerPhysicalAddress),&px,&py,&pz) == 0) {
			// No geo-info so no change
			return false;
		}

		// Find member closest to this peer
		const uint64_t now = RR->node->now();
		std::vector<InetAddress> best; // initial "best" is for peer to stay put
		const double currentDistance = _dist3d(_x,_y,_z,px,py,pz);
		double bestDistance = (offload ? 2147483648.0 : currentDistance);
		unsigned int bestMember = _id;
		{
			Mutex::Lock _l(_memberIds_m);
			for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
				_Member &m = _members[*mid];
				Mutex::Lock _ml(m.lock);

				// Consider member if it's alive and has sent us a location and one or more physical endpoints to send peers to
				if ( ((now - m.lastReceivedAliveAnnouncement) < ZT_CLUSTER_TIMEOUT) && ((m.x != 0)||(m.y != 0)||(m.z != 0)) && (m.zeroTierPhysicalEndpoints.size() > 0) ) {
					double mdist = _dist3d(m.x,m.y,m.z,px,py,pz);
					if (mdist < bestDistance) {
						bestMember = *mid;
						best = m.zeroTierPhysicalEndpoints;
					}
				}
			}
		}

		if (best.size() > 0) {
			TRACE("peer %s is at [%d,%d,%d], distance to us is %f, sending to %u instead for better distance %f",peer->address().toString().c_str(),px,py,pz,currentDistance,bestMember,bestDistance);

			/* if (peer->remoteVersionProtocol() >= 5) {
				// If it's a newer peer send VERB_PUSH_DIRECT_PATHS which is more idiomatic
			} else { */
				// Otherwise send VERB_RENDEZVOUS for ourselves, which will trick peers into trying other endpoints for us even if they're too old for PUSH_DIRECT_PATHS
				for(std::vector<InetAddress>::const_iterator a(best.begin());a!=best.end();++a) {
					if ((a->ss_family == AF_INET)||(a->ss_family == AF_INET6)) {
						Packet outp(peer->address(),RR->identity.address(),Packet::VERB_RENDEZVOUS);
						outp.append((uint8_t)0); // no flags
						RR->identity.address().appendTo(outp); // HACK: rendezvous with ourselves! with really old peers this will only work if I'm a root server!
						outp.append((uint16_t)a->port());
						if (a->ss_family == AF_INET) {
							outp.append((uint8_t)4);
							outp.append(a->rawIpData(),4);
						} else {
							outp.append((uint8_t)16);
							outp.append(a->rawIpData(),16);
						}
						RR->sw->send(outp,true,0);
					}
				}
			//}

			return true;
		} else {
			TRACE("peer %s is at [%d,%d,%d], distance to us is %f and this seems to be the best",peer->address().toString().c_str(),px,py,pz,currentDistance);
			return false;
		}
	} else {
		// TODO: pick based on load if no location info?
		return false;
	}
}

void Cluster::_send(uint16_t memberId,StateMessageType type,const void *msg,unsigned int len)
{
	_Member &m = _members[memberId];
	// assumes m.lock is locked!
	if ((m.q.size() + len + 3) > ZT_CLUSTER_MAX_MESSAGE_LENGTH)
		_flush(memberId);
	m.q.append((uint16_t)(len + 1));
	m.q.append((uint8_t)type);
	m.q.append(msg,len);
}

void Cluster::_flush(uint16_t memberId)
{
	_Member &m = _members[memberId];
	// assumes m.lock is locked!
	if (m.q.size() > (24 + 2 + 2)) { // 16-byte IV + 8-byte MAC + 2 byte from-member-ID + 2 byte to-member-ID
		// Create key from member's key and IV
		char keytmp[32];
		memcpy(keytmp,m.key,32);
		for(int i=0;i<8;++i)
			keytmp[i] ^= m.q[i];
		Salsa20 s20(keytmp,256,m.q.field(8,8));
		Utils::burn(keytmp,sizeof(keytmp));

		// One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
		char polykey[ZT_POLY1305_KEY_LEN];
		memset(polykey,0,sizeof(polykey));
		s20.encrypt12(polykey,polykey,sizeof(polykey));

		// Encrypt m.q in place
		s20.encrypt12(reinterpret_cast<const char *>(m.q.data()) + 24,const_cast<char *>(reinterpret_cast<const char *>(m.q.data())) + 24,m.q.size() - 24);

		// Add MAC for authentication (encrypt-then-MAC)
		char mac[ZT_POLY1305_MAC_LEN];
		Poly1305::compute(mac,reinterpret_cast<const char *>(m.q.data()) + 24,m.q.size() - 24,polykey);
		memcpy(m.q.field(16,8),mac,8);

		// Send!
		_sendFunction(_sendFunctionArg,memberId,m.q.data(),m.q.size());

		// Prepare for more
		m.q.clear();
		char iv[16];
		Utils::getSecureRandom(iv,16);
		m.q.append(iv,16);
		m.q.addSize(8); // room for MAC
		m.q.append((uint16_t)_id); // from member ID
		m.q.append((uint16_t)memberId); // to member ID
	}
}

} // namespace ZeroTier

#endif // ZT_ENABLE_CLUSTER