Cluster.cpp 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716
  1. /*
  2. * ZeroTier One - Network Virtualization Everywhere
  3. * Copyright (C) 2011-2015 ZeroTier, Inc.
  4. *
  5. * This program is free software: you can redistribute it and/or modify
  6. * it under the terms of the GNU General Public License as published by
  7. * the Free Software Foundation, either version 3 of the License, or
  8. * (at your option) any later version.
  9. *
  10. * This program is distributed in the hope that it will be useful,
  11. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13. * GNU General Public License for more details.
  14. *
  15. * You should have received a copy of the GNU General Public License
  16. * along with this program. If not, see <http://www.gnu.org/licenses/>.
  17. *
  18. * --
  19. *
  20. * ZeroTier may be used and distributed under the terms of the GPLv3, which
  21. * are available at: http://www.gnu.org/licenses/gpl-3.0.html
  22. *
  23. * If you would like to embed ZeroTier into a commercial application or
  24. * redistribute it in a modified binary form, please contact ZeroTier Networks
  25. * LLC. Start here: http://www.zerotier.com/
  26. */
  27. #ifdef ZT_ENABLE_CLUSTER
  28. #include <stdint.h>
  29. #include <stdio.h>
  30. #include <stdlib.h>
  31. #include <string.h>
  32. #include <math.h>
  33. #include <algorithm>
  34. #include <utility>
  35. #include "../version.h"
  36. #include "Cluster.hpp"
  37. #include "RuntimeEnvironment.hpp"
  38. #include "MulticastGroup.hpp"
  39. #include "CertificateOfMembership.hpp"
  40. #include "Salsa20.hpp"
  41. #include "Poly1305.hpp"
  42. #include "Identity.hpp"
  43. #include "Topology.hpp"
  44. #include "Packet.hpp"
  45. #include "Switch.hpp"
  46. #include "Node.hpp"
  47. namespace ZeroTier {
  48. static inline double _dist3d(int x1,int y1,int z1,int x2,int y2,int z2)
  49. throw()
  50. {
  51. double dx = ((double)x2 - (double)x1);
  52. double dy = ((double)y2 - (double)y1);
  53. double dz = ((double)z2 - (double)z1);
  54. return sqrt((dx * dx) + (dy * dy) + (dz * dz));
  55. }
  56. Cluster::Cluster(
  57. const RuntimeEnvironment *renv,
  58. uint16_t id,
  59. const std::vector<InetAddress> &zeroTierPhysicalEndpoints,
  60. int32_t x,
  61. int32_t y,
  62. int32_t z,
  63. void (*sendFunction)(void *,unsigned int,const void *,unsigned int),
  64. void *sendFunctionArg,
  65. int (*addressToLocationFunction)(void *,const struct sockaddr_storage *,int *,int *,int *),
  66. void *addressToLocationFunctionArg) :
  67. RR(renv),
  68. _sendFunction(sendFunction),
  69. _sendFunctionArg(sendFunctionArg),
  70. _addressToLocationFunction(addressToLocationFunction),
  71. _addressToLocationFunctionArg(addressToLocationFunctionArg),
  72. _x(x),
  73. _y(y),
  74. _z(z),
  75. _id(id),
  76. _zeroTierPhysicalEndpoints(zeroTierPhysicalEndpoints),
  77. _members(new _Member[ZT_CLUSTER_MAX_MEMBERS]),
  78. _peerAffinities(65536)
  79. {
  80. uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];
  81. // Generate master secret by hashing the secret from our Identity key pair
  82. RR->identity.sha512PrivateKey(_masterSecret);
  83. // Generate our inbound message key, which is the master secret XORed with our ID and hashed twice
  84. memcpy(stmp,_masterSecret,sizeof(stmp));
  85. stmp[0] ^= Utils::hton(id);
  86. SHA512::hash(stmp,stmp,sizeof(stmp));
  87. SHA512::hash(stmp,stmp,sizeof(stmp));
  88. memcpy(_key,stmp,sizeof(_key));
  89. Utils::burn(stmp,sizeof(stmp));
  90. }
  91. Cluster::~Cluster()
  92. {
  93. Utils::burn(_masterSecret,sizeof(_masterSecret));
  94. Utils::burn(_key,sizeof(_key));
  95. delete [] _members;
  96. }
  97. void Cluster::handleIncomingStateMessage(const void *msg,unsigned int len)
  98. {
  99. Buffer<ZT_CLUSTER_MAX_MESSAGE_LENGTH> dmsg;
  100. {
  101. // FORMAT: <[16] iv><[8] MAC><... data>
  102. if ((len < 24)||(len > ZT_CLUSTER_MAX_MESSAGE_LENGTH))
  103. return;
  104. // 16-byte IV: first 8 bytes XORed with key, last 8 bytes used as Salsa20 64-bit IV
  105. char keytmp[32];
  106. memcpy(keytmp,_key,32);
  107. for(int i=0;i<8;++i)
  108. keytmp[i] ^= reinterpret_cast<const char *>(msg)[i];
  109. Salsa20 s20(keytmp,256,reinterpret_cast<const char *>(msg) + 8);
  110. Utils::burn(keytmp,sizeof(keytmp));
  111. // One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
  112. char polykey[ZT_POLY1305_KEY_LEN];
  113. memset(polykey,0,sizeof(polykey));
  114. s20.encrypt12(polykey,polykey,sizeof(polykey));
  115. // Compute 16-byte MAC
  116. char mac[ZT_POLY1305_MAC_LEN];
  117. Poly1305::compute(mac,reinterpret_cast<const char *>(msg) + 24,len - 24,polykey);
  118. // Check first 8 bytes of MAC against 64-bit MAC in stream
  119. if (!Utils::secureEq(mac,reinterpret_cast<const char *>(msg) + 16,8))
  120. return;
  121. // Decrypt!
  122. dmsg.setSize(len - 24);
  123. s20.decrypt12(reinterpret_cast<const char *>(msg) + 24,const_cast<void *>(dmsg.data()),dmsg.size());
  124. }
  125. if (dmsg.size() < 4)
  126. return;
  127. const uint16_t fromMemberId = dmsg.at<uint16_t>(0);
  128. unsigned int ptr = 2;
  129. if (fromMemberId == _id) // sanity check: we don't talk to ourselves
  130. return;
  131. const uint16_t toMemberId = dmsg.at<uint16_t>(ptr);
  132. ptr += 2;
  133. if (toMemberId != _id) // sanity check: message not for us?
  134. return;
  135. { // make sure sender is actually considered a member
  136. Mutex::Lock _l3(_memberIds_m);
  137. if (std::find(_memberIds.begin(),_memberIds.end(),fromMemberId) == _memberIds.end())
  138. return;
  139. }
  140. {
  141. _Member &m = _members[fromMemberId];
  142. Mutex::Lock mlck(m.lock);
  143. try {
  144. while (ptr < dmsg.size()) {
  145. const unsigned int mlen = dmsg.at<uint16_t>(ptr); ptr += 2;
  146. const unsigned int nextPtr = ptr + mlen;
  147. if (nextPtr > dmsg.size())
  148. break;
  149. int mtype = -1;
  150. try {
  151. switch((StateMessageType)(mtype = (int)dmsg[ptr++])) {
  152. default:
  153. break;
  154. case STATE_MESSAGE_ALIVE: {
  155. ptr += 7; // skip version stuff, not used yet
  156. m.x = dmsg.at<int32_t>(ptr); ptr += 4;
  157. m.y = dmsg.at<int32_t>(ptr); ptr += 4;
  158. m.z = dmsg.at<int32_t>(ptr); ptr += 4;
  159. ptr += 8; // skip local clock, not used
  160. m.load = dmsg.at<uint64_t>(ptr); ptr += 8;
  161. ptr += 8; // skip flags, unused
  162. #ifdef ZT_TRACE
  163. std::string addrs;
  164. #endif
  165. unsigned int physicalAddressCount = dmsg[ptr++];
  166. m.zeroTierPhysicalEndpoints.clear();
  167. for(unsigned int i=0;i<physicalAddressCount;++i) {
  168. m.zeroTierPhysicalEndpoints.push_back(InetAddress());
  169. ptr += m.zeroTierPhysicalEndpoints.back().deserialize(dmsg,ptr);
  170. if (!(m.zeroTierPhysicalEndpoints.back())) {
  171. m.zeroTierPhysicalEndpoints.pop_back();
  172. }
  173. #ifdef ZT_TRACE
  174. else {
  175. if (addrs.length() > 0)
  176. addrs.push_back(',');
  177. addrs.append(m.zeroTierPhysicalEndpoints.back().toString());
  178. }
  179. #endif
  180. }
  181. #ifdef ZT_TRACE
  182. if ((RR->node->now() - m.lastReceivedAliveAnnouncement) >= ZT_CLUSTER_TIMEOUT) {
  183. TRACE("[%u] I'm alive! peers close to %d,%d,%d can be redirected to: %s",(unsigned int)fromMemberId,m.x,m.y,m.z,addrs.c_str());
  184. }
  185. #endif
  186. m.lastReceivedAliveAnnouncement = RR->node->now();
  187. } break;
  188. case STATE_MESSAGE_HAVE_PEER: {
  189. Identity id;
  190. InetAddress physicalAddress;
  191. ptr += id.deserialize(dmsg,ptr);
  192. ptr += physicalAddress.deserialize(dmsg,ptr);
  193. if (id) {
  194. // Forget any paths that we have to this peer at its address
  195. if (physicalAddress) {
  196. SharedPtr<Peer> myPeerRecord(RR->topology->getPeer(id.address()));
  197. if (myPeerRecord)
  198. myPeerRecord->removePathByAddress(physicalAddress);
  199. }
  200. // Always save identity to update file time
  201. RR->topology->saveIdentity(id);
  202. // Set peer affinity to its new home
  203. {
  204. Mutex::Lock _l2(_peerAffinities_m);
  205. _PA &pa = _peerAffinities[id.address()];
  206. pa.ts = RR->node->now();
  207. pa.mid = fromMemberId;
  208. }
  209. TRACE("[%u] has %s @ %s",(unsigned int)fromMemberId,id.address().toString().c_str(),physicalAddress.toString().c_str());
  210. }
  211. } break;
  212. case STATE_MESSAGE_MULTICAST_LIKE: {
  213. const uint64_t nwid = dmsg.at<uint64_t>(ptr); ptr += 8;
  214. const Address address(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  215. const MAC mac(dmsg.field(ptr,6),6); ptr += 6;
  216. const uint32_t adi = dmsg.at<uint32_t>(ptr); ptr += 4;
  217. RR->mc->add(RR->node->now(),nwid,MulticastGroup(mac,adi),address);
  218. TRACE("[%u] %s likes %s/%.8x on %.16llx",(unsigned int)fromMemberId,address.toString().c_str(),mac.toString().c_str(),(unsigned int)adi,nwid);
  219. } break;
  220. case STATE_MESSAGE_COM: {
  221. CertificateOfMembership com;
  222. ptr += com.deserialize(dmsg,ptr);
  223. if (com) {
  224. TRACE("[%u] COM for %s on %.16llu rev %llu",(unsigned int)fromMemberId,com.issuedTo().toString().c_str(),com.networkId(),com.revision());
  225. }
  226. } break;
  227. case STATE_MESSAGE_PROXY_UNITE: {
  228. const Address localPeerAddress(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  229. const Address remotePeerAddress(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  230. const unsigned int numRemotePeerPaths = dmsg[ptr++];
  231. InetAddress remotePeerPaths[256]; // size is 8-bit, so 256 is max
  232. for(unsigned int i=0;i<numRemotePeerPaths;++i)
  233. ptr += remotePeerPaths[i].deserialize(dmsg,ptr);
  234. TRACE("[%u] requested proxy unite between local peer %s and remote peer %s",(unsigned int)fromMemberId,localPeerAddress.toString().c_str(),remotePeerAddress.toString().c_str());
  235. SharedPtr<Peer> localPeer(RR->topology->getPeer(localPeerAddress));
  236. if ((localPeer)&&(numRemotePeerPaths > 0)) {
  237. InetAddress bestLocalV4,bestLocalV6;
  238. localPeer->getBestActiveAddresses(RR->node->now(),bestLocalV4,bestLocalV6);
  239. InetAddress bestRemoteV4,bestRemoteV6;
  240. for(unsigned int i=0;i<numRemotePeerPaths;++i) {
  241. if ((bestRemoteV4)&&(bestRemoteV6))
  242. break;
  243. switch(remotePeerPaths[i].ss_family) {
  244. case AF_INET:
  245. if (!bestRemoteV4)
  246. bestRemoteV4 = remotePeerPaths[i];
  247. break;
  248. case AF_INET6:
  249. if (!bestRemoteV6)
  250. bestRemoteV6 = remotePeerPaths[i];
  251. break;
  252. }
  253. }
  254. Packet rendezvousForLocal(localPeerAddress,RR->identity.address(),Packet::VERB_RENDEZVOUS);
  255. rendezvousForLocal.append((uint8_t)0);
  256. remotePeerAddress.appendTo(rendezvousForLocal);
  257. Buffer<2048> rendezvousForRemote;
  258. remotePeerAddress.appendTo(rendezvousForRemote);
  259. rendezvousForRemote.append((uint8_t)Packet::VERB_RENDEZVOUS);
  260. const unsigned int rendezvousForOtherEndPayloadSizePtr = rendezvousForRemote.size();
  261. rendezvousForRemote.addSize(2); // space for actual packet payload length
  262. rendezvousForRemote.append((uint8_t)0); // flags == 0
  263. localPeerAddress.appendTo(rendezvousForRemote);
  264. bool haveMatch = false;
  265. if ((bestLocalV6)&&(bestRemoteV6)) {
  266. haveMatch = true;
  267. rendezvousForLocal.append((uint16_t)bestRemoteV6.port());
  268. rendezvousForLocal.append((uint8_t)16);
  269. rendezvousForLocal.append(bestRemoteV6.rawIpData(),16);
  270. rendezvousForRemote.append((uint16_t)bestLocalV6.port());
  271. rendezvousForRemote.append((uint8_t)16);
  272. rendezvousForRemote.append(bestLocalV6.rawIpData(),16);
  273. rendezvousForRemote.setAt<uint16_t>(rendezvousForOtherEndPayloadSizePtr,(uint16_t)(9 + 16));
  274. } else if ((bestLocalV4)&&(bestRemoteV4)) {
  275. haveMatch = true;
  276. rendezvousForLocal.append((uint16_t)bestRemoteV4.port());
  277. rendezvousForLocal.append((uint8_t)4);
  278. rendezvousForLocal.append(bestRemoteV4.rawIpData(),4);
  279. rendezvousForRemote.append((uint16_t)bestLocalV4.port());
  280. rendezvousForRemote.append((uint8_t)4);
  281. rendezvousForRemote.append(bestLocalV4.rawIpData(),4);
  282. rendezvousForRemote.setAt<uint16_t>(rendezvousForOtherEndPayloadSizePtr,(uint16_t)(9 + 4));
  283. }
  284. if (haveMatch) {
  285. _send(fromMemberId,STATE_MESSAGE_PROXY_SEND,rendezvousForRemote.data(),rendezvousForRemote.size());
  286. RR->sw->send(rendezvousForLocal,true,0);
  287. }
  288. }
  289. } break;
  290. case STATE_MESSAGE_PROXY_SEND: {
  291. const Address rcpt(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  292. const Packet::Verb verb = (Packet::Verb)dmsg[ptr++];
  293. const unsigned int len = dmsg.at<uint16_t>(ptr); ptr += 2;
  294. Packet outp(rcpt,RR->identity.address(),verb);
  295. outp.append(dmsg.field(ptr,len),len); ptr += len;
  296. RR->sw->send(outp,true,0);
  297. TRACE("[%u] proxy send %s to %s length %u",(unsigned int)fromMemberId,Packet::verbString(verb),rcpt.toString().c_str(),len);
  298. } break;
  299. }
  300. } catch ( ... ) {
  301. TRACE("invalid message of size %u type %d (inner decode), discarding",mlen,mtype);
  302. // drop invalids
  303. }
  304. ptr = nextPtr;
  305. }
  306. } catch ( ... ) {
  307. TRACE("invalid message (outer loop), discarding");
  308. // drop invalids
  309. }
  310. }
  311. }
  312. bool Cluster::sendViaCluster(const Address &fromPeerAddress,const Address &toPeerAddress,const void *data,unsigned int len,bool unite)
  313. {
  314. if (len > 16384) // sanity check
  315. return false;
  316. const uint64_t now = RR->node->now();
  317. unsigned int canHasPeer = 0;
  318. { // Anyone got this peer?
  319. Mutex::Lock _l2(_peerAffinities_m);
  320. _PA *pa = _peerAffinities.get(toPeerAddress);
  321. if ((pa)&&(pa->mid != _id)&&((now - pa->ts) < ZT_PEER_ACTIVITY_TIMEOUT))
  322. canHasPeer = pa->mid;
  323. else return false;
  324. }
  325. Buffer<2048> buf;
  326. if (unite) {
  327. InetAddress v4,v6;
  328. if (fromPeerAddress) {
  329. SharedPtr<Peer> fromPeer(RR->topology->getPeer(fromPeerAddress));
  330. if (fromPeer)
  331. fromPeer->getBestActiveAddresses(now,v4,v6);
  332. }
  333. uint8_t addrCount = 0;
  334. if (v4)
  335. ++addrCount;
  336. if (v6)
  337. ++addrCount;
  338. if (addrCount) {
  339. toPeerAddress.appendTo(buf);
  340. fromPeerAddress.appendTo(buf);
  341. buf.append(addrCount);
  342. if (v4)
  343. v4.serialize(buf);
  344. if (v6)
  345. v6.serialize(buf);
  346. }
  347. }
  348. {
  349. Mutex::Lock _l2(_members[canHasPeer].lock);
  350. if (buf.size() > 0)
  351. _send(canHasPeer,STATE_MESSAGE_PROXY_UNITE,buf.data(),buf.size());
  352. if (_members[canHasPeer].zeroTierPhysicalEndpoints.size() > 0)
  353. RR->node->putPacket(InetAddress(),_members[canHasPeer].zeroTierPhysicalEndpoints.front(),data,len);
  354. }
  355. TRACE("sendViaCluster(): relaying %u bytes from %s to %s by way of %u",len,fromPeerAddress.toString().c_str(),toPeerAddress.toString().c_str(),(unsigned int)canHasPeer);
  356. return true;
  357. }
  358. void Cluster::replicateHavePeer(const Identity &peerId,const InetAddress &physicalAddress)
  359. {
  360. const uint64_t now = RR->node->now();
  361. { // Use peer affinity table to track our own last announce time for peers
  362. Mutex::Lock _l2(_peerAffinities_m);
  363. _PA &pa = _peerAffinities[peerId.address()];
  364. if (pa.mid != _id) {
  365. pa.ts = now;
  366. pa.mid = _id;
  367. } else if ((now - pa.ts) < ZT_CLUSTER_HAVE_PEER_ANNOUNCE_PERIOD) {
  368. return;
  369. } else {
  370. pa.ts = now;
  371. }
  372. }
  373. // announcement
  374. Buffer<4096> buf;
  375. peerId.serialize(buf,false);
  376. physicalAddress.serialize(buf);
  377. {
  378. Mutex::Lock _l(_memberIds_m);
  379. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  380. Mutex::Lock _l2(_members[*mid].lock);
  381. _send(*mid,STATE_MESSAGE_HAVE_PEER,buf.data(),buf.size());
  382. }
  383. }
  384. }
  385. void Cluster::replicateMulticastLike(uint64_t nwid,const Address &peerAddress,const MulticastGroup &group)
  386. {
  387. Buffer<2048> buf;
  388. buf.append((uint64_t)nwid);
  389. peerAddress.appendTo(buf);
  390. group.mac().appendTo(buf);
  391. buf.append((uint32_t)group.adi());
  392. TRACE("replicating %s MULTICAST_LIKE %.16llx/%s/%u to all members",peerAddress.toString().c_str(),nwid,group.mac().toString().c_str(),(unsigned int)group.adi());
  393. {
  394. Mutex::Lock _l(_memberIds_m);
  395. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  396. Mutex::Lock _l2(_members[*mid].lock);
  397. _send(*mid,STATE_MESSAGE_MULTICAST_LIKE,buf.data(),buf.size());
  398. }
  399. }
  400. }
  401. void Cluster::replicateCertificateOfNetworkMembership(const CertificateOfMembership &com)
  402. {
  403. Buffer<2048> buf;
  404. com.serialize(buf);
  405. TRACE("replicating %s COM for %.16llx to all members",com.issuedTo().toString().c_str(),com.networkId());
  406. {
  407. Mutex::Lock _l(_memberIds_m);
  408. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  409. Mutex::Lock _l2(_members[*mid].lock);
  410. _send(*mid,STATE_MESSAGE_COM,buf.data(),buf.size());
  411. }
  412. }
  413. }
  414. void Cluster::doPeriodicTasks()
  415. {
  416. const uint64_t now = RR->node->now();
  417. {
  418. Mutex::Lock _l(_memberIds_m);
  419. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  420. Mutex::Lock _l2(_members[*mid].lock);
  421. if ((now - _members[*mid].lastAnnouncedAliveTo) >= ((ZT_CLUSTER_TIMEOUT / 2) - 1000)) {
  422. Buffer<2048> alive;
  423. alive.append((uint16_t)ZEROTIER_ONE_VERSION_MAJOR);
  424. alive.append((uint16_t)ZEROTIER_ONE_VERSION_MINOR);
  425. alive.append((uint16_t)ZEROTIER_ONE_VERSION_REVISION);
  426. alive.append((uint8_t)ZT_PROTO_VERSION);
  427. if (_addressToLocationFunction) {
  428. alive.append((int32_t)_x);
  429. alive.append((int32_t)_y);
  430. alive.append((int32_t)_z);
  431. } else {
  432. alive.append((int32_t)0);
  433. alive.append((int32_t)0);
  434. alive.append((int32_t)0);
  435. }
  436. alive.append((uint64_t)now);
  437. alive.append((uint64_t)0); // TODO: compute and send load average
  438. alive.append((uint64_t)0); // unused/reserved flags
  439. alive.append((uint8_t)_zeroTierPhysicalEndpoints.size());
  440. for(std::vector<InetAddress>::const_iterator pe(_zeroTierPhysicalEndpoints.begin());pe!=_zeroTierPhysicalEndpoints.end();++pe)
  441. pe->serialize(alive);
  442. _send(*mid,STATE_MESSAGE_ALIVE,alive.data(),alive.size());
  443. _members[*mid].lastAnnouncedAliveTo = now;
  444. }
  445. _flush(*mid); // does nothing if nothing to flush
  446. }
  447. }
  448. }
  449. void Cluster::addMember(uint16_t memberId)
  450. {
  451. if ((memberId >= ZT_CLUSTER_MAX_MEMBERS)||(memberId == _id))
  452. return;
  453. Mutex::Lock _l2(_members[memberId].lock);
  454. {
  455. Mutex::Lock _l(_memberIds_m);
  456. if (std::find(_memberIds.begin(),_memberIds.end(),memberId) != _memberIds.end())
  457. return;
  458. _memberIds.push_back(memberId);
  459. std::sort(_memberIds.begin(),_memberIds.end());
  460. }
  461. _members[memberId].clear();
  462. // Generate this member's message key from the master and its ID
  463. uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];
  464. memcpy(stmp,_masterSecret,sizeof(stmp));
  465. stmp[0] ^= Utils::hton(memberId);
  466. SHA512::hash(stmp,stmp,sizeof(stmp));
  467. SHA512::hash(stmp,stmp,sizeof(stmp));
  468. memcpy(_members[memberId].key,stmp,sizeof(_members[memberId].key));
  469. Utils::burn(stmp,sizeof(stmp));
  470. // Prepare q
  471. _members[memberId].q.clear();
  472. char iv[16];
  473. Utils::getSecureRandom(iv,16);
  474. _members[memberId].q.append(iv,16);
  475. _members[memberId].q.addSize(8); // room for MAC
  476. _members[memberId].q.append((uint16_t)_id);
  477. _members[memberId].q.append((uint16_t)memberId);
  478. }
  479. void Cluster::removeMember(uint16_t memberId)
  480. {
  481. Mutex::Lock _l(_memberIds_m);
  482. std::vector<uint16_t> newMemberIds;
  483. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  484. if (*mid != memberId)
  485. newMemberIds.push_back(*mid);
  486. }
  487. _memberIds = newMemberIds;
  488. }
  489. bool Cluster::findBetterEndpoint(InetAddress &redirectTo,const Address &peerAddress,const InetAddress &peerPhysicalAddress,bool offload)
  490. {
  491. if (_addressToLocationFunction) {
  492. // Pick based on location if it can be determined
  493. int px = 0,py = 0,pz = 0;
  494. if (_addressToLocationFunction(_addressToLocationFunctionArg,reinterpret_cast<const struct sockaddr_storage *>(&peerPhysicalAddress),&px,&py,&pz) == 0) {
  495. TRACE("no geolocation data for %s (geo-lookup is lazy/async so it may work next time)",peerPhysicalAddress.toIpString().c_str());
  496. return false;
  497. }
  498. // Find member closest to this peer
  499. const uint64_t now = RR->node->now();
  500. std::vector<InetAddress> best; // initial "best" is for peer to stay put
  501. const double currentDistance = _dist3d(_x,_y,_z,px,py,pz);
  502. double bestDistance = (offload ? 2147483648.0 : currentDistance);
  503. unsigned int bestMember = _id;
  504. {
  505. Mutex::Lock _l(_memberIds_m);
  506. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  507. _Member &m = _members[*mid];
  508. Mutex::Lock _ml(m.lock);
  509. // Consider member if it's alive and has sent us a location and one or more physical endpoints to send peers to
  510. if ( ((now - m.lastReceivedAliveAnnouncement) < ZT_CLUSTER_TIMEOUT) && ((m.x != 0)||(m.y != 0)||(m.z != 0)) && (m.zeroTierPhysicalEndpoints.size() > 0) ) {
  511. double mdist = _dist3d(m.x,m.y,m.z,px,py,pz);
  512. if (mdist < bestDistance) {
  513. bestDistance = mdist;
  514. bestMember = *mid;
  515. best = m.zeroTierPhysicalEndpoints;
  516. }
  517. }
  518. }
  519. }
  520. // Suggestion redirection if a closer member was found
  521. for(std::vector<InetAddress>::const_iterator a(best.begin());a!=best.end();++a) {
  522. if (a->ss_family == peerPhysicalAddress.ss_family) {
  523. TRACE("%s at [%d,%d,%d] is %f from us but %f from %u, can redirect to %s",peerAddress.toString().c_str(),px,py,pz,currentDistance,bestDistance,bestMember,a->toString().c_str());
  524. redirectTo = *a;
  525. return true;
  526. }
  527. }
  528. TRACE("%s at [%d,%d,%d] is %f from us, no better endpoints found",peerAddress.toString().c_str(),px,py,pz,currentDistance);
  529. return false;
  530. } else {
  531. // TODO: pick based on load if no location info?
  532. return false;
  533. }
  534. }
  535. void Cluster::status(ZT_ClusterStatus &status) const
  536. {
  537. const uint64_t now = RR->node->now();
  538. memset(&status,0,sizeof(ZT_ClusterStatus));
  539. ZT_ClusterMemberStatus *ms[ZT_CLUSTER_MAX_MEMBERS];
  540. memset(ms,0,sizeof(ms));
  541. status.myId = _id;
  542. ms[_id] = &(status.members[status.clusterSize++]);
  543. ms[_id]->id = _id;
  544. ms[_id]->alive = 1;
  545. ms[_id]->x = _x;
  546. ms[_id]->y = _y;
  547. ms[_id]->z = _z;
  548. ms[_id]->peers = RR->topology->countAlive();
  549. for(std::vector<InetAddress>::const_iterator ep(_zeroTierPhysicalEndpoints.begin());ep!=_zeroTierPhysicalEndpoints.end();++ep) {
  550. if (ms[_id]->numZeroTierPhysicalEndpoints >= ZT_CLUSTER_MAX_ZT_PHYSICAL_ADDRESSES) // sanity check
  551. break;
  552. memcpy(&(ms[_id]->zeroTierPhysicalEndpoints[ms[_id]->numZeroTierPhysicalEndpoints++]),&(*ep),sizeof(struct sockaddr_storage));
  553. }
  554. {
  555. Mutex::Lock _l1(_memberIds_m);
  556. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  557. if (status.clusterSize >= ZT_CLUSTER_MAX_MEMBERS) // sanity check
  558. break;
  559. ZT_ClusterMemberStatus *s = ms[*mid] = &(status.members[status.clusterSize++]);
  560. _Member &m = _members[*mid];
  561. Mutex::Lock ml(m.lock);
  562. s->id = *mid;
  563. s->msSinceLastHeartbeat = (unsigned int)std::min((uint64_t)(~((unsigned int)0)),(now - m.lastReceivedAliveAnnouncement));
  564. s->alive = (s->msSinceLastHeartbeat < ZT_CLUSTER_TIMEOUT) ? 1 : 0;
  565. s->x = m.x;
  566. s->y = m.y;
  567. s->z = m.z;
  568. s->load = m.load;
  569. for(std::vector<InetAddress>::const_iterator ep(m.zeroTierPhysicalEndpoints.begin());ep!=m.zeroTierPhysicalEndpoints.end();++ep) {
  570. if (s->numZeroTierPhysicalEndpoints >= ZT_CLUSTER_MAX_ZT_PHYSICAL_ADDRESSES) // sanity check
  571. break;
  572. memcpy(&(s->zeroTierPhysicalEndpoints[s->numZeroTierPhysicalEndpoints++]),&(*ep),sizeof(struct sockaddr_storage));
  573. }
  574. }
  575. }
  576. {
  577. Mutex::Lock _l2(_peerAffinities_m);
  578. Address *k = (Address *)0;
  579. _PA *v = (_PA *)0;
  580. Hashtable< Address,_PA >::Iterator i(const_cast<Cluster *>(this)->_peerAffinities);
  581. while (i.next(k,v)) {
  582. if ( (ms[v->mid]) && (v->mid != _id) && ((now - v->ts) < ZT_PEER_ACTIVITY_TIMEOUT) )
  583. ++ms[v->mid]->peers;
  584. }
  585. }
  586. }
  587. void Cluster::_send(uint16_t memberId,StateMessageType type,const void *msg,unsigned int len)
  588. {
  589. if ((len + 3) > (ZT_CLUSTER_MAX_MESSAGE_LENGTH - (24 + 2 + 2))) // sanity check
  590. return;
  591. _Member &m = _members[memberId];
  592. // assumes m.lock is locked!
  593. if ((m.q.size() + len + 3) > ZT_CLUSTER_MAX_MESSAGE_LENGTH)
  594. _flush(memberId);
  595. m.q.append((uint16_t)(len + 1));
  596. m.q.append((uint8_t)type);
  597. m.q.append(msg,len);
  598. }
  599. void Cluster::_flush(uint16_t memberId)
  600. {
  601. _Member &m = _members[memberId];
  602. // assumes m.lock is locked!
  603. if (m.q.size() > (24 + 2 + 2)) { // 16-byte IV + 8-byte MAC + 2 byte from-member-ID + 2 byte to-member-ID
  604. // Create key from member's key and IV
  605. char keytmp[32];
  606. memcpy(keytmp,m.key,32);
  607. for(int i=0;i<8;++i)
  608. keytmp[i] ^= m.q[i];
  609. Salsa20 s20(keytmp,256,m.q.field(8,8));
  610. Utils::burn(keytmp,sizeof(keytmp));
  611. // One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
  612. char polykey[ZT_POLY1305_KEY_LEN];
  613. memset(polykey,0,sizeof(polykey));
  614. s20.encrypt12(polykey,polykey,sizeof(polykey));
  615. // Encrypt m.q in place
  616. 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);
  617. // Add MAC for authentication (encrypt-then-MAC)
  618. char mac[ZT_POLY1305_MAC_LEN];
  619. Poly1305::compute(mac,reinterpret_cast<const char *>(m.q.data()) + 24,m.q.size() - 24,polykey);
  620. memcpy(m.q.field(16,8),mac,8);
  621. // Send!
  622. _sendFunction(_sendFunctionArg,memberId,m.q.data(),m.q.size());
  623. // Prepare for more
  624. m.q.clear();
  625. char iv[16];
  626. Utils::getSecureRandom(iv,16);
  627. m.q.append(iv,16);
  628. m.q.addSize(8); // room for MAC
  629. m.q.append((uint16_t)_id); // from member ID
  630. m.q.append((uint16_t)memberId); // to member ID
  631. }
  632. }
  633. } // namespace ZeroTier
  634. #endif // ZT_ENABLE_CLUSTER