Cluster.cpp 28 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794
  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. _lastFlushed(0),
  79. _lastCleanedRemotePeers(0)
  80. {
  81. uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];
  82. // Generate master secret by hashing the secret from our Identity key pair
  83. RR->identity.sha512PrivateKey(_masterSecret);
  84. // Generate our inbound message key, which is the master secret XORed with our ID and hashed twice
  85. memcpy(stmp,_masterSecret,sizeof(stmp));
  86. stmp[0] ^= Utils::hton(id);
  87. SHA512::hash(stmp,stmp,sizeof(stmp));
  88. SHA512::hash(stmp,stmp,sizeof(stmp));
  89. memcpy(_key,stmp,sizeof(_key));
  90. Utils::burn(stmp,sizeof(stmp));
  91. }
  92. Cluster::~Cluster()
  93. {
  94. Utils::burn(_masterSecret,sizeof(_masterSecret));
  95. Utils::burn(_key,sizeof(_key));
  96. delete [] _members;
  97. for(std::multimap<Address,_SQE *>::iterator qi(_sendViaClusterQueue.begin());qi!=_sendViaClusterQueue.end();)
  98. delete qi->second;
  99. }
  100. void Cluster::handleIncomingStateMessage(const void *msg,unsigned int len)
  101. {
  102. Buffer<ZT_CLUSTER_MAX_MESSAGE_LENGTH> dmsg;
  103. {
  104. // FORMAT: <[16] iv><[8] MAC><... data>
  105. if ((len < 24)||(len > ZT_CLUSTER_MAX_MESSAGE_LENGTH))
  106. return;
  107. // 16-byte IV: first 8 bytes XORed with key, last 8 bytes used as Salsa20 64-bit IV
  108. char keytmp[32];
  109. memcpy(keytmp,_key,32);
  110. for(int i=0;i<8;++i)
  111. keytmp[i] ^= reinterpret_cast<const char *>(msg)[i];
  112. Salsa20 s20(keytmp,256,reinterpret_cast<const char *>(msg) + 8);
  113. Utils::burn(keytmp,sizeof(keytmp));
  114. // One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
  115. char polykey[ZT_POLY1305_KEY_LEN];
  116. memset(polykey,0,sizeof(polykey));
  117. s20.encrypt12(polykey,polykey,sizeof(polykey));
  118. // Compute 16-byte MAC
  119. char mac[ZT_POLY1305_MAC_LEN];
  120. Poly1305::compute(mac,reinterpret_cast<const char *>(msg) + 24,len - 24,polykey);
  121. // Check first 8 bytes of MAC against 64-bit MAC in stream
  122. if (!Utils::secureEq(mac,reinterpret_cast<const char *>(msg) + 16,8))
  123. return;
  124. // Decrypt!
  125. dmsg.setSize(len - 24);
  126. s20.decrypt12(reinterpret_cast<const char *>(msg) + 24,const_cast<void *>(dmsg.data()),dmsg.size());
  127. }
  128. if (dmsg.size() < 4)
  129. return;
  130. const uint16_t fromMemberId = dmsg.at<uint16_t>(0);
  131. unsigned int ptr = 2;
  132. if (fromMemberId == _id) // sanity check: we don't talk to ourselves
  133. return;
  134. const uint16_t toMemberId = dmsg.at<uint16_t>(ptr);
  135. ptr += 2;
  136. if (toMemberId != _id) // sanity check: message not for us?
  137. return;
  138. { // make sure sender is actually considered a member
  139. Mutex::Lock _l3(_memberIds_m);
  140. if (std::find(_memberIds.begin(),_memberIds.end(),fromMemberId) == _memberIds.end())
  141. return;
  142. }
  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 CLUSTER_MESSAGE_ALIVE: {
  155. _Member &m = _members[fromMemberId];
  156. Mutex::Lock mlck(m.lock);
  157. ptr += 7; // skip version stuff, not used yet
  158. m.x = dmsg.at<int32_t>(ptr); ptr += 4;
  159. m.y = dmsg.at<int32_t>(ptr); ptr += 4;
  160. m.z = dmsg.at<int32_t>(ptr); ptr += 4;
  161. ptr += 8; // skip local clock, not used
  162. m.load = dmsg.at<uint64_t>(ptr); ptr += 8;
  163. m.peers = dmsg.at<uint64_t>(ptr); ptr += 8;
  164. ptr += 8; // skip flags, unused
  165. #ifdef ZT_TRACE
  166. std::string addrs;
  167. #endif
  168. unsigned int physicalAddressCount = dmsg[ptr++];
  169. m.zeroTierPhysicalEndpoints.clear();
  170. for(unsigned int i=0;i<physicalAddressCount;++i) {
  171. m.zeroTierPhysicalEndpoints.push_back(InetAddress());
  172. ptr += m.zeroTierPhysicalEndpoints.back().deserialize(dmsg,ptr);
  173. if (!(m.zeroTierPhysicalEndpoints.back())) {
  174. m.zeroTierPhysicalEndpoints.pop_back();
  175. }
  176. #ifdef ZT_TRACE
  177. else {
  178. if (addrs.length() > 0)
  179. addrs.push_back(',');
  180. addrs.append(m.zeroTierPhysicalEndpoints.back().toString());
  181. }
  182. #endif
  183. }
  184. #ifdef ZT_TRACE
  185. if ((RR->node->now() - m.lastReceivedAliveAnnouncement) >= ZT_CLUSTER_TIMEOUT) {
  186. 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());
  187. }
  188. #endif
  189. m.lastReceivedAliveAnnouncement = RR->node->now();
  190. } break;
  191. case CLUSTER_MESSAGE_HAVE_PEER: {
  192. Identity id;
  193. ptr += id.deserialize(dmsg,ptr);
  194. if (id) {
  195. RR->topology->saveIdentity(id);
  196. {
  197. Mutex::Lock _l(_remotePeers_m);
  198. _remotePeers[std::pair<Address,unsigned int>(id.address(),(unsigned int)fromMemberId)] = RR->node->now();
  199. }
  200. std::pair<Address,_SQE *> q[ZT_CLUSTER_MAX_QUEUE_PER_SENDER];
  201. unsigned int qc = 0;
  202. {
  203. Mutex::Lock _l(_sendViaClusterQueue_m);
  204. std::pair< std::multimap<Address,_SQE *>::iterator,std::multimap<Address,_SQE *>::iterator > er(_sendViaClusterQueue.equal_range(id.address()));
  205. for(std::multimap<Address,_SQE *>::iterator qi(er.first);qi!=er.second;) {
  206. if (qc >= ZT_CLUSTER_MAX_QUEUE_PER_SENDER) // sanity check
  207. break;
  208. q[qc++] = *qi;
  209. _sendViaClusterQueue.erase(qi++);
  210. }
  211. }
  212. for(unsigned int i=0;i<qc;++i) {
  213. this->sendViaCluster(q[i].first,q[i].second->toPeerAddress,q[i].second->data,q[i].second->len,q[i].second->unite);
  214. delete q[i].second;
  215. }
  216. TRACE("[%u] has %s",(unsigned int)fromMemberId,id.address().toString().c_str());
  217. }
  218. } break;
  219. case CLUSTER_MESSAGE_WANT_PEER: {
  220. const Address zeroTierAddress(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  221. SharedPtr<Peer> peer(RR->topology->getPeerNoCache(zeroTierAddress));
  222. if ( (peer) && (peer->hasActiveDirectPath(RR->node->now())) ) {
  223. Buffer<1024> buf;
  224. peer->identity().serialize(buf);
  225. Mutex::Lock _l2(_members[fromMemberId].lock);
  226. _send(fromMemberId,CLUSTER_MESSAGE_HAVE_PEER,buf.data(),buf.size());
  227. _flush(fromMemberId);
  228. }
  229. } break;
  230. case CLUSTER_MESSAGE_REMOTE_PACKET: {
  231. const unsigned int plen = dmsg.at<uint16_t>(ptr); ptr += 2;
  232. if (plen) {
  233. Packet remotep(dmsg.field(ptr,plen),plen); ptr += plen;
  234. TRACE("remote %s from %s via %u (%u bytes)",Packet::verbString(remotep.verb()),remotep.source().toString().c_str(),fromMemberId,plen);
  235. switch(remotep.verb()) {
  236. case Packet::VERB_WHOIS: _doREMOTE_WHOIS(fromMemberId,remotep); break;
  237. case Packet::VERB_MULTICAST_GATHER: _doREMOTE_MULTICAST_GATHER(fromMemberId,remotep); break;
  238. default: break; // ignore things we don't care about across cluster
  239. }
  240. }
  241. } break;
  242. case CLUSTER_MESSAGE_PROXY_UNITE: {
  243. const Address localPeerAddress(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  244. const Address remotePeerAddress(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  245. const unsigned int numRemotePeerPaths = dmsg[ptr++];
  246. InetAddress remotePeerPaths[256]; // size is 8-bit, so 256 is max
  247. for(unsigned int i=0;i<numRemotePeerPaths;++i)
  248. ptr += remotePeerPaths[i].deserialize(dmsg,ptr);
  249. TRACE("[%u] requested that we unite local %s with remote %s",(unsigned int)fromMemberId,localPeerAddress.toString().c_str(),remotePeerAddress.toString().c_str());
  250. const uint64_t now = RR->node->now();
  251. SharedPtr<Peer> localPeer(RR->topology->getPeerNoCache(localPeerAddress));
  252. if ((localPeer)&&(numRemotePeerPaths > 0)) {
  253. InetAddress bestLocalV4,bestLocalV6;
  254. localPeer->getBestActiveAddresses(now,bestLocalV4,bestLocalV6);
  255. InetAddress bestRemoteV4,bestRemoteV6;
  256. for(unsigned int i=0;i<numRemotePeerPaths;++i) {
  257. if ((bestRemoteV4)&&(bestRemoteV6))
  258. break;
  259. switch(remotePeerPaths[i].ss_family) {
  260. case AF_INET:
  261. if (!bestRemoteV4)
  262. bestRemoteV4 = remotePeerPaths[i];
  263. break;
  264. case AF_INET6:
  265. if (!bestRemoteV6)
  266. bestRemoteV6 = remotePeerPaths[i];
  267. break;
  268. }
  269. }
  270. Packet rendezvousForLocal(localPeerAddress,RR->identity.address(),Packet::VERB_RENDEZVOUS);
  271. rendezvousForLocal.append((uint8_t)0);
  272. remotePeerAddress.appendTo(rendezvousForLocal);
  273. Buffer<2048> rendezvousForRemote;
  274. remotePeerAddress.appendTo(rendezvousForRemote);
  275. rendezvousForRemote.append((uint8_t)Packet::VERB_RENDEZVOUS);
  276. rendezvousForRemote.addSize(2); // space for actual packet payload length
  277. rendezvousForRemote.append((uint8_t)0); // flags == 0
  278. localPeerAddress.appendTo(rendezvousForRemote);
  279. bool haveMatch = false;
  280. if ((bestLocalV6)&&(bestRemoteV6)) {
  281. haveMatch = true;
  282. rendezvousForLocal.append((uint16_t)bestRemoteV6.port());
  283. rendezvousForLocal.append((uint8_t)16);
  284. rendezvousForLocal.append(bestRemoteV6.rawIpData(),16);
  285. rendezvousForRemote.append((uint16_t)bestLocalV6.port());
  286. rendezvousForRemote.append((uint8_t)16);
  287. rendezvousForRemote.append(bestLocalV6.rawIpData(),16);
  288. rendezvousForRemote.setAt<uint16_t>(ZT_ADDRESS_LENGTH + 1,(uint16_t)(9 + 16));
  289. } else if ((bestLocalV4)&&(bestRemoteV4)) {
  290. haveMatch = true;
  291. rendezvousForLocal.append((uint16_t)bestRemoteV4.port());
  292. rendezvousForLocal.append((uint8_t)4);
  293. rendezvousForLocal.append(bestRemoteV4.rawIpData(),4);
  294. rendezvousForRemote.append((uint16_t)bestLocalV4.port());
  295. rendezvousForRemote.append((uint8_t)4);
  296. rendezvousForRemote.append(bestLocalV4.rawIpData(),4);
  297. rendezvousForRemote.setAt<uint16_t>(ZT_ADDRESS_LENGTH + 1,(uint16_t)(9 + 4));
  298. }
  299. if (haveMatch) {
  300. {
  301. Mutex::Lock _l2(_members[fromMemberId].lock);
  302. _send(fromMemberId,CLUSTER_MESSAGE_PROXY_SEND,rendezvousForRemote.data(),rendezvousForRemote.size());
  303. _flush(fromMemberId);
  304. }
  305. RR->sw->send(rendezvousForLocal,true,0);
  306. }
  307. }
  308. } break;
  309. case CLUSTER_MESSAGE_PROXY_SEND: {
  310. const Address rcpt(dmsg.field(ptr,ZT_ADDRESS_LENGTH),ZT_ADDRESS_LENGTH); ptr += ZT_ADDRESS_LENGTH;
  311. const Packet::Verb verb = (Packet::Verb)dmsg[ptr++];
  312. const unsigned int len = dmsg.at<uint16_t>(ptr); ptr += 2;
  313. Packet outp(rcpt,RR->identity.address(),verb);
  314. outp.append(dmsg.field(ptr,len),len); ptr += len;
  315. RR->sw->send(outp,true,0);
  316. //TRACE("[%u] proxy send %s to %s length %u",(unsigned int)fromMemberId,Packet::verbString(verb),rcpt.toString().c_str(),len);
  317. } break;
  318. }
  319. } catch ( ... ) {
  320. TRACE("invalid message of size %u type %d (inner decode), discarding",mlen,mtype);
  321. // drop invalids
  322. }
  323. ptr = nextPtr;
  324. }
  325. } catch ( ... ) {
  326. TRACE("invalid message (outer loop), discarding");
  327. // drop invalids
  328. }
  329. }
  330. void Cluster::sendViaCluster(const Address &fromPeerAddress,const Address &toPeerAddress,const void *data,unsigned int len,bool unite)
  331. {
  332. if (len > ZT_PROTO_MAX_PACKET_LENGTH) // sanity check
  333. return;
  334. _sendViaClusterQueue_m.lock();
  335. const unsigned long queueCount = _sendViaClusterQueue.count(fromPeerAddress);
  336. _sendViaClusterQueue_m.unlock();
  337. if (queueCount > ZT_CLUSTER_MAX_QUEUE_PER_SENDER)
  338. return;
  339. const uint64_t now = RR->node->now();
  340. uint64_t mostRecentTs = 0;
  341. unsigned int mostRecentMemberId = 0xffffffff;
  342. {
  343. Mutex::Lock _l2(_remotePeers_m);
  344. std::map< std::pair<Address,unsigned int>,uint64_t >::const_iterator rpe(_remotePeers.lower_bound(std::pair<Address,unsigned int>(fromPeerAddress,0)));
  345. for(;;) {
  346. if ((rpe == _remotePeers.end())||(rpe->first.first != fromPeerAddress))
  347. break;
  348. else if (rpe->second > mostRecentTs) {
  349. mostRecentTs = rpe->second;
  350. mostRecentMemberId = rpe->first.second;
  351. }
  352. }
  353. }
  354. const uint64_t age = now - mostRecentTs;
  355. if (age >= (ZT_PEER_ACTIVITY_TIMEOUT / 3)) {
  356. const bool enqueueAndWait = ((age >= ZT_PEER_ACTIVITY_TIMEOUT)||(mostRecentMemberId > 0xffff));
  357. // Poll everyone with WANT_PEER if the age of our most recent entry is
  358. // approaching expiration (or has expired, or does not exist).
  359. char tmp[ZT_ADDRESS_LENGTH];
  360. toPeerAddress.copyTo(tmp,ZT_ADDRESS_LENGTH);
  361. {
  362. Mutex::Lock _l(_memberIds_m);
  363. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  364. Mutex::Lock _l2(_members[*mid].lock);
  365. _send(*mid,CLUSTER_MESSAGE_WANT_PEER,tmp,ZT_ADDRESS_LENGTH);
  366. if ((enqueueAndWait)&&(queueCount == 0))
  367. _flush(*mid);
  368. }
  369. }
  370. // If there isn't a good place to send via, then enqueue this for retrying
  371. // later and return after having broadcasted a WANT_PEER.
  372. if (enqueueAndWait) {
  373. Mutex::Lock _l(_sendViaClusterQueue_m);
  374. _sendViaClusterQueue.insert(std::pair<Address,_SQE *>(fromPeerAddress,new _SQE(now,toPeerAddress,data,len,unite)));
  375. return;
  376. }
  377. }
  378. Buffer<1024> buf;
  379. if (unite) {
  380. InetAddress v4,v6;
  381. if (fromPeerAddress) {
  382. SharedPtr<Peer> fromPeer(RR->topology->getPeerNoCache(fromPeerAddress));
  383. if (fromPeer)
  384. fromPeer->getBestActiveAddresses(now,v4,v6);
  385. }
  386. uint8_t addrCount = 0;
  387. if (v4)
  388. ++addrCount;
  389. if (v6)
  390. ++addrCount;
  391. if (addrCount) {
  392. toPeerAddress.appendTo(buf);
  393. fromPeerAddress.appendTo(buf);
  394. buf.append(addrCount);
  395. if (v4)
  396. v4.serialize(buf);
  397. if (v6)
  398. v6.serialize(buf);
  399. }
  400. }
  401. {
  402. Mutex::Lock _l2(_members[mostRecentMemberId].lock);
  403. if (buf.size() > 0) {
  404. _send(mostRecentMemberId,CLUSTER_MESSAGE_PROXY_UNITE,buf.data(),buf.size());
  405. _flush(mostRecentMemberId);
  406. }
  407. if (_members[mostRecentMemberId].zeroTierPhysicalEndpoints.size() > 0)
  408. RR->node->putPacket(InetAddress(),_members[mostRecentMemberId].zeroTierPhysicalEndpoints.front(),data,len);
  409. }
  410. TRACE("sendViaCluster(): relaying %u bytes from %s to %s by way of %u",len,fromPeerAddress.toString().c_str(),toPeerAddress.toString().c_str(),(unsigned int)mostRecentMemberId);
  411. }
  412. void Cluster::sendDistributedQuery(const Packet &pkt)
  413. {
  414. Buffer<4096> buf;
  415. buf.append((uint16_t)pkt.size());
  416. buf.append(pkt.data(),pkt.size());
  417. Mutex::Lock _l(_memberIds_m);
  418. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  419. Mutex::Lock _l2(_members[*mid].lock);
  420. _send(*mid,CLUSTER_MESSAGE_REMOTE_PACKET,buf.data(),buf.size());
  421. _flush(*mid);
  422. }
  423. }
  424. void Cluster::doPeriodicTasks()
  425. {
  426. const uint64_t now = RR->node->now();
  427. if ((now - _lastFlushed) >= ZT_CLUSTER_FLUSH_PERIOD) {
  428. _lastFlushed = now;
  429. {
  430. Mutex::Lock _l2(_sendViaClusterQueue_m);
  431. for(std::multimap<Address,_SQE *>::iterator qi(_sendViaClusterQueue.begin());qi!=_sendViaClusterQueue.end();) {
  432. if ((now - qi->second->timestamp) >= ZT_CLUSTER_QUEUE_EXPIRATION) {
  433. delete qi->second;
  434. _sendViaClusterQueue.erase(qi++);
  435. } else ++qi;
  436. }
  437. }
  438. Mutex::Lock _l(_memberIds_m);
  439. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  440. Mutex::Lock _l2(_members[*mid].lock);
  441. if ((now - _members[*mid].lastAnnouncedAliveTo) >= ((ZT_CLUSTER_TIMEOUT / 2) - 1000)) {
  442. _members[*mid].lastAnnouncedAliveTo = now;
  443. Buffer<2048> alive;
  444. alive.append((uint16_t)ZEROTIER_ONE_VERSION_MAJOR);
  445. alive.append((uint16_t)ZEROTIER_ONE_VERSION_MINOR);
  446. alive.append((uint16_t)ZEROTIER_ONE_VERSION_REVISION);
  447. alive.append((uint8_t)ZT_PROTO_VERSION);
  448. if (_addressToLocationFunction) {
  449. alive.append((int32_t)_x);
  450. alive.append((int32_t)_y);
  451. alive.append((int32_t)_z);
  452. } else {
  453. alive.append((int32_t)0);
  454. alive.append((int32_t)0);
  455. alive.append((int32_t)0);
  456. }
  457. alive.append((uint64_t)now);
  458. alive.append((uint64_t)0); // TODO: compute and send load average
  459. alive.append((uint64_t)RR->topology->countActive());
  460. alive.append((uint64_t)0); // unused/reserved flags
  461. alive.append((uint8_t)_zeroTierPhysicalEndpoints.size());
  462. for(std::vector<InetAddress>::const_iterator pe(_zeroTierPhysicalEndpoints.begin());pe!=_zeroTierPhysicalEndpoints.end();++pe)
  463. pe->serialize(alive);
  464. _send(*mid,CLUSTER_MESSAGE_ALIVE,alive.data(),alive.size());
  465. }
  466. _flush(*mid);
  467. }
  468. }
  469. if ((now - _lastCleanedRemotePeers) >= (ZT_PEER_ACTIVITY_TIMEOUT * 2)) {
  470. _lastCleanedRemotePeers = now;
  471. Mutex::Lock _l(_remotePeers_m);
  472. for(std::map< std::pair<Address,unsigned int>,uint64_t >::iterator rp(_remotePeers.begin());rp!=_remotePeers.end();) {
  473. if ((now - rp->second) >= ZT_PEER_ACTIVITY_TIMEOUT)
  474. _remotePeers.erase(rp++);
  475. else ++rp;
  476. }
  477. }
  478. }
  479. void Cluster::addMember(uint16_t memberId)
  480. {
  481. if ((memberId >= ZT_CLUSTER_MAX_MEMBERS)||(memberId == _id))
  482. return;
  483. Mutex::Lock _l2(_members[memberId].lock);
  484. {
  485. Mutex::Lock _l(_memberIds_m);
  486. if (std::find(_memberIds.begin(),_memberIds.end(),memberId) != _memberIds.end())
  487. return;
  488. _memberIds.push_back(memberId);
  489. std::sort(_memberIds.begin(),_memberIds.end());
  490. }
  491. _members[memberId].clear();
  492. // Generate this member's message key from the master and its ID
  493. uint16_t stmp[ZT_SHA512_DIGEST_LEN / sizeof(uint16_t)];
  494. memcpy(stmp,_masterSecret,sizeof(stmp));
  495. stmp[0] ^= Utils::hton(memberId);
  496. SHA512::hash(stmp,stmp,sizeof(stmp));
  497. SHA512::hash(stmp,stmp,sizeof(stmp));
  498. memcpy(_members[memberId].key,stmp,sizeof(_members[memberId].key));
  499. Utils::burn(stmp,sizeof(stmp));
  500. // Prepare q
  501. _members[memberId].q.clear();
  502. char iv[16];
  503. Utils::getSecureRandom(iv,16);
  504. _members[memberId].q.append(iv,16);
  505. _members[memberId].q.addSize(8); // room for MAC
  506. _members[memberId].q.append((uint16_t)_id);
  507. _members[memberId].q.append((uint16_t)memberId);
  508. }
  509. void Cluster::removeMember(uint16_t memberId)
  510. {
  511. Mutex::Lock _l(_memberIds_m);
  512. std::vector<uint16_t> newMemberIds;
  513. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  514. if (*mid != memberId)
  515. newMemberIds.push_back(*mid);
  516. }
  517. _memberIds = newMemberIds;
  518. }
  519. bool Cluster::findBetterEndpoint(InetAddress &redirectTo,const Address &peerAddress,const InetAddress &peerPhysicalAddress,bool offload)
  520. {
  521. if (_addressToLocationFunction) {
  522. // Pick based on location if it can be determined
  523. int px = 0,py = 0,pz = 0;
  524. if (_addressToLocationFunction(_addressToLocationFunctionArg,reinterpret_cast<const struct sockaddr_storage *>(&peerPhysicalAddress),&px,&py,&pz) == 0) {
  525. TRACE("no geolocation data for %s (geo-lookup is lazy/async so it may work next time)",peerPhysicalAddress.toIpString().c_str());
  526. return false;
  527. }
  528. // Find member closest to this peer
  529. const uint64_t now = RR->node->now();
  530. std::vector<InetAddress> best;
  531. const double currentDistance = _dist3d(_x,_y,_z,px,py,pz);
  532. double bestDistance = (offload ? 2147483648.0 : currentDistance);
  533. unsigned int bestMember = _id;
  534. {
  535. Mutex::Lock _l(_memberIds_m);
  536. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  537. _Member &m = _members[*mid];
  538. Mutex::Lock _ml(m.lock);
  539. // Consider member if it's alive and has sent us a location and one or more physical endpoints to send peers to
  540. if ( ((now - m.lastReceivedAliveAnnouncement) < ZT_CLUSTER_TIMEOUT) && ((m.x != 0)||(m.y != 0)||(m.z != 0)) && (m.zeroTierPhysicalEndpoints.size() > 0) ) {
  541. const double mdist = _dist3d(m.x,m.y,m.z,px,py,pz);
  542. if (mdist < bestDistance) {
  543. bestDistance = mdist;
  544. bestMember = *mid;
  545. best = m.zeroTierPhysicalEndpoints;
  546. }
  547. }
  548. }
  549. }
  550. // Redirect to a closer member if it has a ZeroTier endpoint address in the same ss_family
  551. for(std::vector<InetAddress>::const_iterator a(best.begin());a!=best.end();++a) {
  552. if (a->ss_family == peerPhysicalAddress.ss_family) {
  553. 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());
  554. redirectTo = *a;
  555. return true;
  556. }
  557. }
  558. TRACE("%s at [%d,%d,%d] is %f from us, no better endpoints found",peerAddress.toString().c_str(),px,py,pz,currentDistance);
  559. return false;
  560. } else {
  561. // TODO: pick based on load if no location info?
  562. return false;
  563. }
  564. }
  565. void Cluster::status(ZT_ClusterStatus &status) const
  566. {
  567. const uint64_t now = RR->node->now();
  568. memset(&status,0,sizeof(ZT_ClusterStatus));
  569. status.myId = _id;
  570. {
  571. ZT_ClusterMemberStatus *const s = &(status.members[status.clusterSize++]);
  572. s->id = _id;
  573. s->alive = 1;
  574. s->x = _x;
  575. s->y = _y;
  576. s->z = _z;
  577. s->load = 0; // TODO
  578. s->peers = RR->topology->countActive();
  579. for(std::vector<InetAddress>::const_iterator ep(_zeroTierPhysicalEndpoints.begin());ep!=_zeroTierPhysicalEndpoints.end();++ep) {
  580. if (s->numZeroTierPhysicalEndpoints >= ZT_CLUSTER_MAX_ZT_PHYSICAL_ADDRESSES) // sanity check
  581. break;
  582. memcpy(&(s->zeroTierPhysicalEndpoints[s->numZeroTierPhysicalEndpoints++]),&(*ep),sizeof(struct sockaddr_storage));
  583. }
  584. }
  585. {
  586. Mutex::Lock _l1(_memberIds_m);
  587. for(std::vector<uint16_t>::const_iterator mid(_memberIds.begin());mid!=_memberIds.end();++mid) {
  588. if (status.clusterSize >= ZT_CLUSTER_MAX_MEMBERS) // sanity check
  589. break;
  590. _Member &m = _members[*mid];
  591. Mutex::Lock ml(m.lock);
  592. ZT_ClusterMemberStatus *const s = &(status.members[status.clusterSize++]);
  593. s->id = *mid;
  594. s->msSinceLastHeartbeat = (unsigned int)std::min((uint64_t)(~((unsigned int)0)),(now - m.lastReceivedAliveAnnouncement));
  595. s->alive = (s->msSinceLastHeartbeat < ZT_CLUSTER_TIMEOUT) ? 1 : 0;
  596. s->x = m.x;
  597. s->y = m.y;
  598. s->z = m.z;
  599. s->load = m.load;
  600. s->peers = m.peers;
  601. for(std::vector<InetAddress>::const_iterator ep(m.zeroTierPhysicalEndpoints.begin());ep!=m.zeroTierPhysicalEndpoints.end();++ep) {
  602. if (s->numZeroTierPhysicalEndpoints >= ZT_CLUSTER_MAX_ZT_PHYSICAL_ADDRESSES) // sanity check
  603. break;
  604. memcpy(&(s->zeroTierPhysicalEndpoints[s->numZeroTierPhysicalEndpoints++]),&(*ep),sizeof(struct sockaddr_storage));
  605. }
  606. }
  607. }
  608. }
  609. void Cluster::_send(uint16_t memberId,StateMessageType type,const void *msg,unsigned int len)
  610. {
  611. if ((len + 3) > (ZT_CLUSTER_MAX_MESSAGE_LENGTH - (24 + 2 + 2))) // sanity check
  612. return;
  613. _Member &m = _members[memberId];
  614. // assumes m.lock is locked!
  615. if ((m.q.size() + len + 3) > ZT_CLUSTER_MAX_MESSAGE_LENGTH)
  616. _flush(memberId);
  617. m.q.append((uint16_t)(len + 1));
  618. m.q.append((uint8_t)type);
  619. m.q.append(msg,len);
  620. }
  621. void Cluster::_flush(uint16_t memberId)
  622. {
  623. _Member &m = _members[memberId];
  624. // assumes m.lock is locked!
  625. if (m.q.size() > (24 + 2 + 2)) { // 16-byte IV + 8-byte MAC + 2 byte from-member-ID + 2 byte to-member-ID
  626. // Create key from member's key and IV
  627. char keytmp[32];
  628. memcpy(keytmp,m.key,32);
  629. for(int i=0;i<8;++i)
  630. keytmp[i] ^= m.q[i];
  631. Salsa20 s20(keytmp,256,m.q.field(8,8));
  632. Utils::burn(keytmp,sizeof(keytmp));
  633. // One-time-use Poly1305 key from first 32 bytes of Salsa20 keystream (as per DJB/NaCl "standard")
  634. char polykey[ZT_POLY1305_KEY_LEN];
  635. memset(polykey,0,sizeof(polykey));
  636. s20.encrypt12(polykey,polykey,sizeof(polykey));
  637. // Encrypt m.q in place
  638. 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);
  639. // Add MAC for authentication (encrypt-then-MAC)
  640. char mac[ZT_POLY1305_MAC_LEN];
  641. Poly1305::compute(mac,reinterpret_cast<const char *>(m.q.data()) + 24,m.q.size() - 24,polykey);
  642. memcpy(m.q.field(16,8),mac,8);
  643. // Send!
  644. _sendFunction(_sendFunctionArg,memberId,m.q.data(),m.q.size());
  645. // Prepare for more
  646. m.q.clear();
  647. char iv[16];
  648. Utils::getSecureRandom(iv,16);
  649. m.q.append(iv,16);
  650. m.q.addSize(8); // room for MAC
  651. m.q.append((uint16_t)_id); // from member ID
  652. m.q.append((uint16_t)memberId); // to member ID
  653. }
  654. }
  655. void Cluster::_doREMOTE_WHOIS(uint64_t fromMemberId,const Packet &remotep)
  656. {
  657. if (remotep.payloadLength() >= ZT_ADDRESS_LENGTH) {
  658. Identity queried(RR->topology->getIdentity(Address(remotep.payload(),ZT_ADDRESS_LENGTH)));
  659. if (queried) {
  660. Buffer<1024> routp;
  661. remotep.source().appendTo(routp);
  662. routp.append((uint8_t)Packet::VERB_OK);
  663. routp.addSize(2); // space for length
  664. routp.append((uint8_t)Packet::VERB_WHOIS);
  665. routp.append(remotep.packetId());
  666. queried.serialize(routp);
  667. Mutex::Lock _l2(_members[fromMemberId].lock);
  668. routp.setAt<uint16_t>(ZT_ADDRESS_LENGTH + 1,(uint16_t)(routp.size() - ZT_ADDRESS_LENGTH - 3));
  669. _send(fromMemberId,CLUSTER_MESSAGE_PROXY_SEND,routp.data(),routp.size());
  670. _flush(fromMemberId);
  671. }
  672. }
  673. }
  674. void Cluster::_doREMOTE_MULTICAST_GATHER(uint64_t fromMemberId,const Packet &remotep)
  675. {
  676. const uint64_t nwid = remotep.at<uint64_t>(ZT_PROTO_VERB_MULTICAST_GATHER_IDX_NETWORK_ID);
  677. const MulticastGroup mg(MAC(remotep.field(ZT_PROTO_VERB_MULTICAST_GATHER_IDX_MAC,6),6),remotep.at<uint32_t>(ZT_PROTO_VERB_MULTICAST_GATHER_IDX_ADI));
  678. unsigned int gatherLimit = remotep.at<uint32_t>(ZT_PROTO_VERB_MULTICAST_GATHER_IDX_GATHER_LIMIT);
  679. const Address remotePeerAddress(remotep.source());
  680. if (gatherLimit) {
  681. Buffer<ZT_PROTO_MAX_PACKET_LENGTH> routp;
  682. remotePeerAddress.appendTo(routp);
  683. routp.append((uint8_t)Packet::VERB_OK);
  684. routp.addSize(2); // space for length
  685. routp.append((uint8_t)Packet::VERB_MULTICAST_GATHER);
  686. routp.append(remotep.packetId());
  687. routp.append(nwid);
  688. mg.mac().appendTo(routp);
  689. routp.append((uint32_t)mg.adi());
  690. if (gatherLimit > ((ZT_CLUSTER_MAX_MESSAGE_LENGTH - 80) / 5))
  691. gatherLimit = ((ZT_CLUSTER_MAX_MESSAGE_LENGTH - 80) / 5);
  692. if (RR->mc->gather(remotePeerAddress,nwid,mg,routp,gatherLimit)) {
  693. routp.setAt<uint16_t>(ZT_ADDRESS_LENGTH + 1,(uint16_t)(routp.size() - ZT_ADDRESS_LENGTH - 3));
  694. Mutex::Lock _l2(_members[fromMemberId].lock);
  695. _send(fromMemberId,CLUSTER_MESSAGE_PROXY_SEND,routp.data(),routp.size());
  696. }
  697. }
  698. }
  699. } // namespace ZeroTier
  700. #endif // ZT_ENABLE_CLUSTER