Bond.cpp 70 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036
  1. /*
  2. * Copyright (c)2013-2021 ZeroTier, Inc.
  3. *
  4. * Use of this software is governed by the Business Source License included
  5. * in the LICENSE.TXT file in the project's root directory.
  6. *
  7. * Change Date: 2026-01-01
  8. *
  9. * On the date above, in accordance with the Business Source License, use
  10. * of this software will be governed by version 2.0 of the Apache License.
  11. */
  12. /****/
  13. #include "Bond.hpp"
  14. #include "Switch.hpp"
  15. #include <cinttypes> // for PRId64, etc. macros
  16. #include <cmath>
  17. #include <cstdio>
  18. #include <string>
  19. // FIXME: remove this suppression and actually fix warnings
  20. #ifdef __GNUC__
  21. #pragma GCC diagnostic ignored "-Wsign-compare"
  22. #endif
  23. namespace ZeroTier {
  24. static unsigned char s_freeRandomByteCounter = 0;
  25. int Bond::_minReqMonitorInterval = ZT_BOND_FAILOVER_DEFAULT_INTERVAL;
  26. uint8_t Bond::_defaultPolicy = ZT_BOND_POLICY_NONE;
  27. Phy<Bond*>* Bond::_phy;
  28. Binder* Bond::_binder;
  29. Mutex Bond::_bonds_m;
  30. Mutex Bond::_links_m;
  31. std::string Bond::_defaultPolicyStr;
  32. std::map<int64_t, SharedPtr<Bond> > Bond::_bonds;
  33. std::map<int64_t, std::string> Bond::_policyTemplateAssignments;
  34. std::map<std::string, SharedPtr<Bond> > Bond::_bondPolicyTemplates;
  35. std::map<std::string, std::vector<SharedPtr<Link> > > Bond::_linkDefinitions;
  36. std::map<std::string, std::map<std::string, SharedPtr<Link> > > Bond::_interfaceToLinkMap;
  37. bool Bond::linkAllowed(std::string& policyAlias, SharedPtr<Link> link)
  38. {
  39. if (! link) {
  40. return false;
  41. }
  42. bool foundInDefinitions = false;
  43. if (_linkDefinitions.count(policyAlias)) {
  44. auto it = _linkDefinitions[policyAlias].begin();
  45. while (it != _linkDefinitions[policyAlias].end()) {
  46. if (link->ifname() == (*it)->ifname()) {
  47. foundInDefinitions = true;
  48. break;
  49. }
  50. ++it;
  51. }
  52. }
  53. return _linkDefinitions[policyAlias].empty() || foundInDefinitions;
  54. }
  55. void Bond::addCustomLink(std::string& policyAlias, SharedPtr<Link> link)
  56. {
  57. Mutex::Lock _l(_links_m);
  58. _linkDefinitions[policyAlias].push_back(link);
  59. auto search = _interfaceToLinkMap[policyAlias].find(link->ifname());
  60. if (search == _interfaceToLinkMap[policyAlias].end()) {
  61. link->setAsUserSpecified(true);
  62. _interfaceToLinkMap[policyAlias].insert(std::pair<std::string, SharedPtr<Link> >(link->ifname(), link));
  63. }
  64. }
  65. bool Bond::addCustomPolicy(const SharedPtr<Bond>& newBond)
  66. {
  67. Mutex::Lock _l(_bonds_m);
  68. if (! _bondPolicyTemplates.count(newBond->policyAlias())) {
  69. _bondPolicyTemplates[newBond->policyAlias()] = newBond;
  70. return true;
  71. }
  72. return false;
  73. }
  74. bool Bond::assignBondingPolicyToPeer(int64_t identity, const std::string& policyAlias)
  75. {
  76. Mutex::Lock _l(_bonds_m);
  77. if (! _policyTemplateAssignments.count(identity)) {
  78. _policyTemplateAssignments[identity] = policyAlias;
  79. return true;
  80. }
  81. return false;
  82. }
  83. SharedPtr<Bond> Bond::getBondByPeerId(int64_t identity)
  84. {
  85. Mutex::Lock _l(_bonds_m);
  86. return _bonds.count(identity) ? _bonds[identity] : SharedPtr<Bond>();
  87. }
  88. bool Bond::setAllMtuByTuple(uint16_t mtu, const std::string& ifStr, const std::string& ipStr)
  89. {
  90. Mutex::Lock _l(_bonds_m);
  91. std::map<int64_t, SharedPtr<Bond> >::iterator bondItr = _bonds.begin();
  92. bool found = false;
  93. while (bondItr != _bonds.end()) {
  94. if (bondItr->second->setMtuByTuple(mtu, ifStr, ipStr)) {
  95. found = true;
  96. }
  97. ++bondItr;
  98. }
  99. return found;
  100. }
  101. bool Bond::setMtuByTuple(uint16_t mtu, const std::string& ifStr, const std::string& ipStr)
  102. {
  103. Mutex::Lock _lp(_paths_m);
  104. bool found = false;
  105. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  106. if (_paths[i].p) {
  107. SharedPtr<Link> sl = getLink(_paths[i].p);
  108. if (sl) {
  109. if (sl->ifname() == ifStr) {
  110. char ipBuf[64] = { 0 };
  111. _paths[i].p->address().toIpString(ipBuf);
  112. std::string newString = std::string(ipBuf);
  113. if (newString == ipStr) {
  114. _paths[i].p->_mtu = mtu;
  115. found = true;
  116. }
  117. }
  118. }
  119. }
  120. }
  121. return found;
  122. }
  123. SharedPtr<Bond> Bond::createBond(const RuntimeEnvironment* renv, const SharedPtr<Peer>& peer)
  124. {
  125. Mutex::Lock _l(_bonds_m);
  126. int64_t identity = peer->identity().address().toInt();
  127. Bond* bond = nullptr;
  128. if (! _bonds.count(identity)) {
  129. if (! _policyTemplateAssignments.count(identity)) {
  130. if (_defaultPolicy) {
  131. bond = new Bond(renv, _defaultPolicy, peer);
  132. bond->debug("new default bond");
  133. }
  134. if (! _defaultPolicy && _defaultPolicyStr.length()) {
  135. bond = new Bond(renv, _bondPolicyTemplates[_defaultPolicyStr].ptr(), peer);
  136. bond->debug("new default custom bond (based on %s)", bond->getPolicyStrByCode(bond->policy()).c_str());
  137. }
  138. }
  139. else {
  140. if (! _bondPolicyTemplates[_policyTemplateAssignments[identity]]) {
  141. bond = new Bond(renv, _defaultPolicy, peer);
  142. bond->debug("peer-specific bond, was specified as %s but the bond definition was not found, using default %s", _policyTemplateAssignments[identity].c_str(), getPolicyStrByCode(_defaultPolicy).c_str());
  143. }
  144. else {
  145. bond = new Bond(renv, _bondPolicyTemplates[_policyTemplateAssignments[identity]].ptr(), peer);
  146. bond->debug("new default bond");
  147. }
  148. }
  149. }
  150. if (bond) {
  151. _bonds[identity] = bond;
  152. /**
  153. * Determine if user has specified anything that could affect the bonding policy's decisions
  154. */
  155. if (_interfaceToLinkMap.count(bond->policyAlias())) {
  156. std::map<std::string, SharedPtr<Link> >::iterator it = _interfaceToLinkMap[bond->policyAlias()].begin();
  157. while (it != _interfaceToLinkMap[bond->policyAlias()].end()) {
  158. if (it->second->isUserSpecified()) {
  159. bond->_userHasSpecifiedLinks = true;
  160. }
  161. if (it->second->isUserSpecified() && it->second->primary()) {
  162. bond->_userHasSpecifiedPrimaryLink = true;
  163. }
  164. if (it->second->isUserSpecified() && it->second->userHasSpecifiedFailoverInstructions()) {
  165. bond->_userHasSpecifiedFailoverInstructions = true;
  166. }
  167. if (it->second->isUserSpecified() && (it->second->capacity() > 0)) {
  168. bond->_userHasSpecifiedLinkCapacities = true;
  169. }
  170. ++it;
  171. }
  172. }
  173. bond->startBond();
  174. return bond;
  175. }
  176. return SharedPtr<Bond>();
  177. }
  178. void Bond::destroyBond(uint64_t peerId)
  179. {
  180. Mutex::Lock _l(_bonds_m);
  181. auto iter = _bonds.find(peerId);
  182. if (iter != _bonds.end()) {
  183. iter->second->stopBond();
  184. _bonds.erase(iter);
  185. }
  186. }
  187. void Bond::stopBond()
  188. {
  189. debug("stopping bond");
  190. _run = false;
  191. }
  192. void Bond::startBond()
  193. {
  194. debug("starting bond");
  195. _run = true;
  196. }
  197. SharedPtr<Link> Bond::getLinkBySocket(const std::string& policyAlias, uint64_t localSocket, bool createIfNeeded = false)
  198. {
  199. Mutex::Lock _l(_links_m);
  200. char ifname[ZT_MAX_PHYSIFNAME] = {};
  201. _binder->getIfName((PhySocket*)((uintptr_t)localSocket), ifname, sizeof(ifname) - 1);
  202. std::string ifnameStr(ifname);
  203. auto search = _interfaceToLinkMap[policyAlias].find(ifnameStr);
  204. if (search == _interfaceToLinkMap[policyAlias].end()) {
  205. if (createIfNeeded) {
  206. SharedPtr<Link> s = new Link(ifnameStr, 0, 0, 0, true, ZT_BOND_SLAVE_MODE_PRIMARY, "");
  207. _interfaceToLinkMap[policyAlias].insert(std::pair<std::string, SharedPtr<Link> >(ifnameStr, s));
  208. return s;
  209. }
  210. else {
  211. return SharedPtr<Link>();
  212. }
  213. }
  214. else {
  215. return search->second;
  216. }
  217. }
  218. SharedPtr<Link> Bond::getLinkByName(const std::string& policyAlias, const std::string& ifname)
  219. {
  220. Mutex::Lock _l(_links_m);
  221. auto search = _interfaceToLinkMap[policyAlias].find(ifname);
  222. if (search != _interfaceToLinkMap[policyAlias].end()) {
  223. return search->second;
  224. }
  225. return SharedPtr<Link>();
  226. }
  227. void Bond::processBackgroundTasks(void* tPtr, const int64_t now)
  228. {
  229. unsigned long _currMinReqMonitorInterval = ZT_BOND_FAILOVER_DEFAULT_INTERVAL;
  230. Mutex::Lock _l(_bonds_m);
  231. std::map<int64_t, SharedPtr<Bond> >::iterator bondItr = _bonds.begin();
  232. while (bondItr != _bonds.end()) {
  233. // Update Bond Controller's background processing timer
  234. _currMinReqMonitorInterval = std::min(_currMinReqMonitorInterval, (unsigned long)(bondItr->second->monitorInterval()));
  235. bondItr->second->processBackgroundBondTasks(tPtr, now);
  236. ++bondItr;
  237. }
  238. _minReqMonitorInterval = std::min(_currMinReqMonitorInterval, (unsigned long)ZT_BOND_FAILOVER_DEFAULT_INTERVAL);
  239. }
  240. Bond::Bond(const RuntimeEnvironment* renv) : RR(renv)
  241. {
  242. initTimers();
  243. }
  244. Bond::Bond(const RuntimeEnvironment* renv, int policy, const SharedPtr<Peer>& peer) : RR(renv), _freeRandomByte((unsigned char)((uintptr_t)this >> 4) ^ ++s_freeRandomByteCounter), _peer(peer), _peerId(_peer->_id.address().toInt())
  245. {
  246. initTimers();
  247. setBondParameters(policy, SharedPtr<Bond>(), false);
  248. _policyAlias = getPolicyStrByCode(policy);
  249. }
  250. Bond::Bond(const RuntimeEnvironment* renv, std::string& basePolicy, std::string& policyAlias, const SharedPtr<Peer>& peer) : RR(renv), _policyAlias(policyAlias), _peer(peer)
  251. {
  252. initTimers();
  253. setBondParameters(getPolicyCodeByStr(basePolicy), SharedPtr<Bond>(), false);
  254. }
  255. Bond::Bond(const RuntimeEnvironment* renv, SharedPtr<Bond> originalBond, const SharedPtr<Peer>& peer)
  256. : RR(renv)
  257. , _freeRandomByte((unsigned char)((uintptr_t)this >> 4) ^ ++s_freeRandomByteCounter)
  258. , _peer(peer)
  259. , _peerId(_peer->_id.address().toInt())
  260. {
  261. initTimers();
  262. setBondParameters(originalBond->_policy, originalBond, true);
  263. }
  264. void Bond::nominatePathToBond(const SharedPtr<Path>& path, int64_t now)
  265. {
  266. Mutex::Lock _l(_paths_m);
  267. debug("attempting to nominate link %s", pathToStr(path).c_str());
  268. /**
  269. * Ensure the link is allowed and the path is not already present
  270. */
  271. if (! RR->bc->linkAllowed(_policyAlias, getLinkBySocket(_policyAlias, path->localSocket(), true))) {
  272. debug("link %s is not allowed according to user-specified rules", pathToStr(path).c_str());
  273. return;
  274. }
  275. bool alreadyPresent = false;
  276. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  277. // Sanity check
  278. if (path.ptr() == _paths[i].p.ptr()) {
  279. alreadyPresent = true;
  280. debug("link %s already exists", pathToStr(path).c_str());
  281. break;
  282. }
  283. }
  284. if (! alreadyPresent) {
  285. SharedPtr<Link> link = getLink(path);
  286. if (link) {
  287. std::string ifnameStr = std::string(link->ifname());
  288. memset(path->_ifname, 0x0, ZT_MAX_PHYSIFNAME);
  289. memcpy(path->_ifname, ifnameStr.c_str(), std::min((int)ifnameStr.length(), ZT_MAX_PHYSIFNAME));
  290. }
  291. /**
  292. * Find somewhere to stick it
  293. */
  294. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  295. if (! _paths[i].p) {
  296. _paths[i].set(now, path);
  297. /**
  298. * Set user preferences and update state variables of other paths on the same link
  299. */
  300. SharedPtr<Link> sl = getLink(_paths[i].p);
  301. if (sl) {
  302. // Determine if there are any other paths on this link
  303. bool bFoundCommonLink = false;
  304. SharedPtr<Link> commonLink = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  305. if (commonLink) {
  306. for (unsigned int j = 0; j < ZT_MAX_PEER_NETWORK_PATHS; ++j) {
  307. if (_paths[j].p && _paths[j].p.ptr() != _paths[i].p.ptr()) {
  308. if (RR->bc->getLinkBySocket(_policyAlias, _paths[j].p->localSocket(), true) == commonLink) {
  309. bFoundCommonLink = true;
  310. _paths[j].onlyPathOnLink = false;
  311. }
  312. }
  313. }
  314. _paths[i].ipvPref = sl->ipvPref();
  315. _paths[i].mode = sl->mode();
  316. _paths[i].enabled = sl->enabled();
  317. _paths[i].localPort = _phy->getLocalPort((PhySocket*)((uintptr_t)path->localSocket()));
  318. _paths[i].onlyPathOnLink = ! bFoundCommonLink;
  319. }
  320. }
  321. log("nominated link %s", pathToStr(path).c_str());
  322. break;
  323. }
  324. }
  325. }
  326. curateBond(now, true);
  327. estimatePathQuality(now);
  328. }
  329. void Bond::addPathToBond(int nominatedIdx, int bondedIdx)
  330. {
  331. // Map bonded set to nominated set
  332. _realIdxMap[bondedIdx] = nominatedIdx;
  333. // Tell the bonding layer that we can now use this path for traffic
  334. _paths[nominatedIdx].bonded = true;
  335. }
  336. SharedPtr<Path> Bond::getAppropriatePath(int64_t now, int32_t flowId)
  337. {
  338. Mutex::Lock _l(_paths_m);
  339. /**
  340. * active-backup
  341. */
  342. if (_policy == ZT_BOND_POLICY_ACTIVE_BACKUP) {
  343. if (_abPathIdx != ZT_MAX_PEER_NETWORK_PATHS && _paths[_abPathIdx].p) {
  344. return _paths[_abPathIdx].p;
  345. }
  346. }
  347. /**
  348. * broadcast
  349. */
  350. if (_policy == ZT_BOND_POLICY_BROADCAST) {
  351. return SharedPtr<Path>(); // Handled in Switch::_trySend()
  352. }
  353. if (! _numBondedPaths) {
  354. return SharedPtr<Path>(); // No paths assigned to bond yet, cannot balance traffic
  355. }
  356. /**
  357. * balance-rr
  358. */
  359. if (_policy == ZT_BOND_POLICY_BALANCE_RR) {
  360. if (_packetsPerLink == 0) {
  361. // Randomly select a path
  362. return _paths[_realIdxMap[_freeRandomByte % _numBondedPaths]].p;
  363. }
  364. if (_rrPacketsSentOnCurrLink < _packetsPerLink) {
  365. // Continue to use this link
  366. ++_rrPacketsSentOnCurrLink;
  367. return _paths[_realIdxMap[_rrIdx]].p;
  368. }
  369. // Reset striping counter
  370. _rrPacketsSentOnCurrLink = 0;
  371. if (_numBondedPaths == 1 || _rrIdx >= (ZT_MAX_PEER_NETWORK_PATHS - 1)) {
  372. _rrIdx = 0;
  373. }
  374. else {
  375. int _tempIdx = _rrIdx;
  376. for (int searchCount = 0; searchCount < (_numBondedPaths - 1); searchCount++) {
  377. _tempIdx = (_tempIdx == (_numBondedPaths - 1)) ? 0 : _tempIdx + 1;
  378. if (_realIdxMap[_tempIdx] != ZT_MAX_PEER_NETWORK_PATHS) {
  379. if (_paths[_realIdxMap[_tempIdx]].p && _paths[_realIdxMap[_tempIdx]].eligible) {
  380. _rrIdx = _tempIdx;
  381. break;
  382. }
  383. }
  384. }
  385. }
  386. if (_paths[_realIdxMap[_rrIdx]].p) {
  387. return _paths[_realIdxMap[_rrIdx]].p;
  388. }
  389. }
  390. /**
  391. * balance-xor/aware
  392. */
  393. if (_policy == ZT_BOND_POLICY_BALANCE_XOR || _policy == ZT_BOND_POLICY_BALANCE_AWARE) {
  394. if (flowId == -1) {
  395. // No specific path required for unclassified traffic, send on anything
  396. int m_idx = _realIdxMap[_freeRandomByte % _numBondedPaths];
  397. return _paths[m_idx].p;
  398. }
  399. Mutex::Lock _l(_flows_m);
  400. std::map<int16_t, SharedPtr<Flow> >::iterator it = _flows.find(flowId);
  401. if (likely(it != _flows.end())) {
  402. it->second->lastActivity = now;
  403. return _paths[it->second->assignedPath].p;
  404. }
  405. else {
  406. unsigned char entropy;
  407. Utils::getSecureRandom(&entropy, 1);
  408. SharedPtr<Flow> flow = createFlow(ZT_MAX_PEER_NETWORK_PATHS, flowId, entropy, now);
  409. _flows[flowId] = flow;
  410. return _paths[flow->assignedPath].p;
  411. }
  412. }
  413. return SharedPtr<Path>();
  414. }
  415. void Bond::recordIncomingInvalidPacket(const SharedPtr<Path>& path)
  416. {
  417. Mutex::Lock _l(_paths_m);
  418. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  419. if (_paths[i].p == path) {
  420. //_paths[i].packetValiditySamples.push(false);
  421. }
  422. }
  423. }
  424. void Bond::recordOutgoingPacket(const SharedPtr<Path>& path, uint64_t packetId, uint16_t payloadLength, const Packet::Verb verb, const int32_t flowId, int64_t now)
  425. {
  426. _freeRandomByte += (unsigned char)(packetId >> 8); // Grab entropy to use in path selection logic
  427. bool isFrame = (verb == Packet::Packet::VERB_ECHO || verb == Packet::VERB_FRAME || verb == Packet::VERB_EXT_FRAME);
  428. bool shouldRecord = (packetId & (ZT_QOS_ACK_DIVISOR - 1) && (verb != Packet::VERB_ACK) && (verb != Packet::VERB_QOS_MEASUREMENT));
  429. if (isFrame || shouldRecord) {
  430. Mutex::Lock _l(_paths_m);
  431. int pathIdx = getNominatedPathIdx(path);
  432. if (pathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  433. return;
  434. }
  435. if (isFrame) {
  436. ++(_paths[pathIdx].packetsOut);
  437. _lastFrame = now;
  438. }
  439. if (shouldRecord) {
  440. //_paths[pathIdx].expectingAckAsOf = now;
  441. //_paths[pathIdx].totalBytesSentSinceLastAckReceived += payloadLength;
  442. //_paths[pathIdx].unackedBytes += payloadLength;
  443. if (_paths[pathIdx].qosStatsOut.size() < ZT_QOS_MAX_PENDING_RECORDS) {
  444. _paths[pathIdx].qosStatsOut[packetId] = now;
  445. }
  446. }
  447. }
  448. if (flowId != ZT_QOS_NO_FLOW) {
  449. Mutex::Lock _l(_flows_m);
  450. if (_flows.count(flowId)) {
  451. _flows[flowId]->bytesOut += payloadLength;
  452. }
  453. }
  454. }
  455. void Bond::recordIncomingPacket(const SharedPtr<Path>& path, uint64_t packetId, uint16_t payloadLength, Packet::Verb verb, int32_t flowId, int64_t now)
  456. {
  457. bool isFrame = (verb == Packet::Packet::VERB_ECHO || verb == Packet::VERB_FRAME || verb == Packet::VERB_EXT_FRAME);
  458. bool shouldRecord = (packetId & (ZT_QOS_ACK_DIVISOR - 1) && (verb != Packet::VERB_ACK) && (verb != Packet::VERB_QOS_MEASUREMENT));
  459. Mutex::Lock _l(_paths_m);
  460. int pathIdx = getNominatedPathIdx(path);
  461. if (pathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  462. return;
  463. }
  464. // Take note of the time that this previously-dead path received a packet
  465. if (! _paths[pathIdx].alive) {
  466. _paths[pathIdx].lastAliveToggle = now;
  467. }
  468. if (isFrame || shouldRecord) {
  469. if (_paths[pathIdx].allowed()) {
  470. if (isFrame) {
  471. ++(_paths[pathIdx].packetsIn);
  472. _lastFrame = now;
  473. }
  474. if (shouldRecord) {
  475. if (_paths[pathIdx].qosStatsIn.size() < ZT_QOS_MAX_PENDING_RECORDS) {
  476. // debug("Recording QoS information (table size = %d)", _paths[pathIdx].qosStatsIn.size());
  477. _paths[pathIdx].qosStatsIn[packetId] = now;
  478. ++(_paths[pathIdx].packetsReceivedSinceLastQoS);
  479. //_paths[pathIdx].packetValiditySamples.push(true);
  480. }
  481. else {
  482. // debug("QoS buffer full, will not record information");
  483. }
  484. /*
  485. if (_paths[pathIdx].ackStatsIn.size() < ZT_ACK_MAX_PENDING_RECORDS) {
  486. //debug("Recording ACK information (table size = %d)", _paths[pathIdx].ackStatsIn.size());
  487. _paths[pathIdx].ackStatsIn[packetId] = payloadLength;
  488. ++(_paths[pathIdx].packetsReceivedSinceLastAck);
  489. }
  490. else {
  491. debug("ACK buffer full, will not record information");
  492. }
  493. */
  494. }
  495. }
  496. }
  497. /**
  498. * Learn new flows and pro-actively create entries for them in the bond so
  499. * that the next time we send a packet out that is part of a flow we know
  500. * which path to use.
  501. */
  502. if ((flowId != ZT_QOS_NO_FLOW) && (_policy == ZT_BOND_POLICY_BALANCE_RR || _policy == ZT_BOND_POLICY_BALANCE_XOR || _policy == ZT_BOND_POLICY_BALANCE_AWARE)) {
  503. Mutex::Lock _l(_flows_m);
  504. SharedPtr<Flow> flow;
  505. if (! _flows.count(flowId)) {
  506. flow = createFlow(pathIdx, flowId, 0, now);
  507. }
  508. else {
  509. flow = _flows[flowId];
  510. }
  511. if (flow) {
  512. flow->bytesIn += payloadLength;
  513. }
  514. }
  515. }
  516. void Bond::receivedQoS(const SharedPtr<Path>& path, int64_t now, int count, uint64_t* rx_id, uint16_t* rx_ts)
  517. {
  518. Mutex::Lock _l(_paths_m);
  519. int pathIdx = getNominatedPathIdx(path);
  520. if (pathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  521. return;
  522. }
  523. _paths[pathIdx].lastQoSReceived = now;
  524. // debug("received QoS packet (sampling %d frames) via %s", count, pathToStr(path).c_str());
  525. // Look up egress times and compute latency values for each record
  526. std::map<uint64_t, uint64_t>::iterator it;
  527. for (int j = 0; j < count; j++) {
  528. it = _paths[pathIdx].qosStatsOut.find(rx_id[j]);
  529. if (it != _paths[pathIdx].qosStatsOut.end()) {
  530. _paths[pathIdx].latencySamples.push(((uint16_t)(now - it->second) - rx_ts[j]) / 2);
  531. // if (_paths[pathIdx].shouldAvoid) {
  532. // debug("RX sample on avoided path %d", pathIdx);
  533. // }
  534. _paths[pathIdx].qosStatsOut.erase(it);
  535. }
  536. }
  537. _paths[pathIdx].qosRecordSize.push(count);
  538. }
  539. void Bond::receivedAck(int pathIdx, int64_t now, int32_t ackedBytes)
  540. {
  541. /*
  542. Mutex::Lock _l(_paths_m);
  543. debug("received ACK of %d bytes on path %s, there are still %d un-acked bytes", ackedBytes, pathToStr(_paths[pathIdx].p).c_str(), _paths[pathIdx].unackedBytes);
  544. _paths[pathIdx].lastAckReceived = now;
  545. _paths[pathIdx].unackedBytes = (ackedBytes > _paths[pathIdx].unackedBytes) ? 0 : _paths[pathIdx].unackedBytes - ackedBytes;
  546. */
  547. }
  548. int32_t Bond::generateQoSPacket(int pathIdx, int64_t now, char* qosBuffer)
  549. {
  550. int32_t len = 0;
  551. std::map<uint64_t, uint64_t>::iterator it = _paths[pathIdx].qosStatsIn.begin();
  552. int i = 0;
  553. int numRecords = std::min(_paths[pathIdx].packetsReceivedSinceLastQoS, ZT_QOS_TABLE_SIZE);
  554. // debug("numRecords=%3d, packetsReceivedSinceLastQoS=%3d, _paths[pathIdx].qosStatsIn.size()=%3zu", numRecords, _paths[pathIdx].packetsReceivedSinceLastQoS, _paths[pathIdx].qosStatsIn.size());
  555. while (i < numRecords && it != _paths[pathIdx].qosStatsIn.end()) {
  556. uint64_t id = it->first;
  557. memcpy(qosBuffer, &id, sizeof(uint64_t));
  558. qosBuffer += sizeof(uint64_t);
  559. uint16_t holdingTime = (uint16_t)(now - it->second);
  560. memcpy(qosBuffer, &holdingTime, sizeof(uint16_t));
  561. qosBuffer += sizeof(uint16_t);
  562. len += sizeof(uint64_t) + sizeof(uint16_t);
  563. _paths[pathIdx].qosStatsIn.erase(it++);
  564. ++i;
  565. }
  566. return len;
  567. }
  568. bool Bond::assignFlowToBondedPath(SharedPtr<Flow>& flow, int64_t now, bool reassign = false)
  569. {
  570. if (! _numBondedPaths) {
  571. debug("unable to assign flow %x (bond has no links)", flow->id);
  572. return false;
  573. }
  574. unsigned int bondedIdx = ZT_MAX_PEER_NETWORK_PATHS;
  575. if (_policy == ZT_BOND_POLICY_BALANCE_XOR) {
  576. bondedIdx = abs((int)(flow->id % _numBondedPaths));
  577. flow->assignPath(_realIdxMap[bondedIdx], now);
  578. ++(_paths[_realIdxMap[bondedIdx]].assignedFlowCount);
  579. }
  580. if (_policy == ZT_BOND_POLICY_BALANCE_AWARE) {
  581. /** balance-aware generally works like balance-xor except that it will try to
  582. * take into account user preferences (or default sane limits) that will discourage
  583. * allocating traffic to links with a lesser perceived "quality" */
  584. int offset = 0;
  585. float bestQuality = 0.0;
  586. int nextBestQualIdx = ZT_MAX_PEER_NETWORK_PATHS;
  587. if (reassign) {
  588. log("attempting to re-assign out-flow %04x previously on idx %d (%u / %zu flows)", flow->id, flow->assignedPath, _paths[_realIdxMap[flow->assignedPath]].assignedFlowCount, _flows.size());
  589. }
  590. else {
  591. debug("attempting to assign flow for the first time");
  592. }
  593. unsigned char entropy;
  594. Utils::getSecureRandom(&entropy, 1);
  595. float randomLinkCapacity = ((float)entropy / 255.0); // Used to random but proportional choices
  596. while (offset < _numBondedPaths) {
  597. unsigned char entropy;
  598. Utils::getSecureRandom(&entropy, 1);
  599. if (reassign) {
  600. bondedIdx = (flow->assignedPath + offset) % (_numBondedPaths);
  601. }
  602. else {
  603. bondedIdx = abs((int)((entropy + offset) % (_numBondedPaths)));
  604. }
  605. // debug("idx=%d, offset=%d, randomCap=%f, actualCap=%f", bondedIdx, offset, randomLinkCapacity, _paths[_realIdxMap[bondedIdx]].relativeLinkCapacity);
  606. if (! _paths[_realIdxMap[bondedIdx]].p) {
  607. continue;
  608. }
  609. if (! _paths[_realIdxMap[bondedIdx]].shouldAvoid && randomLinkCapacity <= _paths[_realIdxMap[bondedIdx]].relativeLinkCapacity) {
  610. // debug(" assign out-flow %04x to link %s (%u / %zu flows)", flow->id, pathToStr(_paths[_realIdxMap[bondedIdx]].p).c_str(), _paths[_realIdxMap[bondedIdx]].assignedFlowCount, _flows.size());
  611. break; // Acceptable -- No violation of quality spec
  612. }
  613. if (_paths[_realIdxMap[bondedIdx]].relativeQuality > bestQuality) {
  614. bestQuality = _paths[_realIdxMap[bondedIdx]].relativeQuality;
  615. nextBestQualIdx = bondedIdx;
  616. // debug(" recording next-best link %f idx %d", _paths[_realIdxMap[bondedIdx]].relativeQuality, bondedIdx);
  617. }
  618. ++offset;
  619. }
  620. if (offset < _numBondedPaths) {
  621. // We were (able) to find a path that didn't violate any of the user's quality requirements
  622. flow->assignPath(_realIdxMap[bondedIdx], now);
  623. ++(_paths[_realIdxMap[bondedIdx]].assignedFlowCount);
  624. // debug(" ABLE to find optimal link %f idx %d", _paths[_realIdxMap[bondedIdx]].relativeQuality, bondedIdx);
  625. }
  626. else {
  627. // We were (unable) to find a path that didn't violate at least one quality requirement, will choose next best option
  628. flow->assignPath(_realIdxMap[nextBestQualIdx], now);
  629. ++(_paths[_realIdxMap[nextBestQualIdx]].assignedFlowCount);
  630. // debug(" UNABLE to find, will use link %f idx %d", _paths[_realIdxMap[nextBestQualIdx]].relativeQuality, nextBestQualIdx);
  631. }
  632. }
  633. if (_policy == ZT_BOND_POLICY_ACTIVE_BACKUP) {
  634. if (_abPathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  635. log("unable to assign out-flow %x (no active backup link)", flow->id);
  636. }
  637. flow->assignPath(_abPathIdx, now);
  638. }
  639. log("assign out-flow %04x to link %s (%u / %zu flows)", flow->id, pathToStr(_paths[flow->assignedPath].p).c_str(), _paths[flow->assignedPath].assignedFlowCount, _flows.size());
  640. return true;
  641. }
  642. SharedPtr<Bond::Flow> Bond::createFlow(int pathIdx, int32_t flowId, unsigned char entropy, int64_t now)
  643. {
  644. if (! _numBondedPaths) {
  645. debug("unable to assign flow %04x (bond has no links)", flowId);
  646. return SharedPtr<Flow>();
  647. }
  648. if (_flows.size() >= ZT_FLOW_MAX_COUNT) {
  649. debug("forget oldest flow (max flows reached: %d)", ZT_FLOW_MAX_COUNT);
  650. forgetFlowsWhenNecessary(0, true, now);
  651. }
  652. SharedPtr<Flow> flow = new Flow(flowId, now);
  653. _flows[flowId] = flow;
  654. /**
  655. * Add a flow with a given Path already provided. This is the case when a packet
  656. * is received on a path but no flow exists, in this case we simply assign the path
  657. * that the remote peer chose for us.
  658. */
  659. if (pathIdx != ZT_MAX_PEER_NETWORK_PATHS) {
  660. flow->assignPath(pathIdx, now);
  661. _paths[pathIdx].assignedFlowCount++;
  662. debug("assign in-flow %04x to link %s (%u / %zu)", flow->id, pathToStr(_paths[pathIdx].p).c_str(), _paths[pathIdx].assignedFlowCount, _flows.size());
  663. }
  664. /**
  665. * Add a flow when no path was provided. This means that it is an outgoing packet
  666. * and that it is up to the local peer to decide how to load-balance its transmission.
  667. */
  668. else {
  669. assignFlowToBondedPath(flow, now);
  670. }
  671. return flow;
  672. }
  673. void Bond::forgetFlowsWhenNecessary(uint64_t age, bool oldest, int64_t now)
  674. {
  675. std::map<int16_t, SharedPtr<Flow> >::iterator it = _flows.begin();
  676. std::map<int16_t, SharedPtr<Flow> >::iterator oldestFlow = _flows.end();
  677. SharedPtr<Flow> expiredFlow;
  678. if (age) { // Remove by specific age
  679. while (it != _flows.end()) {
  680. if (it->second->age(now) > age) {
  681. debug("forget flow %04x (age %" PRId64 ") (%u / %zu)", it->first, it->second->age(now), _paths[it->second->assignedPath].assignedFlowCount, (_flows.size() - 1));
  682. _paths[it->second->assignedPath].assignedFlowCount--;
  683. it = _flows.erase(it);
  684. }
  685. else {
  686. ++it;
  687. }
  688. }
  689. }
  690. else if (oldest) { // Remove single oldest by natural expiration
  691. uint64_t maxAge = 0;
  692. while (it != _flows.end()) {
  693. if (it->second->age(now) > maxAge) {
  694. maxAge = (now - it->second->age(now));
  695. oldestFlow = it;
  696. }
  697. ++it;
  698. }
  699. if (oldestFlow != _flows.end()) {
  700. debug("forget oldest flow %04x (age %" PRId64 ") (total flows: %zu)", oldestFlow->first, oldestFlow->second->age(now), _flows.size() - 1);
  701. _paths[oldestFlow->second->assignedPath].assignedFlowCount--;
  702. _flows.erase(oldestFlow);
  703. }
  704. }
  705. }
  706. void Bond::processIncomingPathNegotiationRequest(uint64_t now, SharedPtr<Path>& path, int16_t remoteUtility)
  707. {
  708. char pathStr[64] = { 0 };
  709. if (_abLinkSelectMethod != ZT_BOND_RESELECTION_POLICY_OPTIMIZE) {
  710. return;
  711. }
  712. Mutex::Lock _l(_paths_m);
  713. int pathIdx = getNominatedPathIdx(path);
  714. if (pathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  715. return;
  716. }
  717. _paths[pathIdx].p->address().toString(pathStr);
  718. if (! _lastPathNegotiationCheck) {
  719. return;
  720. }
  721. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[pathIdx].p->localSocket());
  722. if (link) {
  723. if (remoteUtility > _localUtility) {
  724. _paths[pathIdx].p->address().toString(pathStr);
  725. debug("peer suggests alternate link %s/%s, remote utility (%d) greater than local utility (%d), switching to suggested link\n", link->ifname().c_str(), pathStr, remoteUtility, _localUtility);
  726. _negotiatedPathIdx = pathIdx;
  727. }
  728. if (remoteUtility < _localUtility) {
  729. debug("peer suggests alternate link %s/%s, remote utility (%d) less than local utility (%d), not switching\n", link->ifname().c_str(), pathStr, remoteUtility, _localUtility);
  730. }
  731. if (remoteUtility == _localUtility) {
  732. debug("peer suggests alternate link %s/%s, remote utility (%d) equal to local utility (%d)\n", link->ifname().c_str(), pathStr, remoteUtility, _localUtility);
  733. if (_peer->_id.address().toInt() > RR->node->identity().address().toInt()) {
  734. debug("agree with peer to use alternate link %s/%s\n", link->ifname().c_str(), pathStr);
  735. _negotiatedPathIdx = pathIdx;
  736. }
  737. else {
  738. debug("ignore petition from peer to use alternate link %s/%s\n", link->ifname().c_str(), pathStr);
  739. }
  740. }
  741. }
  742. }
  743. void Bond::pathNegotiationCheck(void* tPtr, int64_t now)
  744. {
  745. int maxInPathIdx = ZT_MAX_PEER_NETWORK_PATHS;
  746. int maxOutPathIdx = ZT_MAX_PEER_NETWORK_PATHS;
  747. uint64_t maxInCount = 0;
  748. uint64_t maxOutCount = 0;
  749. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  750. if (! _paths[i].p) {
  751. continue;
  752. }
  753. if (_paths[i].packetsIn > maxInCount) {
  754. maxInCount = _paths[i].packetsIn;
  755. maxInPathIdx = i;
  756. }
  757. if (_paths[i].packetsOut > maxOutCount) {
  758. maxOutCount = _paths[i].packetsOut;
  759. maxOutPathIdx = i;
  760. }
  761. _paths[i].resetPacketCounts();
  762. }
  763. bool _peerLinksSynchronized = ((maxInPathIdx != ZT_MAX_PEER_NETWORK_PATHS) && (maxOutPathIdx != ZT_MAX_PEER_NETWORK_PATHS) && (maxInPathIdx != maxOutPathIdx)) ? false : true;
  764. /**
  765. * Determine utility and attempt to petition remote peer to switch to our chosen path
  766. */
  767. if (! _peerLinksSynchronized) {
  768. _localUtility = _paths[maxOutPathIdx].failoverScore - _paths[maxInPathIdx].failoverScore;
  769. if (_paths[maxOutPathIdx].negotiated) {
  770. _localUtility -= ZT_BOND_FAILOVER_HANDICAP_NEGOTIATED;
  771. }
  772. if ((now - _lastSentPathNegotiationRequest) > ZT_PATH_NEGOTIATION_CUTOFF_TIME) {
  773. // fprintf(stderr, "BT: (sync) it's been long enough, sending more requests.\n");
  774. _numSentPathNegotiationRequests = 0;
  775. }
  776. if (_numSentPathNegotiationRequests < ZT_PATH_NEGOTIATION_TRY_COUNT) {
  777. if (_localUtility >= 0) {
  778. // fprintf(stderr, "BT: (sync) paths appear to be out of sync (utility=%d)\n", _localUtility);
  779. sendPATH_NEGOTIATION_REQUEST(tPtr, _paths[maxOutPathIdx].p);
  780. ++_numSentPathNegotiationRequests;
  781. _lastSentPathNegotiationRequest = now;
  782. // fprintf(stderr, "sending request to use %s on %s, ls=%llx, utility=%d\n", pathStr, link->ifname().c_str(), _paths[maxOutPathIdx].p->localSocket(), _localUtility);
  783. }
  784. }
  785. /**
  786. * Give up negotiating and consider switching
  787. */
  788. else if ((now - _lastSentPathNegotiationRequest) > (2 * ZT_BOND_OPTIMIZE_INTERVAL)) {
  789. if (_localUtility == 0) {
  790. // There's no loss to us, just switch without sending a another request
  791. // fprintf(stderr, "BT: (sync) giving up, switching to remote peer's path.\n");
  792. _negotiatedPathIdx = maxInPathIdx;
  793. }
  794. }
  795. }
  796. }
  797. void Bond::sendPATH_NEGOTIATION_REQUEST(void* tPtr, int pathIdx)
  798. {
  799. debug("send link negotiation request to peer via link %s, local utility is %d", pathToStr(_paths[pathIdx].p).c_str(), _localUtility);
  800. if (_abLinkSelectMethod != ZT_BOND_RESELECTION_POLICY_OPTIMIZE) {
  801. return;
  802. }
  803. Packet outp(_peer->_id.address(), RR->identity.address(), Packet::VERB_PATH_NEGOTIATION_REQUEST);
  804. outp.append<int16_t>(_localUtility);
  805. if (_paths[pathIdx].p->address()) {
  806. Metrics::pkt_path_negotiation_request_out++;
  807. outp.armor(_peer->key(), false, _peer->aesKeysIfSupported());
  808. RR->node->putPacket(tPtr, _paths[pathIdx].p->localSocket(), _paths[pathIdx].p->address(), outp.data(), outp.size());
  809. _overheadBytes += outp.size();
  810. }
  811. }
  812. void Bond::sendACK(void* tPtr, int pathIdx, int64_t localSocket, const InetAddress& atAddress, int64_t now)
  813. {
  814. /*
  815. Packet outp(_peer->_id.address(), RR->identity.address(), Packet::VERB_ACK);
  816. int32_t bytesToAck = 0;
  817. std::map<uint64_t, uint64_t>::iterator it = _paths[pathIdx].ackStatsIn.begin();
  818. while (it != _paths[pathIdx].ackStatsIn.end()) {
  819. bytesToAck += it->second;
  820. ++it;
  821. }
  822. debug("sending ACK of %d bytes on path %s (table size = %zu)", bytesToAck, pathToStr(_paths[pathIdx].p).c_str(), _paths[pathIdx].ackStatsIn.size());
  823. outp.append<uint32_t>(bytesToAck);
  824. if (atAddress) {
  825. outp.armor(_peer->key(), false, _peer->aesKeysIfSupported());
  826. RR->node->putPacket(tPtr, localSocket, atAddress, outp.data(), outp.size());
  827. }
  828. else {
  829. RR->sw->send(tPtr, outp, false);
  830. }
  831. _paths[pathIdx].ackStatsIn.clear();
  832. _paths[pathIdx].packetsReceivedSinceLastAck = 0;
  833. _paths[pathIdx].lastAckSent = now;
  834. */
  835. }
  836. void Bond::sendQOS_MEASUREMENT(void* tPtr, int pathIdx, int64_t localSocket, const InetAddress& atAddress, int64_t now)
  837. {
  838. int64_t _now = RR->node->now();
  839. Packet outp(_peer->_id.address(), RR->identity.address(), Packet::VERB_QOS_MEASUREMENT);
  840. char qosData[ZT_QOS_MAX_PACKET_SIZE];
  841. int16_t len = generateQoSPacket(pathIdx, _now, qosData);
  842. if (len) {
  843. // debug("sending QOS via link %s (len=%d)", pathToStr(_paths[pathIdx].p).c_str(), len);
  844. outp.append(qosData, len);
  845. if (atAddress) {
  846. outp.armor(_peer->key(), false, _peer->aesKeysIfSupported());
  847. RR->node->putPacket(tPtr, localSocket, atAddress, outp.data(), outp.size());
  848. }
  849. else {
  850. RR->sw->send(tPtr, outp, false);
  851. }
  852. Metrics::pkt_qos_out++;
  853. _paths[pathIdx].packetsReceivedSinceLastQoS = 0;
  854. _paths[pathIdx].lastQoSMeasurement = now;
  855. _overheadBytes += outp.size();
  856. }
  857. }
  858. void Bond::processBackgroundBondTasks(void* tPtr, int64_t now)
  859. {
  860. if (! _run) {
  861. return;
  862. }
  863. if (! _peer->_localMultipathSupported || (now - _lastBackgroundTaskCheck) < ZT_BOND_BACKGROUND_TASK_MIN_INTERVAL) {
  864. return;
  865. }
  866. _lastBackgroundTaskCheck = now;
  867. Mutex::Lock _l(_paths_m);
  868. curateBond(now, false);
  869. if ((now - _lastQualityEstimation) > _qualityEstimationInterval) {
  870. _lastQualityEstimation = now;
  871. estimatePathQuality(now);
  872. }
  873. dumpInfo(now, false);
  874. // Send ambient monitoring traffic
  875. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  876. if (_paths[i].p && _paths[i].allowed()) {
  877. if (_isLeaf) {
  878. if ((_monitorInterval > 0) && (((now - _paths[i].p->_lastIn) >= (_paths[i].alive ? _monitorInterval : _failoverInterval)))) {
  879. if ((_peer->remoteVersionProtocol() >= 5) && (! ((_peer->remoteVersionMajor() == 1) && (_peer->remoteVersionMinor() == 1) && (_peer->remoteVersionRevision() == 0)))) {
  880. Packet outp(_peer->address(), RR->identity.address(), Packet::VERB_ECHO); // ECHO (this is our bond's heartbeat)
  881. outp.armor(_peer->key(), true, _peer->aesKeysIfSupported());
  882. RR->node->expectReplyTo(outp.packetId());
  883. RR->node->putPacket(tPtr, _paths[i].p->localSocket(), _paths[i].p->address(), outp.data(), outp.size());
  884. _paths[i].p->_lastOut = now;
  885. _overheadBytes += outp.size();
  886. Metrics::pkt_echo_out++;
  887. // debug("tx: verb 0x%-2x of len %4d via %s (ECHO)", Packet::VERB_ECHO, outp.size(), pathToStr(_paths[i].p).c_str());
  888. }
  889. }
  890. // QOS
  891. if (_paths[i].needsToSendQoS(now, _qosSendInterval)) {
  892. sendQOS_MEASUREMENT(tPtr, i, _paths[i].p->localSocket(), _paths[i].p->address(), now);
  893. }
  894. // ACK
  895. /*
  896. if (_paths[i].needsToSendAck(now, _ackSendInterval)) {
  897. sendACK(tPtr, i, _paths[i].p->localSocket(), _paths[i].p->address(), now);
  898. }
  899. */
  900. }
  901. }
  902. }
  903. // Perform periodic background tasks unique to each bonding policy
  904. switch (_policy) {
  905. case ZT_BOND_POLICY_ACTIVE_BACKUP:
  906. processActiveBackupTasks(tPtr, now);
  907. break;
  908. case ZT_BOND_POLICY_BROADCAST:
  909. break;
  910. case ZT_BOND_POLICY_BALANCE_RR:
  911. case ZT_BOND_POLICY_BALANCE_XOR:
  912. case ZT_BOND_POLICY_BALANCE_AWARE:
  913. processBalanceTasks(now);
  914. break;
  915. default:
  916. break;
  917. }
  918. // Check whether or not a path negotiation needs to be performed
  919. if (((now - _lastPathNegotiationCheck) > ZT_BOND_OPTIMIZE_INTERVAL) && _allowPathNegotiation) {
  920. _lastPathNegotiationCheck = now;
  921. pathNegotiationCheck(tPtr, now);
  922. }
  923. }
  924. void Bond::curateBond(int64_t now, bool rebuildBond)
  925. {
  926. uint8_t tmpNumAliveLinks = 0;
  927. uint8_t tmpNumTotalLinks = 0;
  928. /**
  929. * Update path state variables. State variables are used so that critical
  930. * blocks that perform fast packet processing won't need to make as many
  931. * function calls or computations.
  932. */
  933. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  934. if (! _paths[i].p) {
  935. continue;
  936. }
  937. // Whether this path is still in its trial period
  938. bool inTrial = (now - _paths[i].whenNominated) < ZT_BOND_OPTIMIZE_INTERVAL;
  939. /**
  940. * Remove expired or invalid links from bond
  941. */
  942. SharedPtr<Link> link = getLink(_paths[i].p);
  943. if (! link) {
  944. log("link is no longer valid, removing from bond");
  945. _paths[i].p->_valid = false;
  946. _paths[i] = NominatedPath();
  947. _paths[i].p = SharedPtr<Path>();
  948. continue;
  949. }
  950. if ((now - _paths[i].lastEligibility) > (ZT_PEER_EXPIRED_PATH_TRIAL_PERIOD) && ! inTrial) {
  951. log("link (%s) has expired or is invalid, removing from bond", pathToStr(_paths[i].p).c_str());
  952. _paths[i] = NominatedPath();
  953. _paths[i].p = SharedPtr<Path>();
  954. continue;
  955. }
  956. tmpNumTotalLinks++;
  957. if (_paths[i].eligible) {
  958. tmpNumAliveLinks++;
  959. }
  960. /**
  961. * Determine aliveness
  962. */
  963. _paths[i].alive = _isLeaf ? (now - _paths[i].p->_lastIn) < _failoverInterval : (now - _paths[i].p->_lastIn) < ZT_PEER_PATH_EXPIRATION;
  964. /**
  965. * Determine current eligibility
  966. */
  967. bool currEligibility = false;
  968. // Simple RX age (driven by packets of any type and gratuitous VERB_HELLOs)
  969. bool acceptableAge = _isLeaf ? (_paths[i].p->age(now) < (_failoverInterval + _downDelay)) : _paths[i].alive;
  970. // Whether we've waited long enough since the link last came online
  971. bool satisfiedUpDelay = (now - _paths[i].lastAliveToggle) >= _upDelay;
  972. // How long since the last QoS was received (Must be less than ZT_PEER_PATH_EXPIRATION since the remote peer's _qosSendInterval isn't known)
  973. bool acceptableQoSAge = (_paths[i].lastQoSReceived == 0 && inTrial) || ((now - _paths[i].lastQoSReceived) < ZT_PEER_EXPIRED_PATH_TRIAL_PERIOD);
  974. currEligibility = _paths[i].allowed() && ((acceptableAge && satisfiedUpDelay && acceptableQoSAge) || inTrial);
  975. if (currEligibility) {
  976. _paths[i].lastEligibility = now;
  977. }
  978. /**
  979. * Note eligibility state change (if any) and take appropriate action
  980. */
  981. if (currEligibility != _paths[i].eligible) {
  982. if (currEligibility == 0) {
  983. log("link %s is no longer eligible", pathToStr(_paths[i].p).c_str());
  984. }
  985. if (currEligibility == 1) {
  986. log("link %s is eligible", pathToStr(_paths[i].p).c_str());
  987. }
  988. debug("\t[%d] allowed=%d, age=%d, qa=%d, ud=%d, trial=%d", i, _paths[i].allowed(), acceptableAge, acceptableQoSAge, satisfiedUpDelay, inTrial);
  989. dumpPathStatus(now, i);
  990. if (currEligibility) {
  991. rebuildBond = true;
  992. }
  993. if (! currEligibility) {
  994. _paths[i].adjustRefractoryPeriod(now, _defaultPathRefractoryPeriod, ! currEligibility);
  995. if (_paths[i].bonded) {
  996. debug("link %s was bonded, flow reallocation will occur soon", pathToStr(_paths[i].p).c_str());
  997. rebuildBond = true;
  998. _paths[i].shouldAvoid = true;
  999. _paths[i].bonded = false;
  1000. }
  1001. }
  1002. }
  1003. if (currEligibility) {
  1004. _paths[i].adjustRefractoryPeriod(now, _defaultPathRefractoryPeriod, false);
  1005. }
  1006. _paths[i].eligible = currEligibility;
  1007. }
  1008. /**
  1009. * Trigger status report if number of links change
  1010. */
  1011. _numAliveLinks = tmpNumAliveLinks;
  1012. _numTotalLinks = tmpNumTotalLinks;
  1013. if ((_numAliveLinks != tmpNumAliveLinks) || (_numTotalLinks != tmpNumTotalLinks)) {
  1014. dumpInfo(now, true);
  1015. }
  1016. /**
  1017. * Check for failure of (all) primary links and inform bond to use spares if present
  1018. */
  1019. bool foundUsablePrimaryPath = false;
  1020. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1021. // debug("[%d], bonded=%d, alive=%d", i, _paths[i].bonded , _paths[i].alive);
  1022. if (_paths[i].p && _paths[i].bonded && _paths[i].alive) {
  1023. foundUsablePrimaryPath = true;
  1024. }
  1025. }
  1026. rebuildBond = rebuildBond ? true : ! foundUsablePrimaryPath;
  1027. /**
  1028. * Curate the set of paths that are part of the bond proper. Select a set of paths
  1029. * per logical link according to eligibility and user-specified constraints.
  1030. */
  1031. if ((_policy == ZT_BOND_POLICY_BALANCE_RR) || (_policy == ZT_BOND_POLICY_BALANCE_XOR) || (_policy == ZT_BOND_POLICY_BALANCE_AWARE)) {
  1032. if (! _numBondedPaths) {
  1033. rebuildBond = true;
  1034. }
  1035. if (rebuildBond) {
  1036. // Clear previous bonded index mapping
  1037. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1038. _realIdxMap[i] = ZT_MAX_PEER_NETWORK_PATHS;
  1039. _paths[i].bonded = false;
  1040. }
  1041. int updatedBondedPathCount = 0;
  1042. // Build map associating paths with local physical links. Will be selected from in next step
  1043. std::map<SharedPtr<Link>, std::vector<int> > linkMap;
  1044. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1045. if (_paths[i].p) {
  1046. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1047. if (link) {
  1048. linkMap[link].push_back(i);
  1049. }
  1050. }
  1051. }
  1052. // Re-form bond from link<->path map
  1053. std::map<SharedPtr<Link>, std::vector<int> >::iterator it = linkMap.begin();
  1054. while (it != linkMap.end()) {
  1055. SharedPtr<Link> link = it->first;
  1056. // Bond a spare link if required (no viable primary links left)
  1057. if (! foundUsablePrimaryPath) {
  1058. // debug("no usable primary links remain, will attempt to use spare if available");
  1059. for (int j = 0; j < it->second.size(); j++) {
  1060. int idx = it->second.at(j);
  1061. if (! _paths[idx].p || ! _paths[idx].eligible || ! _paths[idx].allowed() || ! _paths[idx].isSpare()) {
  1062. continue;
  1063. }
  1064. addPathToBond(idx, updatedBondedPathCount);
  1065. ++updatedBondedPathCount;
  1066. debug("add %s (spare)", pathToStr(_paths[idx].p).c_str());
  1067. }
  1068. }
  1069. int ipvPref = link->ipvPref();
  1070. // If user has no address type preference, then use every path we find on a link
  1071. if (ipvPref == 0) {
  1072. for (int j = 0; j < it->second.size(); j++) {
  1073. int idx = it->second.at(j);
  1074. if (! _paths[idx].p || ! _paths[idx].eligible || ! _paths[idx].allowed() || _paths[idx].isSpare()) {
  1075. continue;
  1076. }
  1077. addPathToBond(idx, updatedBondedPathCount);
  1078. ++updatedBondedPathCount;
  1079. debug("add %s (no user addr preference)", pathToStr(_paths[idx].p).c_str());
  1080. }
  1081. }
  1082. // If the user prefers to only use one address type (IPv4 or IPv6)
  1083. if (ipvPref == 4 || ipvPref == 6) {
  1084. for (int j = 0; j < it->second.size(); j++) {
  1085. int idx = it->second.at(j);
  1086. if (! _paths[idx].p || ! _paths[idx].eligible || _paths[idx].isSpare()) {
  1087. continue;
  1088. }
  1089. if (! _paths[idx].allowed()) {
  1090. debug("did not add %s (user addr preference %d)", pathToStr(_paths[idx].p).c_str(), ipvPref);
  1091. continue;
  1092. }
  1093. addPathToBond(idx, updatedBondedPathCount);
  1094. ++updatedBondedPathCount;
  1095. debug("add path %s (user addr preference %d)", pathToStr(_paths[idx].p).c_str(), ipvPref);
  1096. }
  1097. }
  1098. // If the users prefers one address type to another, try to find at least
  1099. // one path of that type before considering others.
  1100. if (ipvPref == 46 || ipvPref == 64) {
  1101. bool foundPreferredPath = false;
  1102. // Search for preferred paths
  1103. for (int j = 0; j < it->second.size(); j++) {
  1104. int idx = it->second.at(j);
  1105. if (! _paths[idx].p || ! _paths[idx].eligible || ! _paths[idx].allowed() || _paths[idx].isSpare()) {
  1106. continue;
  1107. }
  1108. if (_paths[idx].preferred()) {
  1109. addPathToBond(idx, updatedBondedPathCount);
  1110. ++updatedBondedPathCount;
  1111. debug("add %s (user addr preference %d)", pathToStr(_paths[idx].p).c_str(), ipvPref);
  1112. foundPreferredPath = true;
  1113. }
  1114. }
  1115. // Unable to find a path that matches user preference, settle for another address type
  1116. if (! foundPreferredPath) {
  1117. debug("did not find first-choice path type on link %s (user preference %d)", link->ifname().c_str(), ipvPref);
  1118. for (int j = 0; j < it->second.size(); j++) {
  1119. int idx = it->second.at(j);
  1120. if (! _paths[idx].p || ! _paths[idx].eligible || _paths[idx].isSpare()) {
  1121. continue;
  1122. }
  1123. addPathToBond(idx, updatedBondedPathCount);
  1124. ++updatedBondedPathCount;
  1125. debug("add %s (user addr preference %d)", pathToStr(_paths[idx].p).c_str(), ipvPref);
  1126. foundPreferredPath = true;
  1127. }
  1128. }
  1129. }
  1130. ++it; // Next link
  1131. }
  1132. _numBondedPaths = updatedBondedPathCount;
  1133. if (_policy == ZT_BOND_POLICY_BALANCE_RR) {
  1134. // Cause a RR reset since the current index might no longer be valid
  1135. _rrPacketsSentOnCurrLink = _packetsPerLink;
  1136. _rrIdx = 0;
  1137. }
  1138. }
  1139. }
  1140. }
  1141. void Bond::estimatePathQuality(int64_t now)
  1142. {
  1143. float lat[ZT_MAX_PEER_NETWORK_PATHS] = { 0 };
  1144. float pdv[ZT_MAX_PEER_NETWORK_PATHS] = { 0 };
  1145. float plr[ZT_MAX_PEER_NETWORK_PATHS] = { 0 };
  1146. float per[ZT_MAX_PEER_NETWORK_PATHS] = { 0 };
  1147. float maxLAT = 0;
  1148. float maxPDV = 0;
  1149. float maxPLR = 0;
  1150. float maxPER = 0;
  1151. float absoluteQuality[ZT_MAX_PEER_NETWORK_PATHS] = { 0 };
  1152. float totQuality = 0.0f;
  1153. // Process observation samples, compute summary statistics, and compute relative link qualities
  1154. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1155. if (! _paths[i].p || ! _paths[i].allowed()) {
  1156. continue;
  1157. }
  1158. // Drain unacknowledged QoS records
  1159. int qosRecordTimeout = (_qosSendInterval * 3);
  1160. std::map<uint64_t, uint64_t>::iterator it = _paths[i].qosStatsOut.begin();
  1161. int numDroppedQosOutRecords = 0;
  1162. while (it != _paths[i].qosStatsOut.end()) {
  1163. if ((now - it->second) >= qosRecordTimeout) {
  1164. it = _paths[i].qosStatsOut.erase(it);
  1165. ++numDroppedQosOutRecords;
  1166. }
  1167. else {
  1168. ++it;
  1169. }
  1170. }
  1171. if (numDroppedQosOutRecords) {
  1172. // debug("dropped %d QOS out-records", numDroppedQosOutRecords);
  1173. }
  1174. /*
  1175. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1176. if (! _paths[i].p) {
  1177. continue;
  1178. }
  1179. // if ((now - _paths[i].lastAckReceived) > ackSendInterval) {
  1180. // debug("been a while since ACK");
  1181. // if (_paths[i].unackedBytes > 0) {
  1182. // _paths[i].unackedBytes / _paths[i].bytesSen
  1183. // }
  1184. // }
  1185. }
  1186. */
  1187. it = _paths[i].qosStatsIn.begin();
  1188. int numDroppedQosInRecords = 0;
  1189. while (it != _paths[i].qosStatsIn.end()) {
  1190. if ((now - it->second) >= qosRecordTimeout) {
  1191. it = _paths[i].qosStatsIn.erase(it);
  1192. ++numDroppedQosInRecords;
  1193. }
  1194. else {
  1195. ++it;
  1196. }
  1197. }
  1198. if (numDroppedQosInRecords) {
  1199. // debug("dropped %d QOS in-records", numDroppedQosInRecords);
  1200. }
  1201. absoluteQuality[i] = 0;
  1202. totQuality = 0;
  1203. // Normalize raw observations according to sane limits and/or user specified values
  1204. lat[i] = 1.0 / expf(4 * Utils::normalize(_paths[i].latency, 0, _qw[ZT_QOS_LAT_MAX_IDX], 0, 1));
  1205. pdv[i] = 1.0 / expf(4 * Utils::normalize(_paths[i].latencyVariance, 0, _qw[ZT_QOS_PDV_MAX_IDX], 0, 1));
  1206. plr[i] = 1.0 / expf(4 * Utils::normalize(_paths[i].packetLossRatio, 0, _qw[ZT_QOS_PLR_MAX_IDX], 0, 1));
  1207. per[i] = 1.0 / expf(4 * Utils::normalize(_paths[i].packetErrorRatio, 0, _qw[ZT_QOS_PER_MAX_IDX], 0, 1));
  1208. // Record bond-wide maximums to determine relative values
  1209. maxLAT = lat[i] > maxLAT ? lat[i] : maxLAT;
  1210. maxPDV = pdv[i] > maxPDV ? pdv[i] : maxPDV;
  1211. maxPLR = plr[i] > maxPLR ? plr[i] : maxPLR;
  1212. maxPER = per[i] > maxPER ? per[i] : maxPER;
  1213. }
  1214. // Compute relative user-specified link capacities (may change during life of Bond)
  1215. int maxObservedLinkCap = 0;
  1216. // Find current maximum
  1217. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1218. if (_paths[i].p && _paths[i].allowed()) {
  1219. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1220. if (link) {
  1221. int linkSpeed = link->capacity();
  1222. _paths[i].p->_givenLinkSpeed = linkSpeed;
  1223. _paths[i].p->_mtu = link->mtu() ? link->mtu() : _paths[i].p->_mtu;
  1224. _paths[i].p->_assignedFlowCount = _paths[i].assignedFlowCount;
  1225. maxObservedLinkCap = linkSpeed > maxObservedLinkCap ? linkSpeed : maxObservedLinkCap;
  1226. }
  1227. }
  1228. }
  1229. // Compute relative link capacity (Used for weighting traffic allocations)
  1230. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1231. if (_paths[i].p && _paths[i].allowed()) {
  1232. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1233. if (link) {
  1234. float relativeCapacity = (link->capacity() / (float)maxObservedLinkCap);
  1235. link->setRelativeCapacity(relativeCapacity);
  1236. _paths[i].relativeLinkCapacity = relativeCapacity;
  1237. }
  1238. }
  1239. }
  1240. // Convert metrics to relative quantities and apply contribution weights
  1241. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1242. if (_paths[i].p && _paths[i].bonded) {
  1243. absoluteQuality[i] += ((maxLAT > 0.0f ? lat[i] / maxLAT : 0.0f) * _qw[ZT_QOS_LAT_WEIGHT_IDX]);
  1244. absoluteQuality[i] += ((maxPDV > 0.0f ? pdv[i] / maxPDV : 0.0f) * _qw[ZT_QOS_PDV_WEIGHT_IDX]);
  1245. absoluteQuality[i] += ((maxPLR > 0.0f ? plr[i] / maxPLR : 0.0f) * _qw[ZT_QOS_PLR_WEIGHT_IDX]);
  1246. absoluteQuality[i] += ((maxPER > 0.0f ? per[i] / maxPER : 0.0f) * _qw[ZT_QOS_PER_WEIGHT_IDX]);
  1247. absoluteQuality[i] *= _paths[i].relativeLinkCapacity;
  1248. totQuality += absoluteQuality[i];
  1249. }
  1250. }
  1251. // Compute quality of link relative to all others in the bond (also accounting for stated link capacity)
  1252. if (totQuality > 0.0) {
  1253. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1254. if (_paths[i].p && _paths[i].bonded) {
  1255. _paths[i].relativeQuality = absoluteQuality[i] / totQuality;
  1256. // debug("[%2d], abs=%f, tot=%f, rel=%f, relcap=%f", i, absoluteQuality[i], totQuality, _paths[i].relativeQuality, _paths[i].relativeLinkCapacity);
  1257. }
  1258. }
  1259. }
  1260. // Compute summary statistics
  1261. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1262. if (! _paths[i].p || ! _paths[i].allowed()) {
  1263. continue;
  1264. }
  1265. // Compute/Smooth average of real-world observations
  1266. if (_paths[i].latencySamples.count() >= ZT_QOS_SHORTTERM_SAMPLE_WIN_MIN_REQ_SIZE) {
  1267. _paths[i].latency = _paths[i].latencySamples.mean();
  1268. }
  1269. if (_paths[i].latencySamples.count() >= ZT_QOS_SHORTTERM_SAMPLE_WIN_MIN_REQ_SIZE) {
  1270. _paths[i].latencyVariance = _paths[i].latencySamples.stddev();
  1271. }
  1272. // Write values to external path object so that it can be propagated to the user
  1273. _paths[i].p->_latencyMean = _paths[i].latency;
  1274. _paths[i].p->_latencyVariance = _paths[i].latencyVariance;
  1275. _paths[i].p->_packetLossRatio = _paths[i].packetLossRatio;
  1276. _paths[i].p->_packetErrorRatio = _paths[i].packetErrorRatio;
  1277. _paths[i].p->_bonded = _paths[i].bonded;
  1278. _paths[i].p->_eligible = _paths[i].eligible;
  1279. //_paths[i].packetErrorRatio = 1.0 - (_paths[i].packetValiditySamples.count() ? _paths[i].packetValiditySamples.mean() : 1.0);
  1280. // _valid is written elsewhere
  1281. _paths[i].p->_relativeQuality = _paths[i].relativeQuality;
  1282. _paths[i].p->_localPort = _paths[i].localPort;
  1283. }
  1284. // Flag links for avoidance
  1285. for (unsigned int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1286. if (! _paths[i].p || ! _paths[i].allowed()) {
  1287. continue;
  1288. }
  1289. bool shouldAvoid = false;
  1290. if (! _paths[i].shouldAvoid) {
  1291. if (_paths[i].latency > _qw[ZT_QOS_LAT_MAX_IDX]) {
  1292. log("avoiding link %s because (lat %6.4f > %6.4f)", pathToStr(_paths[i].p).c_str(), _paths[i].latency, _qw[ZT_QOS_LAT_MAX_IDX]);
  1293. shouldAvoid = true;
  1294. }
  1295. if (_paths[i].latencyVariance > _qw[ZT_QOS_PDV_MAX_IDX]) {
  1296. log("avoiding link %s because (pdv %6.4f > %6.4f)", pathToStr(_paths[i].p).c_str(), _paths[i].latencyVariance, _qw[ZT_QOS_PDV_MAX_IDX]);
  1297. shouldAvoid = true;
  1298. }
  1299. if (_paths[i].packetErrorRatio > _qw[ZT_QOS_PER_MAX_IDX]) {
  1300. log("avoiding link %s because (per %6.4f > %6.4f)", pathToStr(_paths[i].p).c_str(), _paths[i].packetErrorRatio, _qw[ZT_QOS_PER_MAX_IDX]);
  1301. shouldAvoid = true;
  1302. }
  1303. if (_paths[i].packetLossRatio > _qw[ZT_QOS_PLR_MAX_IDX]) {
  1304. log("avoiding link %s because (plr %6.4f > %6.4f)", pathToStr(_paths[i].p).c_str(), _paths[i].packetLossRatio, _qw[ZT_QOS_PLR_MAX_IDX]);
  1305. shouldAvoid = true;
  1306. }
  1307. _paths[i].shouldAvoid = shouldAvoid;
  1308. }
  1309. else {
  1310. if (! shouldAvoid) {
  1311. log("no longer avoiding link %s", pathToStr(_paths[i].p).c_str());
  1312. _paths[i].shouldAvoid = false;
  1313. }
  1314. }
  1315. }
  1316. }
  1317. void Bond::processBalanceTasks(int64_t now)
  1318. {
  1319. if (! _numBondedPaths) {
  1320. return;
  1321. }
  1322. /**
  1323. * Clean up and reset flows if necessary
  1324. */
  1325. if ((now - _lastFlowExpirationCheck) > ZT_PEER_PATH_EXPIRATION) {
  1326. Mutex::Lock _l(_flows_m);
  1327. forgetFlowsWhenNecessary(ZT_PEER_PATH_EXPIRATION, false, now);
  1328. std::map<int16_t, SharedPtr<Flow> >::iterator it = _flows.begin();
  1329. while (it != _flows.end()) {
  1330. it->second->resetByteCounts();
  1331. ++it;
  1332. }
  1333. _lastFlowExpirationCheck = now;
  1334. }
  1335. /**
  1336. * Move (all) flows from dead paths
  1337. */
  1338. if (_policy == ZT_BOND_POLICY_BALANCE_XOR || _policy == ZT_BOND_POLICY_BALANCE_AWARE) {
  1339. Mutex::Lock _l(_flows_m);
  1340. std::map<int16_t, SharedPtr<Flow> >::iterator flow_it = _flows.begin();
  1341. while (flow_it != _flows.end()) {
  1342. if (_paths[flow_it->second->assignedPath].p) {
  1343. int originalPathIdx = flow_it->second->assignedPath;
  1344. if (! _paths[originalPathIdx].eligible) {
  1345. log("moving all flows from dead link %s", pathToStr(_paths[originalPathIdx].p).c_str());
  1346. if (assignFlowToBondedPath(flow_it->second, now, true)) {
  1347. _paths[originalPathIdx].assignedFlowCount--;
  1348. }
  1349. }
  1350. }
  1351. ++flow_it;
  1352. }
  1353. }
  1354. /**
  1355. * Move (some) flows from low quality paths
  1356. */
  1357. if (_policy == ZT_BOND_POLICY_BALANCE_AWARE) {
  1358. Mutex::Lock _l(_flows_m);
  1359. std::map<int16_t, SharedPtr<Flow> >::iterator flow_it = _flows.begin();
  1360. while (flow_it != _flows.end()) {
  1361. if (_paths[flow_it->second->assignedPath].p) {
  1362. int originalPathIdx = flow_it->second->assignedPath;
  1363. if (_paths[originalPathIdx].shouldAvoid) {
  1364. if (assignFlowToBondedPath(flow_it->second, now, true)) {
  1365. _paths[originalPathIdx].assignedFlowCount--;
  1366. return; // Only move one flow at a time
  1367. }
  1368. }
  1369. }
  1370. ++flow_it;
  1371. }
  1372. }
  1373. }
  1374. void Bond::dequeueNextActiveBackupPath(uint64_t now)
  1375. {
  1376. if (_abFailoverQueue.empty()) {
  1377. return;
  1378. }
  1379. _abPathIdx = _abFailoverQueue.front();
  1380. _abFailoverQueue.pop_front();
  1381. _lastActiveBackupPathChange = now;
  1382. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1383. if (_paths[i].p) {
  1384. _paths[i].resetPacketCounts();
  1385. }
  1386. }
  1387. }
  1388. bool Bond::abForciblyRotateLink()
  1389. {
  1390. Mutex::Lock _l(_paths_m);
  1391. if (_policy == ZT_BOND_POLICY_ACTIVE_BACKUP) {
  1392. int prevPathIdx = _abPathIdx;
  1393. dequeueNextActiveBackupPath(RR->node->now());
  1394. log("active link rotated from %s to %s", pathToStr(_paths[prevPathIdx].p).c_str(), pathToStr(_paths[_abPathIdx].p).c_str());
  1395. return true;
  1396. }
  1397. return false;
  1398. }
  1399. void Bond::processActiveBackupTasks(void* tPtr, int64_t now)
  1400. {
  1401. int prevActiveBackupPathIdx = _abPathIdx;
  1402. int nonPreferredPathIdx = ZT_MAX_PEER_NETWORK_PATHS;
  1403. bool bFoundPrimaryLink = false;
  1404. if (_abPathIdx != ZT_MAX_PEER_NETWORK_PATHS && ! _paths[_abPathIdx].p) {
  1405. _abPathIdx = ZT_MAX_PEER_NETWORK_PATHS;
  1406. log("main active-backup path has been removed");
  1407. }
  1408. /**
  1409. * Generate periodic status report
  1410. */
  1411. if ((now - _lastBondStatusLog) > ZT_BOND_STATUS_INTERVAL) {
  1412. _lastBondStatusLog = now;
  1413. if (_abPathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  1414. log("no active link");
  1415. }
  1416. else if (_paths[_abPathIdx].p) {
  1417. log("active link is %s, failover queue size is %zu", pathToStr(_paths[_abPathIdx].p).c_str(), _abFailoverQueue.size());
  1418. }
  1419. if (_abFailoverQueue.empty()) {
  1420. log("failover queue is empty, bond is no longer fault-tolerant");
  1421. }
  1422. }
  1423. /**
  1424. * Select initial "active" active-backup link
  1425. */
  1426. if (_abPathIdx == ZT_MAX_PEER_NETWORK_PATHS) {
  1427. /**
  1428. * [Automatic mode]
  1429. * The user has not explicitly specified links or their failover schedule,
  1430. * the bonding policy will now select the first eligible path and set it as
  1431. * its active backup path, if a substantially better path is detected the bonding
  1432. * policy will assign it as the new active backup path. If the path fails it will
  1433. * simply find the next eligible path.
  1434. */
  1435. if (! userHasSpecifiedLinks()) {
  1436. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1437. if (_paths[i].p && _paths[i].eligible) {
  1438. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1439. if (link) {
  1440. log("found eligible link %s", pathToStr(_paths[i].p).c_str());
  1441. _abPathIdx = i;
  1442. break;
  1443. }
  1444. }
  1445. }
  1446. }
  1447. /**
  1448. * [Manual mode]
  1449. * The user has specified links or failover rules that the bonding policy should adhere to.
  1450. */
  1451. else if (userHasSpecifiedLinks()) {
  1452. if (userHasSpecifiedPrimaryLink()) {
  1453. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1454. if (! _paths[i].p) {
  1455. continue;
  1456. }
  1457. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1458. if (link) {
  1459. if (_paths[i].eligible && link->primary()) {
  1460. if (! _paths[i].preferred()) {
  1461. // Found path on primary link, take note in case we don't find a preferred path
  1462. nonPreferredPathIdx = i;
  1463. bFoundPrimaryLink = true;
  1464. }
  1465. if (_paths[i].preferred()) {
  1466. _abPathIdx = i;
  1467. bFoundPrimaryLink = true;
  1468. if (_paths[_abPathIdx].p) {
  1469. SharedPtr<Link> abLink = RR->bc->getLinkBySocket(_policyAlias, _paths[_abPathIdx].p->localSocket());
  1470. if (abLink) {
  1471. log("found preferred primary link %s", pathToStr(_paths[_abPathIdx].p).c_str());
  1472. }
  1473. break; // Found preferred path on primary link
  1474. }
  1475. }
  1476. }
  1477. }
  1478. }
  1479. if (bFoundPrimaryLink && (nonPreferredPathIdx != ZT_MAX_PEER_NETWORK_PATHS)) {
  1480. log("found non-preferred primary link");
  1481. _abPathIdx = nonPreferredPathIdx;
  1482. }
  1483. }
  1484. else if (! userHasSpecifiedPrimaryLink()) {
  1485. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1486. if (_paths[i].p && _paths[i].eligible) {
  1487. _abPathIdx = i;
  1488. break;
  1489. }
  1490. }
  1491. if (_abPathIdx != ZT_MAX_PEER_NETWORK_PATHS) {
  1492. if (_paths[_abPathIdx].p) {
  1493. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[_abPathIdx].p->localSocket());
  1494. if (link) {
  1495. log("select non-primary link %s", pathToStr(_paths[_abPathIdx].p).c_str());
  1496. }
  1497. }
  1498. }
  1499. }
  1500. }
  1501. }
  1502. // Short-circuit if we don't have an active link yet. Everything below is optimization from the base case
  1503. if (_abPathIdx < 0 || _abPathIdx == ZT_MAX_PEER_NETWORK_PATHS || (! _paths[_abPathIdx].p)) {
  1504. return;
  1505. }
  1506. // Remove ineligible paths from the failover link queue
  1507. for (std::deque<int>::iterator it(_abFailoverQueue.begin()); it != _abFailoverQueue.end();) {
  1508. if (! _paths[(*it)].p) {
  1509. log("link is no longer valid, removing from failover queue (%zu links remain in queue)", _abFailoverQueue.size());
  1510. it = _abFailoverQueue.erase(it);
  1511. continue;
  1512. }
  1513. if (_paths[(*it)].p && ! _paths[(*it)].eligible) {
  1514. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[(*it)].p->localSocket());
  1515. it = _abFailoverQueue.erase(it);
  1516. if (link) {
  1517. log("link %s is ineligible, removing from failover queue (%zu links remain in queue)", pathToStr(_paths[_abPathIdx].p).c_str(), _abFailoverQueue.size());
  1518. }
  1519. continue;
  1520. }
  1521. else {
  1522. ++it;
  1523. }
  1524. }
  1525. /**
  1526. * Failover instructions were provided by user, build queue according those as well as IPv
  1527. * preference, disregarding performance.
  1528. */
  1529. if (userHasSpecifiedFailoverInstructions()) {
  1530. /**
  1531. * Clear failover scores
  1532. */
  1533. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1534. if (_paths[i].p) {
  1535. _paths[i].failoverScore = 0;
  1536. }
  1537. }
  1538. // Follow user-specified failover instructions
  1539. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1540. if (! _paths[i].p || ! _paths[i].allowed() || ! _paths[i].eligible) {
  1541. continue;
  1542. }
  1543. SharedPtr<Link> link = RR->bc->getLinkBySocket(_policyAlias, _paths[i].p->localSocket());
  1544. if (! link) {
  1545. continue;
  1546. }
  1547. int failoverScoreHandicap = _paths[i].failoverScore;
  1548. if (_paths[i].preferred()) {
  1549. failoverScoreHandicap += ZT_BOND_FAILOVER_HANDICAP_PREFERRED;
  1550. }
  1551. if (link->primary()) {
  1552. // If using "optimize" primary re-select mode, ignore user link designations
  1553. failoverScoreHandicap += ZT_BOND_FAILOVER_HANDICAP_PRIMARY;
  1554. }
  1555. if (! _paths[i].failoverScore) {
  1556. // If we didn't inherit a failover score from a "parent" that wants to use this path as a failover
  1557. int newHandicap = failoverScoreHandicap ? failoverScoreHandicap : (_paths[i].relativeQuality * 255.0);
  1558. _paths[i].failoverScore = newHandicap;
  1559. }
  1560. SharedPtr<Link> failoverLink;
  1561. if (link->failoverToLink().length()) {
  1562. failoverLink = RR->bc->getLinkByName(_policyAlias, link->failoverToLink());
  1563. }
  1564. if (failoverLink) {
  1565. for (int j = 0; j < ZT_MAX_PEER_NETWORK_PATHS; j++) {
  1566. if (_paths[j].p && getLink(_paths[j].p) == failoverLink.ptr()) {
  1567. int inheritedHandicap = failoverScoreHandicap - 10;
  1568. int newHandicap = _paths[j].failoverScore > inheritedHandicap ? _paths[j].failoverScore : inheritedHandicap;
  1569. if (! _paths[j].preferred()) {
  1570. newHandicap--;
  1571. }
  1572. _paths[j].failoverScore = newHandicap;
  1573. }
  1574. }
  1575. }
  1576. if (_paths[i].p) {
  1577. if (_paths[i].p.ptr() != _paths[_abPathIdx].p.ptr()) {
  1578. bool bFoundPathInQueue = false;
  1579. for (std::deque<int>::iterator it(_abFailoverQueue.begin()); it != _abFailoverQueue.end(); ++it) {
  1580. if (_paths[(*it)].p && (_paths[i].p.ptr() == _paths[(*it)].p.ptr())) {
  1581. bFoundPathInQueue = true;
  1582. }
  1583. }
  1584. if (! bFoundPathInQueue) {
  1585. _abFailoverQueue.push_front(i);
  1586. log("add link %s to failover queue (%zu links in queue)", pathToStr(_paths[i].p).c_str(), _abFailoverQueue.size());
  1587. addPathToBond(0, i);
  1588. }
  1589. }
  1590. }
  1591. }
  1592. }
  1593. /**
  1594. * No failover instructions provided by user, build queue according to performance
  1595. * and IPv preference.
  1596. */
  1597. else if (! userHasSpecifiedFailoverInstructions()) {
  1598. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1599. if (! _paths[i].p || ! _paths[i].allowed() || ! _paths[i].eligible) {
  1600. continue;
  1601. }
  1602. int failoverScoreHandicap = 0;
  1603. if (_paths[i].preferred()) {
  1604. failoverScoreHandicap = ZT_BOND_FAILOVER_HANDICAP_PREFERRED;
  1605. }
  1606. if (! _paths[i].eligible) {
  1607. failoverScoreHandicap = -10000;
  1608. }
  1609. SharedPtr<Link> link = getLink(_paths[i].p);
  1610. if (! link) {
  1611. continue;
  1612. }
  1613. if (link->primary() && _abLinkSelectMethod != ZT_BOND_RESELECTION_POLICY_OPTIMIZE) {
  1614. // If using "optimize" primary re-select mode, ignore user link designations
  1615. failoverScoreHandicap = ZT_BOND_FAILOVER_HANDICAP_PRIMARY;
  1616. }
  1617. /*
  1618. if (_paths[i].p.ptr() == _paths[_negotiatedPathIdx].p.ptr()) {
  1619. _paths[i].negotiated = true;
  1620. failoverScoreHandicap = ZT_BOND_FAILOVER_HANDICAP_NEGOTIATED;
  1621. }
  1622. else {
  1623. _paths[i].negotiated = false;
  1624. }
  1625. */
  1626. _paths[i].failoverScore = _paths[i].relativeQuality + failoverScoreHandicap;
  1627. if (_paths[i].p.ptr() != _paths[_abPathIdx].p.ptr()) {
  1628. bool bFoundPathInQueue = false;
  1629. for (std::deque<int>::iterator it(_abFailoverQueue.begin()); it != _abFailoverQueue.end(); ++it) {
  1630. if (_paths[i].p.ptr() == _paths[(*it)].p.ptr()) {
  1631. bFoundPathInQueue = true;
  1632. }
  1633. }
  1634. if (! bFoundPathInQueue) {
  1635. _abFailoverQueue.push_front(i);
  1636. log("add link %s to failover queue (%zu links in queue)", pathToStr(_paths[i].p).c_str(), _abFailoverQueue.size());
  1637. addPathToBond(0, i);
  1638. }
  1639. }
  1640. }
  1641. }
  1642. // Sort queue based on performance
  1643. if (! _abFailoverQueue.empty()) {
  1644. for (int i = 0; i < _abFailoverQueue.size(); i++) {
  1645. int value_to_insert = _abFailoverQueue[i];
  1646. int hole_position = i;
  1647. while (hole_position > 0 && (_abFailoverQueue[hole_position - 1] > value_to_insert)) {
  1648. _abFailoverQueue[hole_position] = _abFailoverQueue[hole_position - 1];
  1649. hole_position = hole_position - 1;
  1650. }
  1651. _abFailoverQueue[hole_position] = value_to_insert;
  1652. }
  1653. }
  1654. /**
  1655. * Short-circuit if we have no queued paths
  1656. */
  1657. if (_abFailoverQueue.empty()) {
  1658. return;
  1659. }
  1660. /**
  1661. * Fulfill primary re-select obligations
  1662. */
  1663. if (! _paths[_abPathIdx].eligible) { // Implicit ZT_BOND_RESELECTION_POLICY_FAILURE
  1664. log("link %s has failed, select link from failover queue (%zu links in queue)", pathToStr(_paths[_abPathIdx].p).c_str(), _abFailoverQueue.size());
  1665. if (! _abFailoverQueue.empty()) {
  1666. dequeueNextActiveBackupPath(now);
  1667. log("active link switched to %s", pathToStr(_paths[_abPathIdx].p).c_str());
  1668. }
  1669. else {
  1670. log("failover queue is empty, no links to choose from");
  1671. }
  1672. }
  1673. /**
  1674. * Detect change to prevent flopping during later optimization step.
  1675. */
  1676. if (prevActiveBackupPathIdx != _abPathIdx) {
  1677. _lastActiveBackupPathChange = now;
  1678. }
  1679. if (_abFailoverQueue.empty()) {
  1680. return; // No sense in continuing since there are no links to switch to
  1681. }
  1682. if (_abLinkSelectMethod == ZT_BOND_RESELECTION_POLICY_ALWAYS) {
  1683. SharedPtr<Link> abLink = getLink(_paths[_abPathIdx].p);
  1684. if (! _paths[_abFailoverQueue.front()].p) {
  1685. log("invalid link. not switching");
  1686. return;
  1687. }
  1688. SharedPtr<Link> abFailoverLink = getLink(_paths[_abFailoverQueue.front()].p);
  1689. if (abLink && ! abLink->primary() && _paths[_abFailoverQueue.front()].p && abFailoverLink && abFailoverLink->primary()) {
  1690. dequeueNextActiveBackupPath(now);
  1691. log("switch back to available primary link %s (select mode: always)", pathToStr(_paths[_abPathIdx].p).c_str());
  1692. }
  1693. }
  1694. if (_abLinkSelectMethod == ZT_BOND_RESELECTION_POLICY_BETTER) {
  1695. SharedPtr<Link> abLink = getLink(_paths[_abPathIdx].p);
  1696. if (abLink && ! abLink->primary()) {
  1697. // Active backup has switched to "better" primary link according to re-select policy.
  1698. SharedPtr<Link> abFailoverLink = getLink(_paths[_abFailoverQueue.front()].p);
  1699. if (_paths[_abFailoverQueue.front()].p && abFailoverLink && abFailoverLink->primary() && (_paths[_abFailoverQueue.front()].failoverScore > _paths[_abPathIdx].failoverScore)) {
  1700. dequeueNextActiveBackupPath(now);
  1701. log("switch back to user-defined primary link %s (select mode: better)", pathToStr(_paths[_abPathIdx].p).c_str());
  1702. }
  1703. }
  1704. }
  1705. if (_abLinkSelectMethod == ZT_BOND_RESELECTION_POLICY_OPTIMIZE && ! _abFailoverQueue.empty()) {
  1706. /**
  1707. * Implement link negotiation that was previously-decided
  1708. */
  1709. if (_paths[_abFailoverQueue.front()].negotiated) {
  1710. dequeueNextActiveBackupPath(now);
  1711. _lastPathNegotiationCheck = now;
  1712. log("switch negotiated link %s (select mode: optimize)", pathToStr(_paths[_abPathIdx].p).c_str());
  1713. }
  1714. else {
  1715. // Try to find a better path and automatically switch to it -- not too often, though.
  1716. if ((now - _lastActiveBackupPathChange) > ZT_BOND_OPTIMIZE_INTERVAL) {
  1717. if (! _abFailoverQueue.empty()) {
  1718. int newFScore = _paths[_abFailoverQueue.front()].failoverScore;
  1719. int prevFScore = _paths[_abPathIdx].failoverScore;
  1720. // Establish a minimum switch threshold to prevent flapping
  1721. int failoverScoreDifference = _paths[_abFailoverQueue.front()].failoverScore - _paths[_abPathIdx].failoverScore;
  1722. int thresholdQuantity = (int)(ZT_BOND_ACTIVE_BACKUP_OPTIMIZE_MIN_THRESHOLD * (float)_paths[_abPathIdx].relativeQuality);
  1723. if ((failoverScoreDifference > 0) && (failoverScoreDifference > thresholdQuantity)) {
  1724. SharedPtr<Path> oldPath = _paths[_abPathIdx].p;
  1725. dequeueNextActiveBackupPath(now);
  1726. log("switch from %s (score: %d) to better link %s (score: %d) (select mode: optimize)", pathToStr(oldPath).c_str(), prevFScore, pathToStr(_paths[_abPathIdx].p).c_str(), newFScore);
  1727. }
  1728. }
  1729. }
  1730. }
  1731. }
  1732. }
  1733. void Bond::initTimers()
  1734. {
  1735. _lastFlowExpirationCheck = 0;
  1736. _lastFlowRebalance = 0;
  1737. _lastSentPathNegotiationRequest = 0;
  1738. _lastPathNegotiationCheck = 0;
  1739. _lastPathNegotiationReceived = 0;
  1740. _lastQoSRateCheck = 0;
  1741. _lastAckRateCheck = 0;
  1742. _lastQualityEstimation = 0;
  1743. _lastBondStatusLog = 0;
  1744. _lastSummaryDump = 0;
  1745. _lastActiveBackupPathChange = 0;
  1746. _lastFrame = 0;
  1747. _lastBackgroundTaskCheck = 0;
  1748. }
  1749. void Bond::setBondParameters(int policy, SharedPtr<Bond> templateBond, bool useTemplate)
  1750. {
  1751. // Sanity check for policy
  1752. _defaultPolicy = (_defaultPolicy <= ZT_BOND_POLICY_NONE || _defaultPolicy > ZT_BOND_POLICY_BALANCE_AWARE) ? ZT_BOND_POLICY_NONE : _defaultPolicy;
  1753. _policy = (policy <= ZT_BOND_POLICY_NONE || policy > ZT_BOND_POLICY_BALANCE_AWARE) ? _defaultPolicy : policy;
  1754. // Check if non-leaf to prevent spamming infrastructure
  1755. ZT_PeerRole role;
  1756. if (_peer) {
  1757. role = RR->topology->role(_peer->address());
  1758. }
  1759. _isLeaf = _peer ? (role != ZT_PEER_ROLE_PLANET && role != ZT_PEER_ROLE_MOON) : false;
  1760. // Path negotiation
  1761. _allowPathNegotiation = false;
  1762. _pathNegotiationCutoffCount = 0;
  1763. _localUtility = 0;
  1764. _negotiatedPathIdx = 0;
  1765. // User preferences which may override the default bonding algorithm's behavior
  1766. _userHasSpecifiedPrimaryLink = false;
  1767. _userHasSpecifiedFailoverInstructions = false;
  1768. _userHasSpecifiedLinkCapacities = 0;
  1769. // Bond status
  1770. _numAliveLinks = 0;
  1771. _numTotalLinks = 0;
  1772. _numBondedPaths = 0;
  1773. // General parameters
  1774. _downDelay = 0;
  1775. _upDelay = 0;
  1776. _monitorInterval = 0;
  1777. // balance-aware
  1778. _totalBondUnderload = 0;
  1779. _overheadBytes = 0;
  1780. /**
  1781. * Policy defaults
  1782. */
  1783. _abPathIdx = ZT_MAX_PEER_NETWORK_PATHS;
  1784. _abLinkSelectMethod = ZT_BOND_RESELECTION_POLICY_OPTIMIZE;
  1785. _rrPacketsSentOnCurrLink = 0;
  1786. _rrIdx = 0;
  1787. _packetsPerLink = 64;
  1788. // Sane quality defaults
  1789. _qw[ZT_QOS_LAT_MAX_IDX] = 500.0f;
  1790. _qw[ZT_QOS_PDV_MAX_IDX] = 100.0f;
  1791. _qw[ZT_QOS_PLR_MAX_IDX] = 0.001f;
  1792. _qw[ZT_QOS_PER_MAX_IDX] = 0.0001f;
  1793. _qw[ZT_QOS_LAT_WEIGHT_IDX] = 0.25f;
  1794. _qw[ZT_QOS_PDV_WEIGHT_IDX] = 0.25f;
  1795. _qw[ZT_QOS_PLR_WEIGHT_IDX] = 0.25f;
  1796. _qw[ZT_QOS_PER_WEIGHT_IDX] = 0.25f;
  1797. _failoverInterval = ZT_BOND_FAILOVER_DEFAULT_INTERVAL;
  1798. /* If a user has specified custom parameters for this bonding policy, overlay them onto the defaults */
  1799. if (useTemplate) {
  1800. _policyAlias = templateBond->_policyAlias;
  1801. _policy = templateBond->policy();
  1802. _failoverInterval = templateBond->_failoverInterval >= ZT_BOND_FAILOVER_MIN_INTERVAL ? templateBond->_failoverInterval : ZT_BOND_FAILOVER_MIN_INTERVAL;
  1803. _downDelay = templateBond->_downDelay;
  1804. _upDelay = templateBond->_upDelay;
  1805. _abLinkSelectMethod = templateBond->_abLinkSelectMethod;
  1806. memcpy(_qw, templateBond->_qw, ZT_QOS_PARAMETER_SIZE * sizeof(float));
  1807. debug("user link quality spec = {%6.3f, %6.3f, %6.3f, %6.3f, %6.3f, %6.3f, %6.3f, %6.3f}", _qw[0], _qw[1], _qw[2], _qw[3], _qw[4], _qw[5], _qw[6], _qw[7]);
  1808. }
  1809. if (! _isLeaf) {
  1810. _policy = ZT_BOND_POLICY_NONE;
  1811. }
  1812. // Timer geometry
  1813. _monitorInterval = _failoverInterval / ZT_BOND_ECHOS_PER_FAILOVER_INTERVAL;
  1814. _qualityEstimationInterval = _failoverInterval * 2;
  1815. _qosSendInterval = _failoverInterval * 2;
  1816. _ackSendInterval = _failoverInterval * 2;
  1817. _qosCutoffCount = 0;
  1818. _ackCutoffCount = 0;
  1819. _defaultPathRefractoryPeriod = 8000;
  1820. }
  1821. void Bond::setUserLinkQualitySpec(float weights[], int len)
  1822. {
  1823. if (len != ZT_QOS_PARAMETER_SIZE) {
  1824. debug("link quality spec has an invalid number of parameters (%d out of %d), ignoring", len, ZT_QOS_PARAMETER_SIZE);
  1825. return;
  1826. }
  1827. float weightTotal = 0.0;
  1828. for (unsigned int i = 4; i < ZT_QOS_PARAMETER_SIZE; ++i) {
  1829. weightTotal += weights[i];
  1830. }
  1831. if (weightTotal > 0.99 && weightTotal < 1.01) {
  1832. memcpy(_qw, weights, len * sizeof(float));
  1833. }
  1834. }
  1835. SharedPtr<Link> Bond::getLink(const SharedPtr<Path>& path)
  1836. {
  1837. return ! path ? SharedPtr<Link>() : RR->bc->getLinkBySocket(_policyAlias, path->localSocket());
  1838. }
  1839. std::string Bond::pathToStr(const SharedPtr<Path>& path)
  1840. {
  1841. #ifdef ZT_TRACE
  1842. if (path) {
  1843. char pathStr[64] = { 0 };
  1844. char fullPathStr[384] = { 0 };
  1845. path->address().toString(pathStr);
  1846. SharedPtr<Link> link = getLink(path);
  1847. if (link) {
  1848. std::string ifnameStr = std::string(link->ifname());
  1849. snprintf(fullPathStr, 384, "%.16" PRIx64 "-%s/%s", path->localSocket(), ifnameStr.c_str(), pathStr);
  1850. return std::string(fullPathStr);
  1851. }
  1852. }
  1853. return "";
  1854. #else
  1855. return "";
  1856. #endif
  1857. }
  1858. void Bond::dumpPathStatus(int64_t now, int pathIdx)
  1859. {
  1860. #ifdef ZT_TRACE
  1861. std::string aliveOrDead = _paths[pathIdx].alive ? std::string("alive") : std::string("dead");
  1862. std::string eligibleOrNot = _paths[pathIdx].eligible ? std::string("eligible") : std::string("ineligible");
  1863. std::string bondedOrNot = _paths[pathIdx].bonded ? std::string("bonded") : std::string("unbonded");
  1864. log("path[%2u] --- %5s (in %7" PRId64 ", out: %7" PRId64 "), %10s, %8s, flows=%-6u lat=%-8.3f pdv=%-7.3f err=%-6.4f loss=%-6.4f qual=%-6.4f --- (%s) spare=%d",
  1865. pathIdx,
  1866. aliveOrDead.c_str(),
  1867. _paths[pathIdx].p->age(now),
  1868. _paths[pathIdx].p->_lastOut == 0 ? static_cast<int64_t>(0) : now - _paths[pathIdx].p->_lastOut,
  1869. eligibleOrNot.c_str(),
  1870. bondedOrNot.c_str(),
  1871. _paths[pathIdx].assignedFlowCount,
  1872. _paths[pathIdx].latency,
  1873. _paths[pathIdx].latencyVariance,
  1874. _paths[pathIdx].packetErrorRatio,
  1875. _paths[pathIdx].packetLossRatio,
  1876. _paths[pathIdx].relativeQuality,
  1877. pathToStr(_paths[pathIdx].p).c_str(),
  1878. _paths[pathIdx].isSpare());
  1879. #endif
  1880. }
  1881. void Bond::dumpInfo(int64_t now, bool force)
  1882. {
  1883. #ifdef ZT_TRACE
  1884. uint64_t timeSinceLastDump = now - _lastSummaryDump;
  1885. if (! force && timeSinceLastDump < ZT_BOND_STATUS_INTERVAL) {
  1886. return;
  1887. }
  1888. _lastSummaryDump = now;
  1889. float overhead = (_overheadBytes / (timeSinceLastDump / 1000.0f) / 1000.0f);
  1890. _overheadBytes = 0;
  1891. log("bond: bp=%d, fi=%" PRIu64 ", mi=%d, ud=%d, dd=%d, flows=%zu, leaf=%d, overhead=%f KB/s, links=(%d/%d)",
  1892. _policy,
  1893. _failoverInterval,
  1894. _monitorInterval,
  1895. _upDelay,
  1896. _downDelay,
  1897. _flows.size(),
  1898. _isLeaf,
  1899. overhead,
  1900. _numAliveLinks,
  1901. _numTotalLinks);
  1902. for (int i = 0; i < ZT_MAX_PEER_NETWORK_PATHS; ++i) {
  1903. if (_paths[i].p) {
  1904. dumpPathStatus(now, i);
  1905. }
  1906. }
  1907. log("");
  1908. #endif
  1909. }
  1910. } // namespace ZeroTier