CentralDB.cpp 67 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897
  1. /*
  2. * Copyright (c)2019 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 "CentralDB.hpp"
  14. #ifdef ZT_CONTROLLER_USE_LIBPQ
  15. #include "../../node/Constants.hpp"
  16. #include "../../node/SHA512.hpp"
  17. #include "../../version.h"
  18. #include "BigTableStatusWriter.hpp"
  19. #include "ControllerChangeNotifier.hpp"
  20. #include "ControllerConfig.hpp"
  21. #include "CtlUtil.hpp"
  22. #include "EmbeddedNetworkController.hpp"
  23. #include "OtelCarrier.hpp"
  24. #include "PostgresStatusWriter.hpp"
  25. #include "PubSubListener.hpp"
  26. #include "PubSubWriter.hpp"
  27. #include "Redis.hpp"
  28. #include "RedisListener.hpp"
  29. #include "RedisStatusWriter.hpp"
  30. #include "opentelemetry/context/propagation/global_propagator.h"
  31. #include "opentelemetry/trace/provider.h"
  32. #include <chrono>
  33. #include <climits>
  34. #include <iomanip>
  35. #include <libpq-fe.h>
  36. #include <optional>
  37. #include <pqxx/pqxx>
  38. #include <rustybits.h>
  39. #include <sstream>
  40. // #define REDIS_TRACE 1
  41. using json = nlohmann::json;
  42. using namespace ZeroTier;
  43. using Attrs = std::vector<std::pair<std::string, std::string> >;
  44. using Item = std::pair<std::string, Attrs>;
  45. using ItemStream = std::vector<Item>;
  46. CentralDB::CentralDB(
  47. const Identity& myId,
  48. const char* connString,
  49. int listenPort,
  50. CentralDB::ListenerMode listenMode,
  51. CentralDB::StatusWriterMode statusMode,
  52. const ControllerConfig* cc)
  53. : DB()
  54. , _listenerMode(listenMode)
  55. , _statusWriterMode(statusMode)
  56. , _cc(cc)
  57. , _pool()
  58. , _myId(myId)
  59. , _myAddress(myId.address())
  60. , _ready(0)
  61. , _connected(1)
  62. , _run(1)
  63. , _waitNoticePrinted(false)
  64. , _listenPort(listenPort)
  65. , _redis(NULL)
  66. , _cluster(NULL)
  67. , _redisMemberStatus(false)
  68. , _smee(NULL)
  69. {
  70. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  71. auto tracer = provider->GetTracer("CentralDB");
  72. auto span = tracer->StartSpan("CentralDB::CentralDB");
  73. auto scope = tracer->WithActiveSpan(span);
  74. rustybits::init_async_runtime();
  75. char myAddress[64];
  76. _myAddressStr = myId.address().toString(myAddress);
  77. _connString = std::string(connString);
  78. auto f = std::make_shared<PostgresConnFactory>(_connString);
  79. _pool =
  80. std::make_shared<ConnectionPool<PostgresConnection> >(15, 5, std::static_pointer_cast<ConnectionFactory>(f));
  81. memset(_ssoPsk, 0, sizeof(_ssoPsk));
  82. char* const ssoPskHex = getenv("ZT_SSO_PSK");
  83. #ifdef ZT_TRACE
  84. fprintf(stderr, "ZT_SSO_PSK: %s\n", ssoPskHex);
  85. #endif
  86. if (ssoPskHex) {
  87. // SECURITY: note that ssoPskHex will always be null-terminated if libc actually
  88. // returns something non-NULL. If the hex encodes something shorter than 48 bytes,
  89. // it will be padded at the end with zeroes. If longer, it'll be truncated.
  90. Utils::unhex(ssoPskHex, _ssoPsk, sizeof(_ssoPsk));
  91. }
  92. const char* redisMemberStatus = getenv("ZT_REDIS_MEMBER_STATUS");
  93. if (redisMemberStatus && (strcmp(redisMemberStatus, "true") == 0)) {
  94. _redisMemberStatus = true;
  95. fprintf(stderr, "Using redis for member status\n");
  96. }
  97. if ((listenMode == LISTENER_MODE_REDIS || statusMode == STATUS_WRITER_MODE_REDIS) && _cc->redisConfig != NULL) {
  98. auto innerspan = tracer->StartSpan("CentralDB::CentralDB::configureRedis");
  99. auto innerscope = tracer->WithActiveSpan(innerspan);
  100. sw::redis::ConnectionOptions opts;
  101. sw::redis::ConnectionPoolOptions poolOpts;
  102. opts.host = _cc->redisConfig->hostname;
  103. opts.port = _cc->redisConfig->port;
  104. opts.password = _cc->redisConfig->password;
  105. opts.db = 0;
  106. opts.keep_alive = true;
  107. opts.connect_timeout = std::chrono::seconds(3);
  108. poolOpts.size = 25;
  109. poolOpts.wait_timeout = std::chrono::seconds(5);
  110. poolOpts.connection_lifetime = std::chrono::minutes(3);
  111. poolOpts.connection_idle_time = std::chrono::minutes(1);
  112. if (_cc->redisConfig->clusterMode) {
  113. innerspan->SetAttribute("cluster_mode", "true");
  114. fprintf(stderr, "Using Redis in Cluster Mode\n");
  115. _cluster = std::make_shared<sw::redis::RedisCluster>(opts, poolOpts);
  116. }
  117. else {
  118. innerspan->SetAttribute("cluster_mode", "false");
  119. fprintf(stderr, "Using Redis in Standalone Mode\n");
  120. _redis = std::make_shared<sw::redis::Redis>(opts, poolOpts);
  121. }
  122. }
  123. _readyLock.lock();
  124. fprintf(
  125. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL waiting for initial data download..." ZT_EOL_S,
  126. ::_timestr(), (unsigned long long)_myAddress.toInt());
  127. _waitNoticePrinted = true;
  128. initializeNetworks();
  129. initializeMembers();
  130. _heartbeatThread = std::thread(&CentralDB::heartbeat, this);
  131. switch (listenMode) {
  132. case LISTENER_MODE_REDIS:
  133. fprintf(stderr, "Using Redis for change listeners\n");
  134. if (_cc->redisConfig != NULL) {
  135. if (_cc->redisConfig->clusterMode) {
  136. _membersDbWatcher = std::make_shared<RedisMemberListener>(_myAddressStr, _cluster, this);
  137. _networksDbWatcher = std::make_shared<RedisNetworkListener>(_myAddressStr, _cluster, this);
  138. }
  139. else {
  140. _membersDbWatcher = std::make_shared<RedisMemberListener>(_myAddressStr, _redis, this);
  141. _networksDbWatcher = std::make_shared<RedisNetworkListener>(_myAddressStr, _redis, this);
  142. }
  143. }
  144. else {
  145. throw std::runtime_error("CentralDB: Redis listener mode selected but no Redis configuration provided");
  146. }
  147. case LISTENER_MODE_PUBSUB:
  148. fprintf(stderr, "Using PubSub for change listeners\n");
  149. if (cc->pubSubConfig != NULL) {
  150. _membersDbWatcher = std::make_shared<PubSubMemberListener>(
  151. _myAddressStr, cc->pubSubConfig->project_id, cc->pubSubConfig->member_change_recv_topic, this);
  152. _networksDbWatcher = std::make_shared<PubSubNetworkListener>(
  153. _myAddressStr, cc->pubSubConfig->project_id, cc->pubSubConfig->network_change_recv_topic, this);
  154. _changeNotifier = std::make_shared<PubSubChangeNotifier>(
  155. _myAddressStr, cc->pubSubConfig->project_id, cc->pubSubConfig->member_change_send_topic,
  156. cc->pubSubConfig->network_change_send_topic);
  157. }
  158. else {
  159. throw std::runtime_error(
  160. "CentralDB: PubSub listener mode selected but no PubSub configuration provided");
  161. }
  162. break;
  163. case LISTENER_MODE_PGSQL:
  164. default:
  165. fprintf(stderr, "Using PostgreSQL for change listeners\n");
  166. _membersDbWatcher = std::make_shared<PostgresMemberListener>(this, _pool, "member_" + _myAddressStr, 5);
  167. _networksDbWatcher = std::make_shared<PostgresNetworkListener>(this, _pool, "network_" + _myAddressStr, 5);
  168. break;
  169. }
  170. std::shared_ptr<PubSubWriter> pubsubWriter;
  171. switch (statusMode) {
  172. case STATUS_WRITER_MODE_REDIS:
  173. fprintf(stderr, "Using Redis for status writer\n");
  174. if (_cc->redisConfig != NULL) {
  175. if (_cc->redisConfig->clusterMode) {
  176. _statusWriter = std::make_shared<RedisStatusWriter>(_cluster, _myAddressStr);
  177. }
  178. else {
  179. _statusWriter = std::make_shared<RedisStatusWriter>(_redis, _myAddressStr);
  180. }
  181. }
  182. else {
  183. throw std::runtime_error("CentralDB: Redis status mode selected but no Redis configuration provided");
  184. }
  185. break;
  186. case STATUS_WRITER_MODE_BIGTABLE:
  187. fprintf(stderr, "Using BigTable for status writer\n");
  188. if (cc->bigTableConfig == NULL) {
  189. throw std::runtime_error(
  190. "CentralDB: BigTable status mode selected but no BigTable configuration provided");
  191. }
  192. if (cc->pubSubConfig == NULL) {
  193. throw std::runtime_error(
  194. "CentralDB: BigTable status mode selected but no PubSub configuration provided");
  195. }
  196. _statusWriter = std::make_shared<BigTableStatusWriter>(
  197. cc->bigTableConfig->project_id, cc->bigTableConfig->instance_id, cc->bigTableConfig->table_id);
  198. break;
  199. case STATUS_WRITER_MODE_PGSQL:
  200. default:
  201. fprintf(stderr, "Using PostgreSQL for status writer\n");
  202. _statusWriter = std::make_shared<PostgresStatusWriter>(_pool);
  203. break;
  204. }
  205. // start background threads
  206. for (int i = 0; i < ZT_CENTRAL_CONTROLLER_COMMIT_THREADS; ++i) {
  207. _commitThread[i] = std::thread(&CentralDB::commitThread, this);
  208. }
  209. _onlineNotificationThread = std::thread(&CentralDB::onlineNotificationThread, this);
  210. configureSmee();
  211. }
  212. CentralDB::~CentralDB()
  213. {
  214. if (_smee != NULL) {
  215. rustybits::smee_client_delete(_smee);
  216. _smee = NULL;
  217. }
  218. rustybits::shutdown_async_runtime();
  219. _run = 0;
  220. std::this_thread::sleep_for(std::chrono::milliseconds(100));
  221. _heartbeatThread.join();
  222. _commitQueue.stop();
  223. for (int i = 0; i < ZT_CENTRAL_CONTROLLER_COMMIT_THREADS; ++i) {
  224. _commitThread[i].join();
  225. }
  226. _onlineNotificationThread.join();
  227. }
  228. void CentralDB::configureSmee()
  229. {
  230. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  231. auto tracer = provider->GetTracer("CentralDB");
  232. auto span = tracer->StartSpan("CentralDB::configureSmee");
  233. auto scope = tracer->WithActiveSpan(span);
  234. const char* TEMPORAL_SCHEME = "ZT_TEMPORAL_SCHEME";
  235. const char* TEMPORAL_HOST = "ZT_TEMPORAL_HOST";
  236. const char* TEMPORAL_PORT = "ZT_TEMPORAL_PORT";
  237. const char* TEMPORAL_NAMESPACE = "ZT_TEMPORAL_NAMESPACE";
  238. const char* SMEE_TASK_QUEUE = "ZT_SMEE_TASK_QUEUE";
  239. const char* scheme = getenv(TEMPORAL_SCHEME);
  240. if (scheme == NULL) {
  241. scheme = "http";
  242. }
  243. const char* host = getenv(TEMPORAL_HOST);
  244. const char* port = getenv(TEMPORAL_PORT);
  245. const char* ns = getenv(TEMPORAL_NAMESPACE);
  246. const char* task_queue = getenv(SMEE_TASK_QUEUE);
  247. if (scheme != NULL && host != NULL && port != NULL && ns != NULL && task_queue != NULL) {
  248. fprintf(stderr, "creating smee client\n");
  249. std::string hostPort =
  250. std::string(scheme) + std::string("://") + std::string(host) + std::string(":") + std::string(port);
  251. this->_smee = rustybits::smee_client_new(hostPort.c_str(), ns, task_queue);
  252. }
  253. else {
  254. fprintf(stderr, "Smee client not configured\n");
  255. }
  256. }
  257. bool CentralDB::waitForReady()
  258. {
  259. while (_ready < 2) {
  260. _readyLock.lock();
  261. _readyLock.unlock();
  262. }
  263. return true;
  264. }
  265. bool CentralDB::isReady()
  266. {
  267. return ((_ready == 2) && (_connected));
  268. }
  269. bool CentralDB::save(nlohmann::json& record, bool notifyListeners)
  270. {
  271. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  272. auto tracer = provider->GetTracer("CentralDB");
  273. auto span = tracer->StartSpan("CentralDB::save");
  274. auto scope = tracer->WithActiveSpan(span);
  275. bool modified = false;
  276. try {
  277. if (! record.is_object()) {
  278. fprintf(stderr, "record is not an object?!?\n");
  279. return false;
  280. }
  281. const std::string objtype = record["objtype"];
  282. if (objtype == "network") {
  283. auto span = tracer->StartSpan("CentralDB::save::network");
  284. auto scope = tracer->WithActiveSpan(span);
  285. fprintf(stderr, "CentralDB network save %s\n", record["id"].get<std::string>().c_str());
  286. const uint64_t nwid = OSUtils::jsonIntHex(record["id"], 0ULL);
  287. if (nwid) {
  288. nlohmann::json old;
  289. get(nwid, old);
  290. if ((! old.is_object()) || (! _compareRecords(old, record))) {
  291. fprintf(stderr, "posting network change to commit queue\n");
  292. record["revision"] = OSUtils::jsonInt(record["revision"], 0ULL) + 1ULL;
  293. _queueItem qi;
  294. qi.jsonData = record;
  295. qi.notifyListeners = notifyListeners;
  296. OtelCarrier<std::map<std::string, std::string> > carrier(qi.traceContext);
  297. auto current_ctx = opentelemetry::context::RuntimeContext::GetCurrent();
  298. auto propagator =
  299. opentelemetry::context::propagation::GlobalTextMapPropagator::GetGlobalPropagator();
  300. propagator->Inject(carrier, current_ctx);
  301. _commitQueue.post(qi);
  302. modified = true;
  303. }
  304. }
  305. }
  306. else if (objtype == "member") {
  307. auto span = tracer->StartSpan("CentralDB::save::member");
  308. auto scope = tracer->WithActiveSpan(span);
  309. std::string networkId = record["nwid"];
  310. std::string memberId = record["id"];
  311. const uint64_t nwid = OSUtils::jsonIntHex(record["nwid"], 0ULL);
  312. const uint64_t id = OSUtils::jsonIntHex(record["id"], 0ULL);
  313. fprintf(stderr, "member save %s-%s\n", networkId.c_str(), memberId.c_str());
  314. if ((id) && (nwid)) {
  315. nlohmann::json network, old;
  316. get(nwid, network, id, old);
  317. if ((! old.is_object()) || (! _compareRecords(old, record))) {
  318. fprintf(stderr, "posting member change to commit queue\n");
  319. record["revision"] = OSUtils::jsonInt(record["revision"], 0ULL) + 1ULL;
  320. _queueItem qi;
  321. qi.jsonData = record;
  322. qi.notifyListeners = notifyListeners;
  323. OtelCarrier<std::map<std::string, std::string> > carrier(qi.traceContext);
  324. auto current_ctx = opentelemetry::context::RuntimeContext::GetCurrent();
  325. auto propagator =
  326. opentelemetry::context::propagation::GlobalTextMapPropagator::GetGlobalPropagator();
  327. propagator->Inject(carrier, current_ctx);
  328. _commitQueue.post(qi);
  329. modified = true;
  330. }
  331. else {
  332. // fprintf(stderr, "no change\n");
  333. }
  334. }
  335. }
  336. else {
  337. fprintf(stderr, "uhh waaat\n");
  338. }
  339. }
  340. catch (std::exception& e) {
  341. fprintf(stderr, "Error on CentralDB::save: %s\n", e.what());
  342. }
  343. catch (...) {
  344. fprintf(stderr, "Unknown error on CentralDB::save\n");
  345. }
  346. return modified;
  347. }
  348. void CentralDB::eraseNetwork(const uint64_t networkId)
  349. {
  350. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  351. auto tracer = provider->GetTracer("CentralDB");
  352. auto span = tracer->StartSpan("CentralDB::eraseNetwork");
  353. auto scope = tracer->WithActiveSpan(span);
  354. char networkIdStr[17];
  355. span->SetAttribute("network_id", Utils::hex(networkId, networkIdStr));
  356. fprintf(stderr, "CentralDB::eraseNetwork\n");
  357. char tmp2[24];
  358. waitForReady();
  359. Utils::hex(networkId, tmp2);
  360. _queueItem qi;
  361. qi.jsonData["id"] = tmp2;
  362. qi.jsonData["objtype"] = "_delete_network";
  363. qi.notifyListeners = true;
  364. OtelCarrier<std::map<std::string, std::string> > carrier(qi.traceContext);
  365. auto current_ctx = opentelemetry::context::RuntimeContext::GetCurrent();
  366. auto propagator = opentelemetry::context::propagation::GlobalTextMapPropagator::GetGlobalPropagator();
  367. propagator->Inject(carrier, current_ctx);
  368. _commitQueue.post(qi);
  369. nlohmann::json nullJson;
  370. _networkChanged(qi.jsonData, nullJson, true);
  371. }
  372. void CentralDB::eraseMember(const uint64_t networkId, const uint64_t memberId)
  373. {
  374. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  375. auto tracer = provider->GetTracer("CentralDB");
  376. auto span = tracer->StartSpan("CentralDB::eraseMember");
  377. auto scope = tracer->WithActiveSpan(span);
  378. char networkIdStr[17];
  379. char memberIdStr[11];
  380. span->SetAttribute("network_id", Utils::hex(networkId, networkIdStr));
  381. span->SetAttribute("member_id", Utils::hex10(memberId, memberIdStr));
  382. fprintf(stderr, "CentralDB::eraseMember\n");
  383. char tmp2[24];
  384. waitForReady();
  385. _queueItem qi;
  386. qi.jsonData["nwid"] = tmp2;
  387. qi.jsonData["id"] = tmp2;
  388. qi.jsonData["objtype"] = "_delete_member";
  389. qi.notifyListeners = true;
  390. OtelCarrier<std::map<std::string, std::string> > carrier(qi.traceContext);
  391. auto current_ctx = opentelemetry::context::RuntimeContext::GetCurrent();
  392. auto propagator = opentelemetry::context::propagation::GlobalTextMapPropagator::GetGlobalPropagator();
  393. propagator->Inject(carrier, current_ctx);
  394. _commitQueue.post(qi);
  395. nlohmann::json nullJson;
  396. _memberChanged(qi.jsonData, nullJson, true);
  397. }
  398. void CentralDB::nodeIsOnline(
  399. const uint64_t networkId,
  400. const uint64_t memberId,
  401. const InetAddress& physicalAddress,
  402. const char* osArch)
  403. {
  404. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  405. auto tracer = provider->GetTracer("CentralDB");
  406. auto span = tracer->StartSpan("CentralDB::nodeIsOnline");
  407. auto scope = tracer->WithActiveSpan(span);
  408. char networkIdStr[17];
  409. char memberIdStr[11];
  410. char ipStr[INET6_ADDRSTRLEN];
  411. span->SetAttribute("network_id", Utils::hex(networkId, networkIdStr));
  412. span->SetAttribute("member_id", Utils::hex10(memberId, memberIdStr));
  413. span->SetAttribute("physical_address", physicalAddress.toString(ipStr));
  414. span->SetAttribute("os_arch", osArch);
  415. std::lock_guard<std::mutex> l(_lastOnline_l);
  416. NodeOnlineRecord& i = _lastOnline[std::pair<uint64_t, uint64_t>(networkId, memberId)];
  417. i.lastSeen = OSUtils::now();
  418. if (physicalAddress) {
  419. i.physicalAddress = physicalAddress;
  420. }
  421. i.osArch = std::string(osArch);
  422. }
  423. void CentralDB::nodeIsOnline(const uint64_t networkId, const uint64_t memberId, const InetAddress& physicalAddress)
  424. {
  425. this->nodeIsOnline(networkId, memberId, physicalAddress, "unknown/unknown");
  426. }
  427. AuthInfo CentralDB::getSSOAuthInfo(const nlohmann::json& member, const std::string& redirectURL)
  428. {
  429. if (_cc->ssoEnabled) {
  430. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  431. auto tracer = provider->GetTracer("CentralDB");
  432. auto span = tracer->StartSpan("CentralDB::getSSOAuthInfo");
  433. auto scope = tracer->WithActiveSpan(span);
  434. Metrics::db_get_sso_info++;
  435. // NONCE is just a random character string. no semantic meaning
  436. // state = HMAC SHA384 of Nonce based on shared sso key
  437. //
  438. // need nonce timeout in database? make sure it's used within X time
  439. // X is 5 minutes for now. Make configurable later?
  440. //
  441. // how do we tell when a nonce is used? if auth_expiration_time is set
  442. std::string networkId = member["nwid"];
  443. std::string memberId = member["id"];
  444. char authenticationURL[4096] = { 0 };
  445. AuthInfo info;
  446. info.enabled = true;
  447. // if (memberId == "a10dccea52" && networkId == "8056c2e21c24673d") {
  448. // fprintf(stderr, "invalid authinfo for grant's machine\n");
  449. // info.version=1;
  450. // return info;
  451. // }
  452. // fprintf(stderr, "CentralDB::updateMemberOnLoad: %s-%s\n", networkId.c_str(), memberId.c_str());
  453. std::shared_ptr<PostgresConnection> c;
  454. try {
  455. c = _pool->borrow();
  456. pqxx::work w(*c->c);
  457. char nonceBytes[16] = { 0 };
  458. std::string nonce = "";
  459. // check if the member exists first.
  460. pqxx::row count =
  461. w.exec(
  462. "SELECT count(id) FROM ztc_member WHERE id = $1 AND network_id = $2 AND deleted = false",
  463. pqxx::params { memberId, networkId })
  464. .one_row();
  465. if (count[0].as<int>() == 1) {
  466. // get active nonce, if exists.
  467. pqxx::result r = w.exec(
  468. "SELECT nonce FROM ztc_sso_expiry "
  469. "WHERE network_id = $1 AND member_id = $2 "
  470. "AND ((NOW() AT TIME ZONE 'UTC') <= authentication_expiry_time) AND ((NOW() AT TIME ZONE 'UTC') <= "
  471. "nonce_expiration)",
  472. pqxx::params { networkId, memberId });
  473. if (r.size() == 0) {
  474. // no active nonce.
  475. // find an unused nonce, if one exists.
  476. pqxx::result r = w.exec(
  477. "SELECT nonce FROM ztc_sso_expiry "
  478. "WHERE network_id = $1 AND member_id = $2 "
  479. "AND authentication_expiry_time IS NULL AND ((NOW() AT TIME ZONE 'UTC') <= nonce_expiration)",
  480. pqxx::params { networkId, memberId });
  481. if (r.size() == 1) {
  482. // we have an existing nonce. Use it
  483. nonce = r.at(0)[0].as<std::string>();
  484. Utils::unhex(nonce.c_str(), nonceBytes, sizeof(nonceBytes));
  485. }
  486. else if (r.empty()) {
  487. // create a nonce
  488. Utils::getSecureRandom(nonceBytes, 16);
  489. char nonceBuf[64] = { 0 };
  490. Utils::hex(nonceBytes, sizeof(nonceBytes), nonceBuf);
  491. nonce = std::string(nonceBuf);
  492. pqxx::result ir = w.exec(
  493. "INSERT INTO ztc_sso_expiry "
  494. "(nonce, nonce_expiration, network_id, member_id) VALUES "
  495. "($1, TO_TIMESTAMP($2::double precision/1000), $3, $4)",
  496. pqxx::params { nonce, OSUtils::now() + 300000, networkId, memberId });
  497. w.commit();
  498. }
  499. else {
  500. // > 1 ?!? Thats an error!
  501. fprintf(stderr, "> 1 unused nonce!\n");
  502. exit(6);
  503. }
  504. }
  505. else if (r.size() == 1) {
  506. nonce = r.at(0)[0].as<std::string>();
  507. Utils::unhex(nonce.c_str(), nonceBytes, sizeof(nonceBytes));
  508. }
  509. else {
  510. // more than 1 nonce in use? Uhhh...
  511. fprintf(stderr, "> 1 nonce in use for network member?!?\n");
  512. exit(7);
  513. }
  514. r = w.exec(
  515. "SELECT oc.client_id, oc.authorization_endpoint, oc.issuer, oc.provider, oc.sso_impl_version "
  516. "FROM ztc_network AS n "
  517. "INNER JOIN ztc_org o "
  518. " ON o.owner_id = n.owner_id "
  519. "LEFT OUTER JOIN ztc_network_oidc_config noc "
  520. " ON noc.network_id = n.id "
  521. "LEFT OUTER JOIN ztc_oidc_config oc "
  522. " ON noc.client_id = oc.client_id AND oc.org_id = o.org_id "
  523. "WHERE n.id = $1 AND n.sso_enabled = true",
  524. pqxx::params { networkId });
  525. std::string client_id = "";
  526. std::string authorization_endpoint = "";
  527. std::string issuer = "";
  528. std::string provider = "";
  529. uint64_t sso_version = 0;
  530. if (r.size() == 1) {
  531. client_id = r.at(0)[0].as<std::optional<std::string> >().value_or("");
  532. authorization_endpoint = r.at(0)[1].as<std::optional<std::string> >().value_or("");
  533. issuer = r.at(0)[2].as<std::optional<std::string> >().value_or("");
  534. provider = r.at(0)[3].as<std::optional<std::string> >().value_or("");
  535. sso_version = r.at(0)[4].as<std::optional<uint64_t> >().value_or(1);
  536. }
  537. else if (r.size() > 1) {
  538. fprintf(
  539. stderr, "ERROR: More than one auth endpoint for an organization?!?!? NetworkID: %s\n",
  540. networkId.c_str());
  541. }
  542. else {
  543. fprintf(stderr, "No client or auth endpoint?!?\n");
  544. }
  545. info.version = sso_version;
  546. // no catch all else because we don't actually care if no records exist here. just continue as normal.
  547. if ((! client_id.empty()) && (! authorization_endpoint.empty())) {
  548. uint8_t state[48];
  549. HMACSHA384(_ssoPsk, nonceBytes, sizeof(nonceBytes), state);
  550. char state_hex[256];
  551. Utils::hex(state, 48, state_hex);
  552. if (info.version == 0) {
  553. char url[2048] = { 0 };
  554. OSUtils::ztsnprintf(
  555. url, sizeof(authenticationURL),
  556. "%s?response_type=id_token&response_mode=form_post&scope=openid+email+profile&redirect_uri="
  557. "%s&nonce=%s&state=%s&client_id=%s",
  558. authorization_endpoint.c_str(), url_encode(redirectURL).c_str(), nonce.c_str(), state_hex,
  559. client_id.c_str());
  560. info.authenticationURL = std::string(url);
  561. }
  562. else if (info.version == 1) {
  563. info.ssoClientID = client_id;
  564. info.issuerURL = issuer;
  565. info.ssoProvider = provider;
  566. info.ssoNonce = nonce;
  567. info.ssoState = std::string(state_hex) + "_" + networkId;
  568. info.centralAuthURL = redirectURL;
  569. #ifdef ZT_DEBUG
  570. fprintf(
  571. stderr,
  572. "ssoClientID: %s\nissuerURL: %s\nssoNonce: %s\nssoState: %s\ncentralAuthURL: %s\nprovider: "
  573. "%s\n",
  574. info.ssoClientID.c_str(), info.issuerURL.c_str(), info.ssoNonce.c_str(),
  575. info.ssoState.c_str(), info.centralAuthURL.c_str(), provider.c_str());
  576. #endif
  577. }
  578. }
  579. else {
  580. fprintf(
  581. stderr, "client_id: %s\nauthorization_endpoint: %s\n", client_id.c_str(),
  582. authorization_endpoint.c_str());
  583. }
  584. }
  585. _pool->unborrow(c);
  586. }
  587. catch (std::exception& e) {
  588. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  589. fprintf(stderr, "ERROR: Error updating member on load for network %s: %s\n", networkId.c_str(), e.what());
  590. }
  591. return info; // std::string(authenticationURL);
  592. }
  593. return AuthInfo();
  594. }
  595. void CentralDB::initializeNetworks()
  596. {
  597. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  598. auto tracer = provider->GetTracer("CentralDB");
  599. auto span = tracer->StartSpan("CentralDB::initializeNetworks");
  600. auto scope = tracer->WithActiveSpan(span);
  601. fprintf(stderr, "Initializing networks...\n");
  602. try {
  603. char qbuf[2048];
  604. sprintf(
  605. qbuf,
  606. "SELECT id, name, configuration , (EXTRACT(EPOCH FROM creation_time AT TIME ZONE 'UTC')*1000)::bigint, "
  607. "(EXTRACT(EPOCH FROM last_modified AT TIME ZONE 'UTC')*1000)::bigint, revision, frontend "
  608. "FROM networks_ctl WHERE controller_id = '%s'",
  609. _myAddressStr.c_str());
  610. auto c = _pool->borrow();
  611. pqxx::work w(*c->c);
  612. fprintf(stderr, "Load networks from psql...\n");
  613. auto stream = pqxx::stream_from::query(w, qbuf);
  614. std::tuple<
  615. std::string // network ID
  616. ,
  617. std::optional<std::string> // name
  618. ,
  619. std::string // configuration
  620. ,
  621. std::optional<uint64_t> // creation_time
  622. ,
  623. std::optional<uint64_t> // last_modified
  624. ,
  625. std::optional<uint64_t> // revision
  626. ,
  627. std::string // frontend
  628. >
  629. row;
  630. uint64_t count = 0;
  631. uint64_t total = 0;
  632. while (stream >> row) {
  633. auto start = std::chrono::high_resolution_clock::now();
  634. json empty;
  635. json config;
  636. initNetwork(config);
  637. std::string nwid = std::get<0>(row);
  638. std::string name = std::get<1>(row).value_or("");
  639. json cfgtmp = json::parse(std::get<2>(row));
  640. std::optional<uint64_t> created_at = std::get<3>(row);
  641. std::optional<uint64_t> last_modified = std::get<4>(row);
  642. std::optional<uint64_t> revision = std::get<5>(row);
  643. std::string frontend = std::get<6>(row);
  644. config["id"] = nwid;
  645. config["name"] = name;
  646. config["creationTime"] = created_at.value_or(0);
  647. config["lastModified"] = last_modified.value_or(0);
  648. config["revision"] = revision.value_or(0);
  649. config["capabilities"] = cfgtmp["capabilities"].is_array() ? cfgtmp["capabilities"] : json::array();
  650. config["enableBroadcast"] =
  651. cfgtmp["enableBroadcast"].is_boolean() ? cfgtmp["enableBroadcast"].get<bool>() : false;
  652. config["mtu"] = cfgtmp["mtu"].is_number() ? cfgtmp["mtu"].get<int32_t>() : 2800;
  653. config["multicastLimit"] =
  654. cfgtmp["multicastLimit"].is_number() ? cfgtmp["multicastLimit"].get<int32_t>() : 64;
  655. config["private"] = cfgtmp["private"].is_boolean() ? cfgtmp["private"].get<bool>() : true;
  656. config["remoteTraceLevel"] =
  657. cfgtmp["remoteTraceLevel"].is_number() ? cfgtmp["remoteTraceLevel"].get<int32_t>() : 0;
  658. config["remoteTraceTarget"] =
  659. cfgtmp["remoteTraceTarget"].is_string() ? cfgtmp["remoteTraceTarget"].get<std::string>() : "";
  660. config["revision"] = revision.value_or(0);
  661. config["rules"] = cfgtmp["rules"].is_array() ? cfgtmp["rules"] : json::array();
  662. config["tags"] = cfgtmp["tags"].is_array() ? cfgtmp["tags"] : json::array();
  663. if (cfgtmp["v4AssignMode"].is_object()) {
  664. config["v4AssignMode"] = cfgtmp["v4AssignMode"];
  665. }
  666. else {
  667. config["v4AssignMode"] = json::object();
  668. config["v4AssignMode"]["zt"] = true;
  669. }
  670. if (cfgtmp["v6AssignMode"].is_object()) {
  671. config["v6AssignMode"] = cfgtmp["v6AssignMode"];
  672. }
  673. else {
  674. config["v6AssignMode"] = json::object();
  675. config["v6AssignMode"]["zt"] = true;
  676. config["v6AssignMode"]["6plane"] = true;
  677. config["v6AssignMode"]["rfc4193"] = false;
  678. }
  679. config["ssoEnabled"] = cfgtmp["ssoEnabled"].is_boolean() ? cfgtmp["ssoEnabled"].get<bool>() : false;
  680. if (config["ssoConfig"].is_object()) {
  681. config["ssoConfig"] = cfgtmp["ssoConfig"];
  682. }
  683. else {
  684. config["ssoConfig"] = empty;
  685. }
  686. config["objtype"] = "network";
  687. config["routes"] = cfgtmp["routes"].is_array() ? cfgtmp["routes"] : json::array();
  688. config["clientId"] = cfgtmp["clientId"].is_string() ? cfgtmp["clientId"].get<std::string>() : "";
  689. config["authorizationEndpoint"] =
  690. cfgtmp["authorizationEndpoint"].is_string() ? cfgtmp["authorizationEndpoint"].get<std::string>() : "";
  691. config["provider"] = cfgtmp["ssoProvider"].is_string() ? cfgtmp["ssoProvider"].get<std::string>() : "";
  692. if (! cfgtmp["dns"].is_object()) {
  693. cfgtmp["dns"] = json::object();
  694. cfgtmp["dns"]["domain"] = "";
  695. cfgtmp["dns"]["servers"] = json::array();
  696. }
  697. else {
  698. config["dns"] = cfgtmp["dns"];
  699. }
  700. config["ipAssignmentPools"] =
  701. cfgtmp["ipAssignmentPools"].is_array() ? cfgtmp["ipAssignmentPools"] : json::array();
  702. config["frontend"] = frontend;
  703. Metrics::network_count++;
  704. _networkChanged(empty, config, false);
  705. auto end = std::chrono::high_resolution_clock::now();
  706. auto dur = std::chrono::duration_cast<std::chrono::microseconds>(end - start);
  707. total += dur.count();
  708. ++count;
  709. if (count > 0 && count % 10000 == 0) {
  710. fprintf(stderr, "Averaging %lu us per network\n", (total / count));
  711. }
  712. }
  713. w.commit();
  714. _pool->unborrow(c);
  715. fprintf(stderr, "done.\n");
  716. if (++this->_ready == 2) {
  717. if (_waitNoticePrinted) {
  718. fprintf(
  719. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL data download complete." ZT_EOL_S, _timestr(),
  720. (unsigned long long)_myAddress.toInt());
  721. }
  722. _readyLock.unlock();
  723. }
  724. fprintf(stderr, "network init done\n");
  725. }
  726. catch (std::exception& e) {
  727. fprintf(stderr, "ERROR: Error initializing networks: %s\n", e.what());
  728. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  729. std::this_thread::sleep_for(std::chrono::milliseconds(5000));
  730. exit(-1);
  731. }
  732. }
  733. void CentralDB::initializeMembers()
  734. {
  735. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  736. auto tracer = provider->GetTracer("CentralDB");
  737. auto span = tracer->StartSpan("CentralDB::initializeMembers");
  738. auto scope = tracer->WithActiveSpan(span);
  739. std::string memberId;
  740. std::string networkId;
  741. try {
  742. std::unordered_map<std::string, std::string> networkMembers;
  743. fprintf(stderr, "Initializing Members...\n");
  744. std::string setKeyBase = "network-nodes-all:{" + _myAddressStr + "}:";
  745. if (_redisMemberStatus) {
  746. fprintf(stderr, "Initialize Redis for members...\n");
  747. std::unique_lock<std::shared_mutex> l(_networks_l);
  748. std::unordered_set<std::string> deletes;
  749. for (auto it : _networks) {
  750. uint64_t nwid_i = it.first;
  751. char nwidTmp[64] = { 0 };
  752. OSUtils::ztsnprintf(nwidTmp, sizeof(nwidTmp), "%.16llx", nwid_i);
  753. std::string nwid(nwidTmp);
  754. std::string key = setKeyBase + nwid;
  755. deletes.insert(key);
  756. }
  757. if (! deletes.empty()) {
  758. try {
  759. if (_cc->redisConfig->clusterMode) {
  760. auto tx = _cluster->transaction(_myAddressStr, true, false);
  761. for (std::string k : deletes) {
  762. tx.del(k);
  763. }
  764. tx.exec();
  765. }
  766. else {
  767. auto tx = _redis->transaction(true, false);
  768. for (std::string k : deletes) {
  769. tx.del(k);
  770. }
  771. tx.exec();
  772. }
  773. }
  774. catch (sw::redis::Error& e) {
  775. // ignore
  776. }
  777. }
  778. }
  779. char qbuf[2048];
  780. sprintf(
  781. qbuf,
  782. "SELECT nm.device_id, nm.network_id, nm.authorized, nm.active_bridge, nm.ip_assignments, "
  783. "nm.no_auto_assign_ips, "
  784. "nm.sso_exempt, (EXTRACT(EPOCH FROM nm.authentication_expiry_time AT TIME ZONE 'UTC')*1000)::bigint, "
  785. "(EXTRACT(EPOCH FROM nm.creation_time AT TIME ZONE 'UTC')*1000)::bigint, nm.identity, "
  786. "(EXTRACT(EPOCH FROM nm.last_authorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  787. "(EXTRACT(EPOCH FROM nm.last_deauthorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  788. "nm.remote_trace_level, nm.remote_trace_target, nm.revision, nm.capabilities, nm.tags, "
  789. "nm.frontend "
  790. "FROM network_memberships_ctl nm "
  791. "INNER JOIN networks_ctl n "
  792. " ON nm.network_id = n.id "
  793. "WHERE n.controller_id = '%s'",
  794. _myAddressStr.c_str());
  795. auto c = _pool->borrow();
  796. pqxx::work w(*c->c);
  797. fprintf(stderr, "Load members from psql...\n");
  798. auto stream = pqxx::stream_from::query(w, qbuf);
  799. std::tuple<
  800. std::string // device ID
  801. ,
  802. std::string // network ID
  803. ,
  804. bool // authorized
  805. ,
  806. std::optional<bool> // active_bridge
  807. ,
  808. std::optional<std::string> // ip_assignments
  809. ,
  810. std::optional<bool> // no_auto_assign_ips
  811. ,
  812. std::optional<bool> // sso_exempt
  813. ,
  814. std::optional<uint64_t> // authentication_expiry_time
  815. ,
  816. std::optional<uint64_t> // creation_time
  817. ,
  818. std::optional<std::string> // identity
  819. ,
  820. std::optional<uint64_t> // last_authorized_time
  821. ,
  822. std::optional<uint64_t> // last_deauthorized_time
  823. ,
  824. std::optional<int32_t> // remote_trace_level
  825. ,
  826. std::optional<std::string> // remote_trace_target
  827. ,
  828. std::optional<uint64_t> // revision
  829. ,
  830. std::optional<std::string> // capabilities
  831. ,
  832. std::optional<std::string> // tags
  833. ,
  834. std::string // frontend
  835. >
  836. row;
  837. auto tmp = std::chrono::high_resolution_clock::now();
  838. uint64_t count = 0;
  839. uint64_t total = 0;
  840. while (stream >> row) {
  841. auto start = std::chrono::high_resolution_clock::now();
  842. json empty;
  843. json config;
  844. initMember(config);
  845. memberId = std::get<0>(row);
  846. networkId = std::get<1>(row);
  847. bool authorized = std::get<2>(row);
  848. std::optional<bool> active_bridge = std::get<3>(row);
  849. std::string ip_assignments = std::get<4>(row).value_or("");
  850. std::optional<bool> no_auto_assign_ips = std::get<5>(row);
  851. std::optional<bool> sso_exempt = std::get<6>(row);
  852. std::optional<uint64_t> authentication_expiry_time = std::get<7>(row);
  853. std::optional<uint64_t> creation_time = std::get<8>(row);
  854. std::optional<std::string> identity = std::get<9>(row);
  855. std::optional<uint64_t> last_authorized_time = std::get<10>(row);
  856. std::optional<uint64_t> last_deauthorized_time = std::get<11>(row);
  857. std::optional<int32_t> remote_trace_level = std::get<12>(row);
  858. std::optional<std::string> remote_trace_target = std::get<13>(row);
  859. std::optional<uint64_t> revision = std::get<14>(row);
  860. std::optional<std::string> capabilities = std::get<15>(row);
  861. std::optional<std::string> tags = std::get<16>(row);
  862. networkMembers.insert(std::pair<std::string, std::string>(setKeyBase + networkId, memberId));
  863. config["objtype"] = "member";
  864. config["id"] = memberId;
  865. config["address"] = identity.value_or("");
  866. config["nwid"] = networkId;
  867. config["authorized"] = authorized;
  868. config["activeBridge"] = active_bridge.value_or(false);
  869. config["ipAssignments"] = json::array();
  870. if (ip_assignments != "{}") {
  871. std::string tmp = ip_assignments.substr(1, ip_assignments.length() - 2);
  872. std::vector<std::string> addrs = split(tmp, ',');
  873. for (auto it = addrs.begin(); it != addrs.end(); ++it) {
  874. config["ipAssignments"].push_back(*it);
  875. }
  876. }
  877. config["capabilities"] = json::parse(capabilities.value_or("[]"));
  878. config["creationTime"] = creation_time.value_or(0);
  879. config["lastAuthorizedTime"] = last_authorized_time.value_or(0);
  880. config["lastDeauthorizedTime"] = last_deauthorized_time.value_or(0);
  881. config["noAutoAssignIPs"] = no_auto_assign_ips.value_or(false);
  882. config["remoteTraceLevel"] = remote_trace_level.value_or(0);
  883. config["remoteTraceTarget"] = remote_trace_target.value_or(nullptr);
  884. config["revision"] = revision.value_or(0);
  885. config["ssoExempt"] = sso_exempt.value_or(false);
  886. config["authenticationExpiryTime"] = authentication_expiry_time.value_or(0);
  887. config["tags"] = json::parse(tags.value_or("[]"));
  888. config["frontend"] = std::get<17>(row);
  889. Metrics::member_count++;
  890. _memberChanged(empty, config, false);
  891. memberId = "";
  892. networkId = "";
  893. auto end = std::chrono::high_resolution_clock::now();
  894. auto dur = std::chrono::duration_cast<std::chrono::microseconds>(end - start);
  895. total += dur.count();
  896. ++count;
  897. if (count > 0 && count % 10000 == 0) {
  898. fprintf(stderr, "Averaging %llu us per member\n", (total / count));
  899. }
  900. }
  901. if (count > 0) {
  902. fprintf(stderr, "Took %llu us per member to load\n", (total / count));
  903. }
  904. stream.complete();
  905. w.commit();
  906. _pool->unborrow(c);
  907. fprintf(stderr, "done.\n");
  908. if (_listenerMode == LISTENER_MODE_REDIS)
  909. if (! networkMembers.empty()) {
  910. if (_redisMemberStatus) {
  911. fprintf(stderr, "Load member data into redis...\n");
  912. if (_cc->redisConfig->clusterMode) {
  913. auto tx = _cluster->transaction(_myAddressStr, true, false);
  914. uint64_t count = 0;
  915. for (auto it : networkMembers) {
  916. tx.sadd(it.first, it.second);
  917. if (++count % 30000 == 0) {
  918. tx.exec();
  919. tx = _cluster->transaction(_myAddressStr, true, false);
  920. }
  921. }
  922. tx.exec();
  923. }
  924. else {
  925. auto tx = _redis->transaction(true, false);
  926. uint64_t count = 0;
  927. for (auto it : networkMembers) {
  928. tx.sadd(it.first, it.second);
  929. if (++count % 30000 == 0) {
  930. tx.exec();
  931. tx = _redis->transaction(true, false);
  932. }
  933. }
  934. tx.exec();
  935. }
  936. fprintf(stderr, "done.\n");
  937. }
  938. }
  939. fprintf(stderr, "Done loading members...\n");
  940. if (++this->_ready == 2) {
  941. if (_waitNoticePrinted) {
  942. fprintf(
  943. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL data download complete." ZT_EOL_S, _timestr(),
  944. (unsigned long long)_myAddress.toInt());
  945. }
  946. _readyLock.unlock();
  947. }
  948. }
  949. catch (sw::redis::Error& e) {
  950. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  951. fprintf(stderr, "ERROR: Error initializing members (redis): %s\n", e.what());
  952. exit(-1);
  953. }
  954. catch (std::exception& e) {
  955. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  956. fprintf(stderr, "ERROR: Error initializing member: %s-%s %s\n", networkId.c_str(), memberId.c_str(), e.what());
  957. exit(-1);
  958. }
  959. }
  960. void CentralDB::heartbeat()
  961. {
  962. char publicId[1024];
  963. char hostnameTmp[1024];
  964. _myId.toString(false, publicId);
  965. if (gethostname(hostnameTmp, sizeof(hostnameTmp)) != 0) {
  966. hostnameTmp[0] = (char)0;
  967. }
  968. else {
  969. for (int i = 0; i < (int)sizeof(hostnameTmp); ++i) {
  970. if ((hostnameTmp[i] == '.') || (hostnameTmp[i] == 0)) {
  971. hostnameTmp[i] = (char)0;
  972. break;
  973. }
  974. }
  975. }
  976. const char* controllerId = _myAddressStr.c_str();
  977. const char* publicIdentity = publicId;
  978. const char* hostname = hostnameTmp;
  979. while (_run == 1) {
  980. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  981. auto tracer = provider->GetTracer("CentralDB");
  982. auto span = tracer->StartSpan("CentralDB::heartbeat");
  983. auto scope = tracer->WithActiveSpan(span);
  984. // fprintf(stderr, "%s: heartbeat\n", controllerId);
  985. auto c = _pool->borrow();
  986. int64_t ts = OSUtils::now();
  987. if (c->c) {
  988. std::string major = std::to_string(ZEROTIER_ONE_VERSION_MAJOR);
  989. std::string minor = std::to_string(ZEROTIER_ONE_VERSION_MINOR);
  990. std::string rev = std::to_string(ZEROTIER_ONE_VERSION_REVISION);
  991. std::string version = major + "." + minor + "." + rev;
  992. std::string versionStr = "v" + version;
  993. try {
  994. pqxx::work w { *c->c };
  995. w.exec(
  996. "INSERT INTO controllers_ctl (id, hostname, last_heartbeat, public_identity, version) VALUES "
  997. "($1, $2, TO_TIMESTAMP($3::double precision/1000), $4, $5) "
  998. "ON CONFLICT (id) DO UPDATE SET hostname = EXCLUDED.hostname, last_heartbeat = "
  999. "EXCLUDED.last_heartbeat, "
  1000. "public_identity = EXCLUDED.public_identity, version = EXCLUDED.version",
  1001. pqxx::params { controllerId, hostname, ts, publicIdentity, versionStr })
  1002. .no_rows();
  1003. w.commit();
  1004. }
  1005. catch (std::exception& e) {
  1006. fprintf(stderr, "%s: Heartbeat update failed: %s\n", controllerId, e.what());
  1007. span->End();
  1008. std::this_thread::sleep_for(std::chrono::milliseconds(1000));
  1009. continue;
  1010. }
  1011. }
  1012. _pool->unborrow(c);
  1013. try {
  1014. if (_listenerMode == LISTENER_MODE_REDIS && _redisMemberStatus) {
  1015. if (_cc->redisConfig->clusterMode) {
  1016. _cluster->zadd("controllers", "controllerId", ts);
  1017. }
  1018. else {
  1019. _redis->zadd("controllers", "controllerId", ts);
  1020. }
  1021. }
  1022. }
  1023. catch (sw::redis::Error& e) {
  1024. fprintf(stderr, "ERROR: Redis error in heartbeat thread: %s\n", e.what());
  1025. }
  1026. span->End();
  1027. std::this_thread::sleep_for(std::chrono::milliseconds(1000));
  1028. }
  1029. fprintf(stderr, "Exited heartbeat thread\n");
  1030. }
  1031. void CentralDB::commitThread()
  1032. {
  1033. fprintf(stderr, "%s: commitThread start\n", _myAddressStr.c_str());
  1034. _queueItem qitem;
  1035. while (_commitQueue.get(qitem) & (_run == 1)) {
  1036. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  1037. auto tracer = provider->GetTracer("CentralDB");
  1038. fprintf(stderr, "checking qitem trace context\n");
  1039. for (auto const& kv : qitem.traceContext) {
  1040. fprintf(stderr, "traceContext: %s: %s\n", kv.first.c_str(), kv.second.c_str());
  1041. }
  1042. auto propagator = opentelemetry::context::propagation::GlobalTextMapPropagator::GetGlobalPropagator();
  1043. OtelCarrier<std::map<std::string, std::string> > carrier(qitem.traceContext);
  1044. auto current_ctx = opentelemetry::context::RuntimeContext::GetCurrent();
  1045. auto new_context = propagator->Extract(carrier, current_ctx);
  1046. auto remote_span = opentelemetry::trace::GetSpan(new_context);
  1047. auto remote_scope = tracer->WithActiveSpan(remote_span);
  1048. {
  1049. auto span = tracer->StartSpan("CentralDB::commitThread");
  1050. auto scope = tracer->WithActiveSpan(span);
  1051. fprintf(stderr, "commitThread tick\n");
  1052. if (! qitem.jsonData.is_object()) {
  1053. fprintf(stderr, "not an object\n");
  1054. continue;
  1055. }
  1056. std::shared_ptr<PostgresConnection> c;
  1057. try {
  1058. c = _pool->borrow();
  1059. }
  1060. catch (std::exception& e) {
  1061. fprintf(stderr, "ERROR: %s\n", e.what());
  1062. continue;
  1063. }
  1064. if (! c) {
  1065. fprintf(stderr, "Error getting database connection\n");
  1066. continue;
  1067. }
  1068. Metrics::pgsql_commit_ticks++;
  1069. try {
  1070. nlohmann::json& config = (qitem.jsonData);
  1071. const std::string objtype = config["objtype"];
  1072. if (objtype == "member") {
  1073. auto mspan = tracer->StartSpan("CentralDB::commitThread::member");
  1074. auto mscope = tracer->WithActiveSpan(mspan);
  1075. // fprintf(stderr, "%s: commitThread: member\n", _myAddressStr.c_str());
  1076. std::string memberId;
  1077. std::string networkId;
  1078. try {
  1079. pqxx::work w(*c->c);
  1080. memberId = config["id"];
  1081. networkId = config["nwid"];
  1082. fprintf(stderr, "commit member %s-%s\n", networkId.c_str(), memberId.c_str());
  1083. std::string target = "NULL";
  1084. if (! config["remoteTraceTarget"].is_null()) {
  1085. target = config["remoteTraceTarget"];
  1086. }
  1087. // get network and the frontend it is assigned to
  1088. // if network does not exist, skip member update
  1089. pqxx::row nwrow =
  1090. w.exec(
  1091. "SELECT COUNT(id), frontend FROM networks_ctl WHERE id = $1 GROUP BY frontend",
  1092. pqxx::params { networkId })
  1093. .one_row();
  1094. int nwcount = nwrow[0].as<int>();
  1095. std::string frontend = nwrow[1].as<std::string>();
  1096. if (nwcount != 1) {
  1097. fprintf(stderr, "network %s does not exist. skipping member upsert\n", networkId.c_str());
  1098. w.abort();
  1099. _pool->unborrow(c);
  1100. continue;
  1101. }
  1102. pqxx::row mrow =
  1103. w.exec(
  1104. "SELECT COUNT(device_id) FROM network_memberships_ctl WHERE device_id = $1 "
  1105. "AND network_id = $2",
  1106. pqxx::params { memberId, networkId })
  1107. .one_row();
  1108. int membercount = mrow[0].as<int>();
  1109. bool isNewMember = (membercount == 0);
  1110. std::string change_source;
  1111. if (! config["change_source"].is_null()) {
  1112. change_source = config["change_source"];
  1113. }
  1114. else {
  1115. change_source = "controller";
  1116. }
  1117. if (! isNewMember && change_source != "controller" && frontend != change_source) {
  1118. fprintf(
  1119. stderr, "skipping member %s-%s update. change source: %s, frontend: %s\n",
  1120. networkId.c_str(), memberId.c_str(), change_source.c_str(), frontend.c_str());
  1121. // if it is not a new member and the change source is not the controller and doesn't match
  1122. // the frontend, don't apply the change.
  1123. continue;
  1124. }
  1125. std::vector<std::string> ipAssignments;
  1126. fprintf(
  1127. stderr, "Saving IP Assignments: \n\tipAssignments: %s\n",
  1128. OSUtils::jsonDump(config["ipAssignments"], -1).c_str());
  1129. if (config["ipAssignments"].is_array()) {
  1130. for (auto& ip : config["ipAssignments"]) {
  1131. if (ip.is_string()) {
  1132. ipAssignments.push_back(ip.get<std::string>());
  1133. }
  1134. }
  1135. }
  1136. fprintf(stderr, "member json: %s\n", config.dump().c_str());
  1137. int64_t vMajor = OSUtils::jsonUInt(config["vMajor"], 0);
  1138. int64_t vMinor = OSUtils::jsonUInt(config["vMinor"], 0);
  1139. int64_t vRev = OSUtils::jsonUInt(config["vRev"], 0);
  1140. int64_t vProto = OSUtils::jsonUInt(config["vProto"], 0);
  1141. if (vMajor < 0)
  1142. vMajor = 0;
  1143. if (vMinor < 0)
  1144. vMinor = 0;
  1145. if (vRev < 0)
  1146. vRev = 0;
  1147. if (vProto < 0)
  1148. vProto = 0;
  1149. pqxx::result res =
  1150. w.exec(
  1151. "INSERT INTO network_memberships_ctl (device_id, network_id, authorized, "
  1152. "active_bridge, "
  1153. "ip_assignments, "
  1154. "no_auto_assign_ips, sso_exempt, authentication_expiry_time, capabilities, "
  1155. "creation_time, "
  1156. "identity, last_authorized_time, last_deauthorized_time, "
  1157. "remote_trace_level, remote_trace_target, revision, tags, version_major, "
  1158. "version_minor, "
  1159. "version_revision, version_protocol) "
  1160. "VALUES ($1, $2, $3, $4, $5, $6, $7, TO_TIMESTAMP($8::double precision/1000), $9, "
  1161. "TO_TIMESTAMP($10::double precision/1000), $11, TO_TIMESTAMP($12::double "
  1162. "precision/1000), "
  1163. "TO_TIMESTAMP($13::double precision/1000), $14, $15, $16, $17, $18, $19, $20, $21) "
  1164. "ON CONFLICT (device_id, network_id) DO UPDATE SET "
  1165. "authorized = EXCLUDED.authorized, active_bridge = EXCLUDED.active_bridge, "
  1166. "ip_assignments = EXCLUDED.ip_assignments, no_auto_assign_ips = "
  1167. "EXCLUDED.no_auto_assign_ips, "
  1168. "sso_exempt = EXCLUDED.sso_exempt, authentication_expiry_time = "
  1169. "EXCLUDED.authentication_expiry_time, "
  1170. "capabilities = EXCLUDED.capabilities, creation_time = EXCLUDED.creation_time, "
  1171. "identity = EXCLUDED.identity, last_authorized_time = EXCLUDED.last_authorized_time, "
  1172. "last_deauthorized_time = EXCLUDED.last_deauthorized_time, "
  1173. "remote_trace_level = EXCLUDED.remote_trace_level, remote_trace_target = "
  1174. "EXCLUDED.remote_trace_target, "
  1175. "revision = EXCLUDED.revision, tags = EXCLUDED.tags, version_major = "
  1176. "EXCLUDED.version_major, "
  1177. "version_minor = EXCLUDED.version_minor, version_revision = "
  1178. "EXCLUDED.version_revision, "
  1179. "version_protocol = EXCLUDED.version_protocol",
  1180. pqxx::params { memberId,
  1181. networkId,
  1182. OSUtils::jsonBool(config["authorized"], false),
  1183. OSUtils::jsonBool(config["activeBridge"], false),
  1184. ipAssignments,
  1185. OSUtils::jsonBool(config["noAutoAssignIps"], false),
  1186. OSUtils::jsonBool(config["ssoExempt"], false),
  1187. OSUtils::jsonInt(config["authenticationExpiryTime"], 0),
  1188. OSUtils::jsonDump(config["capabilities"], -1),
  1189. OSUtils::jsonInt(config["creationTime"], OSUtils::now()),
  1190. OSUtils::jsonString(config["identity"], ""),
  1191. OSUtils::jsonInt(config["lastAuthorizedTime"], 0),
  1192. OSUtils::jsonInt(config["lastDeauthorizedTime"], 0),
  1193. OSUtils::jsonInt(config["remoteTraceLevel"], 0),
  1194. target,
  1195. OSUtils::jsonInt(config["revision"], 0),
  1196. OSUtils::jsonDump(config["tags"], -1),
  1197. vMajor,
  1198. vMinor,
  1199. vRev,
  1200. vProto })
  1201. .no_rows();
  1202. w.commit();
  1203. if (_listenerMode == LISTENER_MODE_PUBSUB) {
  1204. // Publish change to pubsub stream
  1205. if (config["change_source"].is_null() || config["change_source"] == "controller") {
  1206. nlohmann::json oldMember;
  1207. nlohmann::json newMember = config;
  1208. if (! isNewMember) {
  1209. oldMember = _getNetworkMember(w, networkId, memberId);
  1210. }
  1211. _changeNotifier->notifyMemberChange(oldMember, newMember, frontend);
  1212. }
  1213. }
  1214. if (_smee != NULL && isNewMember) {
  1215. // TODO: Smee Notifications for New Members
  1216. // pqxx::row row = w.exec_params1(
  1217. // "SELECT "
  1218. // " count(h.hook_id) "
  1219. // "FROM "
  1220. // " ztc_hook h "
  1221. // " INNER JOIN ztc_org o ON o.org_id = h.org_id "
  1222. // " INNER JOIN ztc_network n ON n.owner_id = o.owner_id "
  1223. // " WHERE "
  1224. // "n.id = $1 ",
  1225. // networkId);
  1226. // int64_t hookCount = row[0].as<int64_t>();
  1227. // if (hookCount > 0) {
  1228. // notifyNewMember(networkId, memberId);
  1229. // }
  1230. }
  1231. const uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  1232. const uint64_t memberidInt = OSUtils::jsonIntHex(config["id"], 0ULL);
  1233. if (nwidInt && memberidInt) {
  1234. nlohmann::json nwOrig;
  1235. nlohmann::json memOrig;
  1236. nlohmann::json memNew(config);
  1237. get(nwidInt, nwOrig, memberidInt, memOrig);
  1238. _memberChanged(memOrig, memNew, qitem.notifyListeners);
  1239. }
  1240. else {
  1241. fprintf(
  1242. stderr, "%s: Can't notify of change. Error parsing nwid or memberid: %llu-%llu\n",
  1243. _myAddressStr.c_str(), (unsigned long long)nwidInt, (unsigned long long)memberidInt);
  1244. }
  1245. }
  1246. catch (std::exception& e) {
  1247. fprintf(
  1248. stderr, "%s ERROR: Error updating member %s-%s: %s\n", _myAddressStr.c_str(),
  1249. networkId.c_str(), memberId.c_str(), e.what());
  1250. mspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1251. }
  1252. }
  1253. else if (objtype == "network") {
  1254. auto nspan = tracer->StartSpan("CentralDB::commitThread::network");
  1255. auto nscope = tracer->WithActiveSpan(nspan);
  1256. try {
  1257. // fprintf(stderr, "%s: commitThread: network\n", _myAddressStr.c_str());
  1258. pqxx::work w(*c->c);
  1259. std::string id = config["id"];
  1260. fprintf(stderr, "commit network %s\n", id.c_str());
  1261. pqxx::row nwrow =
  1262. w.exec("SELECT COUNT(id) frontend FROM networks_ctl WHERE id = $1", pqxx::params { id })
  1263. .one_row();
  1264. int nwcount = nwrow[0].as<int>();
  1265. bool isNewNetwork = (nwcount == 0);
  1266. std::string frontend = "";
  1267. if (! isNewNetwork) {
  1268. pqxx::row nwrow =
  1269. w.exec("SELECT frontend FROM networks_ctl WHERE id = $1", pqxx::params { id })
  1270. .one_row();
  1271. frontend = nwrow[0].as<std::string>();
  1272. }
  1273. std::string change_source;
  1274. if (! config["change_source"].is_null()) {
  1275. change_source = config["change_source"];
  1276. }
  1277. if (! isNewNetwork && change_source != "controller" && frontend != change_source) {
  1278. // if it is not a new network and the change source is not the controller and doesn't match
  1279. // the frontend, don't apply the change.
  1280. fprintf(
  1281. stderr,
  1282. "Skipping network update %s. isNewNetwork: %s, change_source: %s, frontend: %s\n",
  1283. id.c_str(), isNewNetwork ? "true" : "false", change_source.c_str(), frontend.c_str());
  1284. continue;
  1285. }
  1286. pqxx::result res = w.exec(
  1287. "INSERT INTO networks_ctl (id, name, configuration, controller_id, revision, frontend) "
  1288. "VALUES ($1, $2, $3, $4, $5, $6) "
  1289. "ON CONFLICT (id) DO UPDATE SET "
  1290. "name = EXCLUDED.name, configuration = EXCLUDED.configuration, revision = "
  1291. "EXCLUDED.revision+1, "
  1292. "frontend = EXCLUDED.frontend",
  1293. pqxx::params { id, OSUtils::jsonString(config["name"], ""), OSUtils::jsonDump(config, -1),
  1294. _myAddressStr, ((uint64_t)config["revision"]), change_source });
  1295. w.commit();
  1296. if (_listenerMode == LISTENER_MODE_PUBSUB) {
  1297. // Publish change to pubsub stream
  1298. if (config["change_source"].is_null() || config["change_source"] == "controller") {
  1299. nlohmann::json oldNetwork;
  1300. nlohmann::json newNetwork = config;
  1301. if (! isNewNetwork) {
  1302. oldNetwork = _getNetwork(w, id);
  1303. }
  1304. _changeNotifier->notifyNetworkChange(oldNetwork, newNetwork, frontend);
  1305. }
  1306. }
  1307. const uint64_t nwidInt = OSUtils::jsonIntHex(config["id"], 0ULL);
  1308. if (nwidInt) {
  1309. nlohmann::json nwOrig;
  1310. nlohmann::json nwNew(config);
  1311. get(nwidInt, nwOrig);
  1312. _networkChanged(nwOrig, nwNew, qitem.notifyListeners);
  1313. }
  1314. else {
  1315. fprintf(
  1316. stderr, "%s: Can't notify network changed: %llu\n", _myAddressStr.c_str(),
  1317. (unsigned long long)nwidInt);
  1318. }
  1319. }
  1320. catch (std::exception& e) {
  1321. nspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1322. fprintf(stderr, "%s ERROR: Error updating network: %s\n", _myAddressStr.c_str(), e.what());
  1323. }
  1324. if (_listenerMode == LISTENER_MODE_REDIS && _redisMemberStatus) {
  1325. try {
  1326. std::string id = config["id"];
  1327. std::string controllerId = _myAddressStr.c_str();
  1328. std::string key = "networks:{" + controllerId + "}";
  1329. if (_cc->redisConfig->clusterMode) {
  1330. _cluster->sadd(key, id);
  1331. }
  1332. else {
  1333. _redis->sadd(key, id);
  1334. }
  1335. }
  1336. catch (sw::redis::Error& e) {
  1337. nspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1338. fprintf(stderr, "ERROR: Error adding network to Redis: %s\n", e.what());
  1339. }
  1340. }
  1341. }
  1342. else if (objtype == "_delete_network") {
  1343. auto dspan = tracer->StartSpan("CentralDB::commitThread::_delete_network");
  1344. auto dscope = tracer->WithActiveSpan(dspan);
  1345. // fprintf(stderr, "%s: commitThread: delete network\n", _myAddressStr.c_str());
  1346. try {
  1347. pqxx::work w(*c->c);
  1348. std::string networkId = config["id"];
  1349. fprintf(stderr, "Deleting network %s\n", networkId.c_str());
  1350. w.exec("DELETE FROM network_memberships_ctl WHERE network_id = $1", pqxx::params { networkId });
  1351. w.exec("DELETE FROM networks_ctl WHERE id = $1", pqxx::params { networkId });
  1352. w.commit();
  1353. uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  1354. json oldConfig;
  1355. get(nwidInt, oldConfig);
  1356. json empty;
  1357. _networkChanged(oldConfig, empty, qitem.notifyListeners);
  1358. }
  1359. catch (std::exception& e) {
  1360. dspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1361. fprintf(stderr, "%s ERROR: Error deleting network: %s\n", _myAddressStr.c_str(), e.what());
  1362. }
  1363. if (_listenerMode == LISTENER_MODE_REDIS && _redisMemberStatus) {
  1364. try {
  1365. std::string id = config["id"];
  1366. std::string controllerId = _myAddressStr.c_str();
  1367. std::string key = "networks:{" + controllerId + "}";
  1368. if (_cc->redisConfig->clusterMode) {
  1369. _cluster->srem(key, id);
  1370. _cluster->del("network-nodes-online:{" + controllerId + "}:" + id);
  1371. }
  1372. else {
  1373. _redis->srem(key, id);
  1374. _redis->del("network-nodes-online:{" + controllerId + "}:" + id);
  1375. }
  1376. }
  1377. catch (sw::redis::Error& e) {
  1378. dspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1379. fprintf(stderr, "ERROR: Error adding network to Redis: %s\n", e.what());
  1380. }
  1381. }
  1382. }
  1383. else if (objtype == "_delete_member") {
  1384. auto mspan = tracer->StartSpan("CentralDB::commitThread::_delete_member");
  1385. auto mscope = tracer->WithActiveSpan(mspan);
  1386. // fprintf(stderr, "%s commitThread: delete member\n", _myAddressStr.c_str());
  1387. try {
  1388. pqxx::work w(*c->c);
  1389. std::string memberId = config["id"];
  1390. std::string networkId = config["nwid"];
  1391. fprintf(stderr, "Deleting member %s-%s\n", networkId.c_str(), memberId.c_str());
  1392. pqxx::result res =
  1393. w.exec(
  1394. "DELETE FROM network_memberships_ctl WHERE device_id = $1 AND network_id = $2",
  1395. pqxx::params { memberId, networkId })
  1396. .no_rows();
  1397. w.commit();
  1398. uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  1399. uint64_t memberidInt = OSUtils::jsonIntHex(config["id"], 0ULL);
  1400. nlohmann::json networkConfig;
  1401. nlohmann::json oldConfig;
  1402. get(nwidInt, networkConfig, memberidInt, oldConfig);
  1403. json empty;
  1404. _memberChanged(oldConfig, empty, qitem.notifyListeners);
  1405. }
  1406. catch (std::exception& e) {
  1407. mspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1408. fprintf(stderr, "%s ERROR: Error deleting member: %s\n", _myAddressStr.c_str(), e.what());
  1409. }
  1410. if (_listenerMode == LISTENER_MODE_REDIS && _redisMemberStatus) {
  1411. try {
  1412. std::string memberId = config["id"];
  1413. std::string networkId = config["nwid"];
  1414. std::string controllerId = _myAddressStr.c_str();
  1415. std::string key = "network-nodes-all:{" + controllerId + "}:" + networkId;
  1416. if (_cc->redisConfig->clusterMode) {
  1417. _cluster->srem(key, memberId);
  1418. _cluster->del("member:{" + controllerId + "}:" + networkId + ":" + memberId);
  1419. }
  1420. else {
  1421. _redis->srem(key, memberId);
  1422. _redis->del("member:{" + controllerId + "}:" + networkId + ":" + memberId);
  1423. }
  1424. }
  1425. catch (sw::redis::Error& e) {
  1426. mspan->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1427. fprintf(stderr, "ERROR: Error deleting member from Redis: %s\n", e.what());
  1428. }
  1429. }
  1430. }
  1431. else {
  1432. fprintf(stderr, "%s ERROR: unknown objtype\n", _myAddressStr.c_str());
  1433. }
  1434. }
  1435. catch (std::exception& e) {
  1436. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  1437. fprintf(stderr, "%s ERROR: Error getting objtype: %s\n", _myAddressStr.c_str(), e.what());
  1438. }
  1439. _pool->unborrow(c);
  1440. c.reset();
  1441. }
  1442. }
  1443. fprintf(stderr, "%s commitThread finished\n", _myAddressStr.c_str());
  1444. }
  1445. void CentralDB::notifyNewMember(const std::string& networkID, const std::string& memberID)
  1446. {
  1447. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  1448. auto tracer = provider->GetTracer("CentralDB");
  1449. auto span = tracer->StartSpan("CentralDB::notifyNewMember");
  1450. auto scope = tracer->WithActiveSpan(span);
  1451. rustybits::smee_client_notify_network_joined(_smee, networkID.c_str(), memberID.c_str());
  1452. }
  1453. void CentralDB::onlineNotificationThread()
  1454. {
  1455. waitForReady();
  1456. while (_run == 1) {
  1457. {
  1458. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  1459. auto tracer = provider->GetTracer("CentralDB");
  1460. auto span = tracer->StartSpan("CentralDB::onlineNotificationThread");
  1461. auto scope = tracer->WithActiveSpan(span);
  1462. try {
  1463. std::unordered_map<std::pair<uint64_t, uint64_t>, NodeOnlineRecord, _PairHasher> lastOnline;
  1464. {
  1465. std::lock_guard<std::mutex> l(_lastOnline_l);
  1466. lastOnline.swap(_lastOnline);
  1467. }
  1468. uint64_t updateCount = 0;
  1469. auto c = _pool->borrow();
  1470. pqxx::work w(*c->c);
  1471. for (auto i = lastOnline.begin(); i != lastOnline.end(); ++i) {
  1472. updateCount += 1;
  1473. uint64_t nwid_i = i->first.first;
  1474. char nwidTmp[64];
  1475. char memTmp[64];
  1476. char ipTmp[64];
  1477. OSUtils::ztsnprintf(nwidTmp, sizeof(nwidTmp), "%.16llx", nwid_i);
  1478. OSUtils::ztsnprintf(memTmp, sizeof(memTmp), "%.10llx", i->first.second);
  1479. nlohmann::json network, member;
  1480. if (! get(nwid_i, network, i->first.second, member)) {
  1481. continue; // skip non existent networks/members
  1482. }
  1483. std::string networkId(nwidTmp);
  1484. std::string memberId(memTmp);
  1485. try {
  1486. // check if the member exists first.
  1487. //
  1488. // exec_params1 will throw pqxx::unexpected_rows if not exactly one row is returned. If that's
  1489. // the case, skip this record and move on.
  1490. pqxx::row r =
  1491. w.exec(
  1492. "SELECT device_id, network_id FROM network_memberships_ctl WHERE network_id = "
  1493. "$1 AND device_id "
  1494. "= $2",
  1495. pqxx::params { networkId, memberId })
  1496. .one_row();
  1497. }
  1498. catch (pqxx::unexpected_rows& e) {
  1499. continue;
  1500. }
  1501. int64_t ts = i->second.lastSeen;
  1502. std::string ipAddr = i->second.physicalAddress.toIpString(ipTmp);
  1503. std::string timestamp = std::to_string(ts);
  1504. std::string osArch = i->second.osArch;
  1505. std::vector<std::string> osArchSplit = split(osArch, '/');
  1506. std::string os = "unknown";
  1507. std::string arch = "unknown";
  1508. std::string frontend = OSUtils::jsonString(network["frontend"], "");
  1509. int vMajor = OSUtils::jsonInt(member["vMajor"], 0);
  1510. int vMinor = OSUtils::jsonInt(member["vMinor"], 0);
  1511. int vRev = OSUtils::jsonInt(member["vRev"], 0);
  1512. std::string version;
  1513. if (vMajor <= 0 && vMinor <= 0 && vRev <= 0) {
  1514. vMajor = 0;
  1515. vMinor = 0;
  1516. vRev = 0;
  1517. version = "unknown";
  1518. }
  1519. else {
  1520. version =
  1521. "v" + std::to_string(vMajor) + "." + std::to_string(vMinor) + "." + std::to_string(vRev);
  1522. }
  1523. if (osArchSplit.size() == 2) {
  1524. os = osArchSplit[0];
  1525. arch = osArchSplit[1];
  1526. }
  1527. _statusWriter->updateNodeStatus(
  1528. networkId, memberId, os, arch, version, i->second.physicalAddress, ts, frontend);
  1529. }
  1530. _statusWriter->writePending();
  1531. w.commit();
  1532. _pool->unborrow(c);
  1533. }
  1534. catch (std::exception& e) {
  1535. fprintf(stderr, "%s: error in onlinenotification thread: %s\n", _myAddressStr.c_str(), e.what());
  1536. }
  1537. }
  1538. std::this_thread::sleep_for(std::chrono::seconds(10));
  1539. }
  1540. }
  1541. nlohmann::json CentralDB::_getNetworkMember(pqxx::work& tx, const std::string networkID, const std::string memberID)
  1542. {
  1543. nlohmann::json out;
  1544. try {
  1545. pqxx::row row =
  1546. tx.exec(
  1547. "SELECT nm.device_id, nm.network_id, nm.authorized, nm.active_bridge, nm.ip_assignments, "
  1548. "nm.no_auto_assign_ips, "
  1549. "nm.sso_exempt, (EXTRACT(EPOCH FROM nm.authentication_expiry_time AT TIME ZONE 'UTC')*1000)::bigint, "
  1550. "(EXTRACT(EPOCH FROM nm.creation_time AT TIME ZONE 'UTC')*1000)::bigint, nm.identity, "
  1551. "(EXTRACT(EPOCH FROM nm.last_authorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  1552. "(EXTRACT(EPOCH FROM nm.last_deauthorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  1553. "nm.remote_trace_level, nm.remote_trace_target, nm.revision, nm.capabilities, nm.tags, "
  1554. "nm.frontend "
  1555. "FROM network_memberships_ctl nm "
  1556. "INNER JOIN networks_ctl n "
  1557. " ON nm.network_id = n.id "
  1558. "WHERE nm.network_id = $1 AND nm.device_id = $2",
  1559. pqxx::params { networkID, memberID })
  1560. .one_row();
  1561. bool authorized = row[2].as<bool>();
  1562. std::optional<bool> active_bridge =
  1563. row[3].is_null() ? std::optional<bool>() : std::optional<bool>(row[3].as<bool>());
  1564. std::string ip_assignments = row[4].is_null() ? "{}" : row[4].as<std::string>();
  1565. std::optional<bool> no_auto_assign_ips =
  1566. row[5].is_null() ? std::optional<bool>() : std::optional<bool>(row[5].as<bool>());
  1567. std::optional<bool> sso_exempt =
  1568. row[6].is_null() ? std::optional<bool>() : std::optional<bool>(row[6].as<bool>());
  1569. std::optional<uint64_t> authentication_expiry_time =
  1570. row[7].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[7].as<uint64_t>());
  1571. std::optional<uint64_t> creation_time =
  1572. row[8].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[8].as<uint64_t>());
  1573. std::optional<std::string> identity =
  1574. row[9].is_null() ? std::optional<std::string>() : std::optional<std::string>(row[9].as<std::string>());
  1575. std::optional<uint64_t> last_authorized_time =
  1576. row[10].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[10].as<uint64_t>());
  1577. std::optional<uint64_t> last_deauthorized_time =
  1578. row[11].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[11].as<uint64_t>());
  1579. std::optional<int32_t> remote_trace_level =
  1580. row[12].is_null() ? std::optional<int32_t>() : std::optional<int32_t>(row[12].as<int32_t>());
  1581. std::optional<std::string> remote_trace_target =
  1582. row[13].is_null() ? std::optional<std::string>() : std::optional<std::string>(row[13].as<std::string>());
  1583. std::optional<uint64_t> revision =
  1584. row[14].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[14].as<uint64_t>());
  1585. std::optional<std::string> capabilities =
  1586. row[15].is_null() ? std::optional<std::string>() : std::optional<std::string>(row[15].as<std::string>());
  1587. std::optional<std::string> tags =
  1588. row[16].is_null() ? std::optional<std::string>() : std::optional<std::string>(row[16].as<std::string>());
  1589. std::string frontend = row[17].is_null() ? "" : row[17].as<std::string>();
  1590. out["objtype"] = "member";
  1591. out["id"] = memberID;
  1592. out["nwid"] = networkID;
  1593. out["address"] = identity.value_or("");
  1594. out["authorized"] = authorized;
  1595. out["activeBridge"] = active_bridge.value_or(false);
  1596. out["ipAssignments"] = json::array();
  1597. if (ip_assignments != "{}" && ip_assignments != "[]") {
  1598. std::string tmp = ip_assignments.substr(1, ip_assignments.length() - 2);
  1599. std::vector<std::string> addrs = split(tmp, ',');
  1600. for (auto it = addrs.begin(); it != addrs.end(); ++it) {
  1601. out["ipAssignments"].push_back(*it);
  1602. }
  1603. }
  1604. out["capabilities"] = json::parse(capabilities.value_or("[]"));
  1605. out["creationTime"] = creation_time.value_or(0);
  1606. out["lastAuthorizedTime"] = last_authorized_time.value_or(0);
  1607. out["lastDeauthorizedTime"] = last_deauthorized_time.value_or(0);
  1608. out["noAutoAssignIps"] = no_auto_assign_ips.value_or(false);
  1609. out["remoteTraceLevel"] = remote_trace_level.value_or(0);
  1610. out["remoteTraceTarget"] = remote_trace_target.value_or(nullptr);
  1611. out["revision"] = revision.value_or(0);
  1612. out["ssoExempt"] = sso_exempt.value_or(false);
  1613. out["authenticationExpiryTime"] = authentication_expiry_time.value_or(0);
  1614. out["tags"] = json::parse(tags.value_or("[]"));
  1615. out["frontend"] = frontend;
  1616. }
  1617. catch (std::exception& e) {
  1618. fprintf(
  1619. stderr, "ERROR: Error getting network member %s-%s: %s\n", networkID.c_str(), memberID.c_str(), e.what());
  1620. return nlohmann::json();
  1621. }
  1622. return out;
  1623. }
  1624. nlohmann::json CentralDB::_getNetwork(pqxx::work& tx, const std::string networkID)
  1625. {
  1626. nlohmann::json out;
  1627. try {
  1628. std::optional<std::string> name;
  1629. std::string cfg;
  1630. std::optional<uint64_t> creation_time;
  1631. std::optional<uint64_t> last_modified;
  1632. std::optional<uint64_t> revision;
  1633. std::string frontend;
  1634. pqxx::row row = tx.exec(
  1635. "SELECT id, name, configuration , (EXTRACT(EPOCH FROM creation_time AT TIME ZONE "
  1636. "'UTC')*1000)::bigint, "
  1637. "(EXTRACT(EPOCH FROM last_modified AT TIME ZONE 'UTC')*1000)::bigint, revision, frontend "
  1638. "FROM networks_ctl WHERE id = $1",
  1639. pqxx::params { networkID })
  1640. .one_row();
  1641. cfg = row[2].as<std::string>();
  1642. creation_time = row[3].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[3].as<uint64_t>());
  1643. last_modified = row[4].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[4].as<uint64_t>());
  1644. revision = row[5].is_null() ? std::optional<uint64_t>() : std::optional<uint64_t>(row[5].as<uint64_t>());
  1645. frontend = row[6].is_null() ? "" : row[6].as<std::string>();
  1646. nlohmann::json cfgtmp = nlohmann::json::parse(cfg);
  1647. if (! cfgtmp.is_object()) {
  1648. fprintf(stderr, "ERROR: Network %s configuration is not a JSON object\n", networkID.c_str());
  1649. return nlohmann::json();
  1650. }
  1651. out["objtype"] = "network";
  1652. out["id"] = row[0].as<std::string>();
  1653. out["name"] = row[1].is_null() ? "" : row[1].as<std::string>();
  1654. out["creationTime"] = creation_time.value_or(0);
  1655. out["lastModified"] = last_modified.value_or(0);
  1656. out["revision"] = revision.value_or(0);
  1657. out["capabilities"] = cfgtmp["capabilities"].is_array() ? cfgtmp["capabilities"] : json::array();
  1658. out["enableBroadcast"] = cfgtmp["enableBroadcast"].is_boolean() ? cfgtmp["enableBroadcast"].get<bool>() : false;
  1659. out["mtu"] = cfgtmp["mtu"].is_number() ? cfgtmp["mtu"].get<int32_t>() : 2800;
  1660. out["multicastLimit"] = cfgtmp["multicastLimit"].is_number() ? cfgtmp["multicastLimit"].get<int32_t>() : 64;
  1661. out["private"] = cfgtmp["private"].is_boolean() ? cfgtmp["private"].get<bool>() : true;
  1662. out["remoteTraceLevel"] =
  1663. cfgtmp["remoteTraceLevel"].is_number() ? cfgtmp["remoteTraceLevel"].get<int32_t>() : 0;
  1664. out["remoteTraceTarget"] =
  1665. cfgtmp["remoteTraceTarget"].is_string() ? cfgtmp["remoteTraceTarget"].get<std::string>() : "";
  1666. out["revision"] = revision.value_or(0);
  1667. out["rules"] = cfgtmp["rules"].is_array() ? cfgtmp["rules"] : json::array();
  1668. out["tags"] = cfgtmp["tags"].is_array() ? cfgtmp["tags"] : json::array();
  1669. if (cfgtmp["v4AssignMode"].is_object()) {
  1670. out["v4AssignMode"] = cfgtmp["v4AssignMode"];
  1671. }
  1672. else {
  1673. out["v4AssignMode"] = json::object();
  1674. out["v4AssignMode"]["zt"] = true;
  1675. }
  1676. if (cfgtmp["v6AssignMode"].is_object()) {
  1677. out["v6AssignMode"] = cfgtmp["v6AssignMode"];
  1678. }
  1679. else {
  1680. out["v6AssignMode"] = json::object();
  1681. out["v6AssignMode"]["zt"] = true;
  1682. out["v6AssignMode"]["6plane"] = true;
  1683. out["v6AssignMode"]["rfc4193"] = false;
  1684. }
  1685. out["ssoEnabled"] = cfgtmp["ssoEnabled"].is_boolean() ? cfgtmp["ssoEnabled"].get<bool>() : false;
  1686. out["objtype"] = "network";
  1687. out["routes"] = cfgtmp["routes"].is_array() ? cfgtmp["routes"] : json::array();
  1688. out["clientId"] = cfgtmp["clientId"].is_string() ? cfgtmp["clientId"].get<std::string>() : "";
  1689. out["authorizationEndpoint"] =
  1690. cfgtmp["authorizationEndpoint"].is_string() ? cfgtmp["authorizationEndpoint"].get<std::string>() : nullptr;
  1691. out["provider"] = cfgtmp["ssoProvider"].is_string() ? cfgtmp["ssoProvider"].get<std::string>() : "";
  1692. if (! cfgtmp["dns"].is_object()) {
  1693. cfgtmp["dns"] = json::object();
  1694. cfgtmp["dns"]["domain"] = "";
  1695. cfgtmp["dns"]["servers"] = json::array();
  1696. }
  1697. else {
  1698. out["dns"] = cfgtmp["dns"];
  1699. }
  1700. out["ipAssignmentPools"] = cfgtmp["ipAssignmentPools"].is_array() ? cfgtmp["ipAssignmentPools"] : json::array();
  1701. out["frontend"] = row[6].as<std::string>();
  1702. }
  1703. catch (std::exception& e) {
  1704. fprintf(stderr, "ERROR: Error getting network %s: %s\n", networkID.c_str(), e.what());
  1705. return nlohmann::json();
  1706. }
  1707. return out;
  1708. }
  1709. #endif // ZT_CONTROLLER_USE_LIBPQ