CV2.cpp 43 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264
  1. /* (c) ZeroTier, Inc.
  2. * See LICENSE.txt in nonfree/
  3. */
  4. #ifdef ZT_CONTROLLER_USE_LIBPQ
  5. #include "CV2.hpp"
  6. #include "../../node/Constants.hpp"
  7. #include "../../node/SHA512.hpp"
  8. #include "../../version.h"
  9. #include "CtlUtil.hpp"
  10. #include "EmbeddedNetworkController.hpp"
  11. #include "opentelemetry/trace/provider.h"
  12. #include <chrono>
  13. #include <climits>
  14. #include <iomanip>
  15. #include <libpq-fe.h>
  16. #include <rustybits.h>
  17. #include <sstream>
  18. using json = nlohmann::json;
  19. namespace {
  20. }
  21. using namespace ZeroTier;
  22. CV2::CV2(const Identity& myId, const char* path, int listenPort)
  23. : DB()
  24. , _pool()
  25. , _myId(myId)
  26. , _myAddress(myId.address())
  27. , _ready(0)
  28. , _connected(1)
  29. , _run(1)
  30. , _waitNoticePrinted(false)
  31. , _listenPort(listenPort)
  32. {
  33. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  34. auto tracer = provider->GetTracer("cv2");
  35. auto span = tracer->StartSpan("cv2::CV2");
  36. auto scope = tracer->WithActiveSpan(span);
  37. rustybits::init_async_runtime();
  38. fprintf(stderr, "CV2::CV2\n");
  39. char myAddress[64];
  40. _myAddressStr = myId.address().toString(myAddress);
  41. _connString = std::string(path);
  42. auto f = std::make_shared<PostgresConnFactory>(_connString);
  43. _pool =
  44. std::make_shared<ConnectionPool<PostgresConnection> >(15, 5, std::static_pointer_cast<ConnectionFactory>(f));
  45. memset(_ssoPsk, 0, sizeof(_ssoPsk));
  46. char* const ssoPskHex = getenv("ZT_SSO_PSK");
  47. #ifdef ZT_TRACE
  48. fprintf(stderr, "ZT_SSO_PSK: %s\n", ssoPskHex);
  49. #endif
  50. if (ssoPskHex) {
  51. // SECURITY: note that ssoPskHex will always be null-terminated if libc actually
  52. // returns something non-NULL. If the hex encodes something shorter than 48 bytes,
  53. // it will be padded at the end with zeroes. If longer, it'll be truncated.
  54. Utils::unhex(ssoPskHex, _ssoPsk, sizeof(_ssoPsk));
  55. }
  56. _readyLock.lock();
  57. fprintf(
  58. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL waiting for initial data download..." ZT_EOL_S,
  59. ::_timestr(), (unsigned long long)_myAddress.toInt());
  60. _waitNoticePrinted = true;
  61. initializeNetworks();
  62. initializeMembers();
  63. _heartbeatThread = std::thread(&CV2::heartbeat, this);
  64. _membersDbWatcher = std::thread(&CV2::membersDbWatcher, this);
  65. _networksDbWatcher = std::thread(&CV2::networksDbWatcher, this);
  66. for (int i = 0; i < ZT_CENTRAL_CONTROLLER_COMMIT_THREADS; ++i) {
  67. _commitThread[i] = std::thread(&CV2::commitThread, this);
  68. }
  69. _onlineNotificationThread = std::thread(&CV2::onlineNotificationThread, this);
  70. }
  71. CV2::~CV2()
  72. {
  73. rustybits::shutdown_async_runtime();
  74. _run = 0;
  75. std::this_thread::sleep_for(std::chrono::milliseconds(100));
  76. _heartbeatThread.join();
  77. _membersDbWatcher.join();
  78. _networksDbWatcher.join();
  79. _commitQueue.stop();
  80. for (int i = 0; i < ZT_CENTRAL_CONTROLLER_COMMIT_THREADS; ++i) {
  81. _commitThread[i].join();
  82. }
  83. _onlineNotificationThread.join();
  84. }
  85. bool CV2::waitForReady()
  86. {
  87. while (_ready < 2) {
  88. _readyLock.lock();
  89. _readyLock.unlock();
  90. }
  91. return true;
  92. }
  93. bool CV2::isReady()
  94. {
  95. return (_ready == 2) && _connected;
  96. }
  97. void CV2::_memberChanged(nlohmann::json& old, nlohmann::json& memberConfig, bool notifyListeners)
  98. {
  99. DB::_memberChanged(old, memberConfig, notifyListeners);
  100. }
  101. void CV2::_networkChanged(nlohmann::json& old, nlohmann::json& networkConfig, bool notifyListeners)
  102. {
  103. DB::_networkChanged(old, networkConfig, notifyListeners);
  104. }
  105. bool CV2::save(nlohmann::json& record, bool notifyListeners)
  106. {
  107. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  108. auto tracer = provider->GetTracer("cv2");
  109. auto span = tracer->StartSpan("cv2::save");
  110. auto scope = tracer->WithActiveSpan(span);
  111. bool modified = false;
  112. try {
  113. if (! record.is_object()) {
  114. fprintf(stderr, "record is not an object?!?\n");
  115. return false;
  116. }
  117. const std::string objtype = record["objtype"];
  118. if (objtype == "network") {
  119. auto nspan = tracer->StartSpan("cv2::save::network");
  120. auto nscope = tracer->WithActiveSpan(nspan);
  121. // fprintf(stderr, "network save\n");
  122. const uint64_t nwid = OSUtils::jsonIntHex(record["id"], 0ULL);
  123. if (nwid) {
  124. nlohmann::json old;
  125. get(nwid, old);
  126. if ((! old.is_object()) || (! _compareRecords(old, record))) {
  127. record["revision"] = OSUtils::jsonInt(record["revision"], 0ULL) + 1ULL;
  128. _commitQueue.post(std::pair<nlohmann::json, bool>(record, notifyListeners));
  129. modified = true;
  130. }
  131. }
  132. }
  133. else if (objtype == "member") {
  134. auto mspan = tracer->StartSpan("cv2::save::member");
  135. auto mscope = tracer->WithActiveSpan(mspan);
  136. std::string networkId = record["nwid"];
  137. std::string memberId = record["id"];
  138. const uint64_t nwid = OSUtils::jsonIntHex(record["nwid"], 0ULL);
  139. const uint64_t id = OSUtils::jsonIntHex(record["id"], 0ULL);
  140. // fprintf(stderr, "member save %s-%s\n", networkId.c_str(), memberId.c_str());
  141. if ((id) && (nwid)) {
  142. nlohmann::json network, old;
  143. get(nwid, network, id, old);
  144. if ((! old.is_object()) || (! _compareRecords(old, record))) {
  145. // fprintf(stderr, "commit queue post\n");
  146. record["revision"] = OSUtils::jsonInt(record["revision"], 0ULL) + 1ULL;
  147. _commitQueue.post(std::pair<nlohmann::json, bool>(record, notifyListeners));
  148. modified = true;
  149. }
  150. else {
  151. // fprintf(stderr, "no change\n");
  152. }
  153. }
  154. }
  155. else {
  156. fprintf(stderr, "uhh waaat\n");
  157. }
  158. }
  159. catch (std::exception& e) {
  160. fprintf(stderr, "Error on PostgreSQL::save: %s\n", e.what());
  161. }
  162. catch (...) {
  163. fprintf(stderr, "Unknown error on PostgreSQL::save\n");
  164. }
  165. return modified;
  166. }
  167. void CV2::eraseNetwork(const uint64_t networkId)
  168. {
  169. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  170. auto tracer = provider->GetTracer("cv2");
  171. auto span = tracer->StartSpan("cv2::eraseNetwork");
  172. auto scope = tracer->WithActiveSpan(span);
  173. char networkIdStr[17];
  174. std::string nwid = Utils::hex(networkId, networkIdStr);
  175. span->SetAttribute("network_id", nwid);
  176. fprintf(stderr, "CV2::eraseNetwork\n");
  177. waitForReady();
  178. std::pair<nlohmann::json, bool> tmp;
  179. tmp.first["id"] = nwid;
  180. tmp.first["objtype"] = "_delete_network";
  181. tmp.second = true;
  182. _commitQueue.post(tmp);
  183. // nlohmann::json nullJson;
  184. //_networkChanged(tmp.first, nullJson, isReady());
  185. }
  186. void CV2::eraseMember(const uint64_t networkId, const uint64_t memberId)
  187. {
  188. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  189. auto tracer = provider->GetTracer("cv2");
  190. auto span = tracer->StartSpan("cv2::eraseMember");
  191. auto scope = tracer->WithActiveSpan(span);
  192. char networkIdStr[17];
  193. char memberIdStr[11];
  194. span->SetAttribute("network_id", Utils::hex(networkId, networkIdStr));
  195. span->SetAttribute("member_id", Utils::hex10(memberId, memberIdStr));
  196. fprintf(stderr, "PostgreSQL::eraseMember\n");
  197. char tmp2[24];
  198. waitForReady();
  199. std::pair<nlohmann::json, bool> tmp, nw;
  200. Utils::hex(networkId, tmp2);
  201. tmp.first["nwid"] = tmp2;
  202. Utils::hex(memberId, tmp2);
  203. tmp.first["id"] = tmp2;
  204. tmp.first["objtype"] = "_delete_member";
  205. tmp.second = true;
  206. _commitQueue.post(tmp);
  207. // nlohmann::json nullJson;
  208. //_memberChanged(tmp.first, nullJson, isReady());
  209. }
  210. void CV2::nodeIsOnline(
  211. const uint64_t networkId,
  212. const uint64_t memberId,
  213. const InetAddress& physicalAddress,
  214. const char* osArch)
  215. {
  216. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  217. auto tracer = provider->GetTracer("cv2");
  218. auto span = tracer->StartSpan("cv2::nodeIsOnline");
  219. auto scope = tracer->WithActiveSpan(span);
  220. char networkIdStr[17];
  221. char memberIdStr[11];
  222. char ipAddressStr[INET6_ADDRSTRLEN];
  223. span->SetAttribute("network_id", Utils::hex(networkId, networkIdStr));
  224. span->SetAttribute("member_id", Utils::hex10(memberId, memberIdStr));
  225. span->SetAttribute("physical_address", ipAddressStr);
  226. span->SetAttribute("os_arch", osArch);
  227. std::lock_guard<std::mutex> l(_lastOnline_l);
  228. NodeOnlineRecord& i = _lastOnline[std::pair<uint64_t, uint64_t>(networkId, memberId)];
  229. i.lastSeen = OSUtils::now();
  230. if (physicalAddress) {
  231. i.physicalAddress = physicalAddress;
  232. }
  233. i.osArch = std::string(osArch);
  234. }
  235. void CV2::nodeIsOnline(const uint64_t networkId, const uint64_t memberId, const InetAddress& physicalAddress)
  236. {
  237. this->nodeIsOnline(networkId, memberId, physicalAddress, "unknown/unknown");
  238. }
  239. AuthInfo CV2::getSSOAuthInfo(const nlohmann::json& member, const std::string& redirectURL)
  240. {
  241. // TODO: Redo this for CV2
  242. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  243. auto tracer = provider->GetTracer("cv2");
  244. auto span = tracer->StartSpan("cv2::getSSOAuthInfo");
  245. auto scope = tracer->WithActiveSpan(span);
  246. Metrics::db_get_sso_info++;
  247. // NONCE is just a random character string. no semantic meaning
  248. // state = HMAC SHA384 of Nonce based on shared sso key
  249. //
  250. // need nonce timeout in database? make sure it's used within X time
  251. // X is 5 minutes for now. Make configurable later?
  252. //
  253. // how do we tell when a nonce is used? if auth_expiration_time is set
  254. std::string networkId = member["nwid"];
  255. std::string memberId = member["id"];
  256. char authenticationURL[4096] = { 0 };
  257. AuthInfo info;
  258. info.enabled = true;
  259. // if (memberId == "a10dccea52" && networkId == "8056c2e21c24673d") {
  260. // fprintf(stderr, "invalid authinfo for grant's machine\n");
  261. // info.version=1;
  262. // return info;
  263. // }
  264. // fprintf(stderr, "PostgreSQL::updateMemberOnLoad: %s-%s\n", networkId.c_str(), memberId.c_str());
  265. std::shared_ptr<PostgresConnection> c;
  266. try {
  267. // c = _pool->borrow();
  268. // pqxx::work w(*c->c);
  269. // char nonceBytes[16] = {0};
  270. // std::string nonce = "";
  271. // // check if the member exists first.
  272. // pqxx::row count = w.exec_params1("SELECT count(id) FROM ztc_member WHERE id = $1 AND network_id = $2 AND
  273. // deleted = false", memberId, networkId); if (count[0].as<int>() == 1) {
  274. // // get active nonce, if exists.
  275. // pqxx::result r = w.exec_params("SELECT nonce FROM ztc_sso_expiry "
  276. // "WHERE network_id = $1 AND member_id = $2 "
  277. // "AND ((NOW() AT TIME ZONE 'UTC') <= authentication_expiry_time) AND ((NOW() AT TIME ZONE 'UTC')
  278. // <= nonce_expiration)", networkId, memberId);
  279. // if (r.size() == 0) {
  280. // // no active nonce.
  281. // // find an unused nonce, if one exists.
  282. // pqxx::result r = w.exec_params("SELECT nonce FROM ztc_sso_expiry "
  283. // "WHERE network_id = $1 AND member_id = $2 "
  284. // "AND authentication_expiry_time IS NULL AND ((NOW() AT TIME ZONE 'UTC') <=
  285. // nonce_expiration)", networkId, memberId);
  286. // if (r.size() == 1) {
  287. // // we have an existing nonce. Use it
  288. // nonce = r.at(0)[0].as<std::string>();
  289. // Utils::unhex(nonce.c_str(), nonceBytes, sizeof(nonceBytes));
  290. // } else if (r.empty()) {
  291. // // create a nonce
  292. // Utils::getSecureRandom(nonceBytes, 16);
  293. // char nonceBuf[64] = {0};
  294. // Utils::hex(nonceBytes, sizeof(nonceBytes), nonceBuf);
  295. // nonce = std::string(nonceBuf);
  296. // pqxx::result ir = w.exec_params0("INSERT INTO ztc_sso_expiry "
  297. // "(nonce, nonce_expiration, network_id, member_id) VALUES "
  298. // "($1, TO_TIMESTAMP($2::double precision/1000), $3, $4)",
  299. // nonce, OSUtils::now() + 300000, networkId, memberId);
  300. // w.commit();
  301. // } else {
  302. // // > 1 ?!? Thats an error!
  303. // fprintf(stderr, "> 1 unused nonce!\n");
  304. // exit(6);
  305. // }
  306. // } else if (r.size() == 1) {
  307. // nonce = r.at(0)[0].as<std::string>();
  308. // Utils::unhex(nonce.c_str(), nonceBytes, sizeof(nonceBytes));
  309. // } else {
  310. // // more than 1 nonce in use? Uhhh...
  311. // fprintf(stderr, "> 1 nonce in use for network member?!?\n");
  312. // exit(7);
  313. // }
  314. // r = w.exec_params(
  315. // "SELECT oc.client_id, oc.authorization_endpoint, oc.issuer, oc.provider, oc.sso_impl_version "
  316. // "FROM ztc_network AS n "
  317. // "INNER JOIN ztc_org o "
  318. // " ON o.owner_id = n.owner_id "
  319. // "LEFT OUTER JOIN ztc_network_oidc_config noc "
  320. // " ON noc.network_id = n.id "
  321. // "LEFT OUTER JOIN ztc_oidc_config oc "
  322. // " ON noc.client_id = oc.client_id AND oc.org_id = o.org_id "
  323. // "WHERE n.id = $1 AND n.sso_enabled = true", networkId);
  324. // std::string client_id = "";
  325. // std::string authorization_endpoint = "";
  326. // std::string issuer = "";
  327. // std::string provider = "";
  328. // uint64_t sso_version = 0;
  329. // if (r.size() == 1) {
  330. // client_id = r.at(0)[0].as<std::optional<std::string>>().value_or("");
  331. // authorization_endpoint = r.at(0)[1].as<std::optional<std::string>>().value_or("");
  332. // issuer = r.at(0)[2].as<std::optional<std::string>>().value_or("");
  333. // provider = r.at(0)[3].as<std::optional<std::string>>().value_or("");
  334. // sso_version = r.at(0)[4].as<std::optional<uint64_t>>().value_or(1);
  335. // } else if (r.size() > 1) {
  336. // fprintf(stderr, "ERROR: More than one auth endpoint for an organization?!?!? NetworkID: %s\n",
  337. // networkId.c_str()); } else { fprintf(stderr, "No client or auth endpoint?!?\n");
  338. // }
  339. // info.version = sso_version;
  340. // // no catch all else because we don't actually care if no records exist here. just continue as
  341. // normal. if ((!client_id.empty())&&(!authorization_endpoint.empty())) {
  342. // uint8_t state[48];
  343. // HMACSHA384(_ssoPsk, nonceBytes, sizeof(nonceBytes), state);
  344. // char state_hex[256];
  345. // Utils::hex(state, 48, state_hex);
  346. // if (info.version == 0) {
  347. // char url[2048] = {0};
  348. // OSUtils::ztsnprintf(url, sizeof(authenticationURL),
  349. // "%s?response_type=id_token&response_mode=form_post&scope=openid+email+profile&redirect_uri=%s&nonce=%s&state=%s&client_id=%s",
  350. // authorization_endpoint.c_str(),
  351. // url_encode(redirectURL).c_str(),
  352. // nonce.c_str(),
  353. // state_hex,
  354. // client_id.c_str());
  355. // info.authenticationURL = std::string(url);
  356. // } else if (info.version == 1) {
  357. // info.ssoClientID = client_id;
  358. // info.issuerURL = issuer;
  359. // info.ssoProvider = provider;
  360. // info.ssoNonce = nonce;
  361. // info.ssoState = std::string(state_hex) + "_" +networkId;
  362. // info.centralAuthURL = redirectURL;
  363. // #ifdef ZT_DEBUG
  364. // fprintf(
  365. // stderr,
  366. // "ssoClientID: %s\nissuerURL: %s\nssoNonce: %s\nssoState: %s\ncentralAuthURL:
  367. // %s\nprovider: %s\n", info.ssoClientID.c_str(), info.issuerURL.c_str(), info.ssoNonce.c_str(),
  368. // info.ssoState.c_str(),
  369. // info.centralAuthURL.c_str(),
  370. // provider.c_str());
  371. // #endif
  372. // }
  373. // } else {
  374. // fprintf(stderr, "client_id: %s\nauthorization_endpoint: %s\n", client_id.c_str(),
  375. // authorization_endpoint.c_str());
  376. // }
  377. // }
  378. // _pool->unborrow(c);
  379. }
  380. catch (std::exception& e) {
  381. fprintf(stderr, "ERROR: Error updating member on load for network %s: %s\n", networkId.c_str(), e.what());
  382. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  383. }
  384. return info; // std::string(authenticationURL);
  385. }
  386. void CV2::initializeNetworks()
  387. {
  388. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  389. auto tracer = provider->GetTracer("cv2");
  390. auto span = tracer->StartSpan("cv2::initializeNetworks");
  391. auto scope = tracer->WithActiveSpan(span);
  392. fprintf(stderr, "Initializing networks...\n");
  393. try {
  394. char qbuf[2048];
  395. sprintf(
  396. qbuf,
  397. "SELECT id, name, configuration , (EXTRACT(EPOCH FROM creation_time AT TIME ZONE 'UTC')*1000)::bigint, "
  398. "(EXTRACT(EPOCH FROM last_modified AT TIME ZONE 'UTC')*1000)::bigint, revision "
  399. "FROM networks_ctl WHERE controller_id = '%s'",
  400. _myAddressStr.c_str());
  401. auto c = _pool->borrow();
  402. pqxx::work w(*c->c);
  403. fprintf(stderr, "Load networks from psql...\n");
  404. auto stream = pqxx::stream_from::query(w, qbuf);
  405. std::tuple<
  406. std::string // network ID
  407. ,
  408. std::optional<std::string> // name
  409. ,
  410. std::string // configuration
  411. ,
  412. std::optional<uint64_t> // creation_time
  413. ,
  414. std::optional<uint64_t> // last_modified
  415. ,
  416. std::optional<uint64_t> // revision
  417. >
  418. row;
  419. uint64_t count = 0;
  420. uint64_t total = 0;
  421. while (stream >> row) {
  422. auto start = std::chrono::high_resolution_clock::now();
  423. json empty;
  424. json config;
  425. initNetwork(config);
  426. std::string nwid = std::get<0>(row);
  427. std::string name = std::get<1>(row).value_or("");
  428. json cfgtmp = json::parse(std::get<2>(row));
  429. std::optional<uint64_t> created_at = std::get<3>(row);
  430. std::optional<uint64_t> last_modified = std::get<4>(row);
  431. std::optional<uint64_t> revision = std::get<5>(row);
  432. config["id"] = nwid;
  433. config["name"] = name;
  434. config["creationTime"] = created_at.value_or(0);
  435. config["lastModified"] = last_modified.value_or(0);
  436. config["revision"] = revision.value_or(0);
  437. config["capabilities"] = cfgtmp["capabilities"].is_array() ? cfgtmp["capabilities"] : json::array();
  438. config["enableBroadcast"] =
  439. cfgtmp["enableBroadcast"].is_boolean() ? cfgtmp["enableBroadcast"].get<bool>() : false;
  440. config["mtu"] = cfgtmp["mtu"].is_number() ? cfgtmp["mtu"].get<int32_t>() : 2800;
  441. config["multicastLimit"] =
  442. cfgtmp["multicastLimit"].is_number() ? cfgtmp["multicastLimit"].get<int32_t>() : 64;
  443. config["private"] = cfgtmp["private"].is_boolean() ? cfgtmp["private"].get<bool>() : true;
  444. config["remoteTraceLevel"] =
  445. cfgtmp["remoteTraceLevel"].is_number() ? cfgtmp["remoteTraceLevel"].get<int32_t>() : 0;
  446. config["remoteTraceTarget"] =
  447. cfgtmp["remoteTraceTarget"].is_string() ? cfgtmp["remoteTraceTarget"].get<std::string>() : "";
  448. config["revision"] = revision.value_or(0);
  449. config["rules"] = cfgtmp["rules"].is_array() ? cfgtmp["rules"] : json::array();
  450. config["tags"] = cfgtmp["tags"].is_array() ? cfgtmp["tags"] : json::array();
  451. if (cfgtmp["v4AssignMode"].is_object()) {
  452. config["v4AssignMode"] = cfgtmp["v4AssignMode"];
  453. }
  454. else {
  455. config["v4AssignMode"] = json::object();
  456. config["v4AssignMode"]["zt"] = true;
  457. }
  458. if (cfgtmp["v6AssignMode"].is_object()) {
  459. config["v6AssignMode"] = cfgtmp["v6AssignMode"];
  460. }
  461. else {
  462. config["v6AssignMode"] = json::object();
  463. config["v6AssignMode"]["zt"] = true;
  464. config["v6AssignMode"]["6plane"] = true;
  465. config["v6AssignMode"]["rfc4193"] = false;
  466. }
  467. config["ssoEnabled"] = cfgtmp["ssoEnabled"].is_boolean() ? cfgtmp["ssoEnabled"].get<bool>() : false;
  468. config["objtype"] = "network";
  469. config["routes"] = cfgtmp["routes"].is_array() ? cfgtmp["routes"] : json::array();
  470. config["clientId"] = cfgtmp["clientId"].is_string() ? cfgtmp["clientId"].get<std::string>() : "";
  471. config["authorizationEndpoint"] = cfgtmp["authorizationEndpoint"].is_string()
  472. ? cfgtmp["authorizationEndpoint"].get<std::string>()
  473. : nullptr;
  474. config["provider"] = cfgtmp["ssoProvider"].is_string() ? cfgtmp["ssoProvider"].get<std::string>() : "";
  475. if (! cfgtmp["dns"].is_object()) {
  476. cfgtmp["dns"] = json::object();
  477. cfgtmp["dns"]["domain"] = "";
  478. cfgtmp["dns"]["servers"] = json::array();
  479. }
  480. else {
  481. config["dns"] = cfgtmp["dns"];
  482. }
  483. config["ipAssignmentPools"] =
  484. cfgtmp["ipAssignmentPools"].is_array() ? cfgtmp["ipAssignmentPools"] : json::array();
  485. Metrics::network_count++;
  486. _networkChanged(empty, config, false);
  487. auto end = std::chrono::high_resolution_clock::now();
  488. auto dur = std::chrono::duration_cast<std::chrono::microseconds>(end - start);
  489. ;
  490. total += dur.count();
  491. ++count;
  492. if (count > 0 && count % 10000 == 0) {
  493. fprintf(stderr, "Averaging %lu us per network\n", (total / count));
  494. }
  495. }
  496. w.commit();
  497. _pool->unborrow(c);
  498. fprintf(stderr, "done.\n");
  499. if (++this->_ready == 2) {
  500. if (_waitNoticePrinted) {
  501. fprintf(
  502. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL data download complete." ZT_EOL_S, _timestr(),
  503. (unsigned long long)_myAddress.toInt());
  504. }
  505. _readyLock.unlock();
  506. }
  507. fprintf(stderr, "network init done\n");
  508. }
  509. catch (std::exception& e) {
  510. fprintf(stderr, "ERROR: Error initializing networks: %s\n", e.what());
  511. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  512. std::this_thread::sleep_for(std::chrono::milliseconds(5000));
  513. exit(-1);
  514. }
  515. }
  516. void CV2::initializeMembers()
  517. {
  518. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  519. auto tracer = provider->GetTracer("cv2");
  520. auto span = tracer->StartSpan("cv2::initializeMembers");
  521. auto scope = tracer->WithActiveSpan(span);
  522. std::string memberId;
  523. std::string networkId;
  524. try {
  525. char qbuf[2048];
  526. sprintf(
  527. qbuf,
  528. "SELECT nm.device_id, nm.network_id, nm.authorized, nm.active_bridge, nm.ip_assignments, "
  529. "nm.no_auto_assign_ips, "
  530. "nm.sso_exempt, (EXTRACT(EPOCH FROM nm.authentication_expiry_time AT TIME ZONE 'UTC')*1000)::bigint, "
  531. "(EXTRACT(EPOCH FROM nm.creation_time AT TIME ZONE 'UTC')*1000)::bigint, nm.identity, "
  532. "(EXTRACT(EPOCH FROM nm.last_authorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  533. "(EXTRACT(EPOCH FROM nm.last_deauthorized_time AT TIME ZONE 'UTC')*1000)::bigint, "
  534. "nm.remote_trace_level, nm.remote_trace_target, nm.revision, nm.capabilities, nm.tags "
  535. "FROM network_memberships_ctl nm "
  536. "INNER JOIN networks_ctl n "
  537. " ON nm.network_id = n.id "
  538. "WHERE n.controller_id = '%s'",
  539. _myAddressStr.c_str());
  540. auto c = _pool->borrow();
  541. pqxx::work w(*c->c);
  542. fprintf(stderr, "Load members from psql...\n");
  543. auto stream = pqxx::stream_from::query(w, qbuf);
  544. std::tuple<
  545. std::string // device ID
  546. ,
  547. std::string // network ID
  548. ,
  549. bool // authorized
  550. ,
  551. std::optional<bool> // active_bridge
  552. ,
  553. std::optional<std::string> // ip_assignments
  554. ,
  555. std::optional<bool> // no_auto_assign_ips
  556. ,
  557. std::optional<bool> // sso_exempt
  558. ,
  559. std::optional<uint64_t> // authentication_expiry_time
  560. ,
  561. std::optional<uint64_t> // creation_time
  562. ,
  563. std::optional<std::string> // identity
  564. ,
  565. std::optional<uint64_t> // last_authorized_time
  566. ,
  567. std::optional<uint64_t> // last_deauthorized_time
  568. ,
  569. std::optional<int32_t> // remote_trace_level
  570. ,
  571. std::optional<std::string> // remote_trace_target
  572. ,
  573. std::optional<uint64_t> // revision
  574. ,
  575. std::optional<std::string> // capabilities
  576. ,
  577. std::optional<std::string> // tags
  578. >
  579. row;
  580. uint64_t count = 0;
  581. uint64_t total = 0;
  582. while (stream >> row) {
  583. auto start = std::chrono::high_resolution_clock::now();
  584. json empty;
  585. json config;
  586. initMember(config);
  587. memberId = std::get<0>(row);
  588. networkId = std::get<1>(row);
  589. bool authorized = std::get<2>(row);
  590. std::optional<bool> active_bridge = std::get<3>(row);
  591. std::string ip_assignments = std::get<4>(row).value_or("");
  592. std::optional<bool> no_auto_assign_ips = std::get<5>(row);
  593. std::optional<bool> sso_exempt = std::get<6>(row);
  594. std::optional<uint64_t> authentication_expiry_time = std::get<7>(row);
  595. std::optional<uint64_t> creation_time = std::get<8>(row);
  596. std::optional<std::string> identity = std::get<9>(row);
  597. std::optional<uint64_t> last_authorized_time = std::get<10>(row);
  598. std::optional<uint64_t> last_deauthorized_time = std::get<11>(row);
  599. std::optional<int32_t> remote_trace_level = std::get<12>(row);
  600. std::optional<std::string> remote_trace_target = std::get<13>(row);
  601. std::optional<uint64_t> revision = std::get<14>(row);
  602. std::optional<std::string> capabilities = std::get<15>(row);
  603. std::optional<std::string> tags = std::get<16>(row);
  604. config["objtype"] = "member";
  605. config["id"] = memberId;
  606. config["address"] = identity.value_or("");
  607. config["nwid"] = networkId;
  608. config["authorized"] = authorized;
  609. config["activeBridge"] = active_bridge.value_or(false);
  610. config["ipAssignments"] = json::array();
  611. if (ip_assignments != "{}") {
  612. std::string tmp = ip_assignments.substr(1, ip_assignments.length() - 2);
  613. std::vector<std::string> addrs = split(tmp, ',');
  614. for (auto it = addrs.begin(); it != addrs.end(); ++it) {
  615. config["ipAssignments"].push_back(*it);
  616. }
  617. }
  618. config["capabilities"] = json::parse(capabilities.value_or("[]"));
  619. config["creationTime"] = creation_time.value_or(0);
  620. config["lastAuthorizedTime"] = last_authorized_time.value_or(0);
  621. config["lastDeauthorizedTime"] = last_deauthorized_time.value_or(0);
  622. config["noAutoAssignIPs"] = no_auto_assign_ips.value_or(false);
  623. config["remoteTraceLevel"] = remote_trace_level.value_or(0);
  624. config["remoteTraceTarget"] = remote_trace_target.value_or(nullptr);
  625. config["revision"] = revision.value_or(0);
  626. config["ssoExempt"] = sso_exempt.value_or(false);
  627. config["authenticationExpiryTime"] = authentication_expiry_time.value_or(0);
  628. config["tags"] = json::parse(tags.value_or("[]"));
  629. Metrics::member_count++;
  630. _memberChanged(empty, config, false);
  631. memberId = "";
  632. networkId = "";
  633. auto end = std::chrono::high_resolution_clock::now();
  634. auto dur = std::chrono::duration_cast<std::chrono::microseconds>(end - start);
  635. total += dur.count();
  636. ++count;
  637. if (count > 0 && count % 10000 == 0) {
  638. fprintf(stderr, "Averaging %lu us per member\n", (total / count));
  639. }
  640. }
  641. if (count > 0) {
  642. fprintf(stderr, "Took %lu us per member to load\n", (total / count));
  643. }
  644. stream.complete();
  645. w.commit();
  646. _pool->unborrow(c);
  647. fprintf(stderr, "done.\n");
  648. if (++this->_ready == 2) {
  649. if (_waitNoticePrinted) {
  650. fprintf(
  651. stderr, "[%s] NOTICE: %.10llx controller PostgreSQL data download complete." ZT_EOL_S, _timestr(),
  652. (unsigned long long)_myAddress.toInt());
  653. }
  654. _readyLock.unlock();
  655. }
  656. fprintf(stderr, "member init done\n");
  657. }
  658. catch (std::exception& e) {
  659. fprintf(stderr, "ERROR: Error initializing member: %s-%s %s\n", networkId.c_str(), memberId.c_str(), e.what());
  660. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  661. exit(-1);
  662. }
  663. }
  664. void CV2::heartbeat()
  665. {
  666. char publicId[1024];
  667. char hostnameTmp[1024];
  668. _myId.toString(false, publicId);
  669. if (gethostname(hostnameTmp, sizeof(hostnameTmp)) != 0) {
  670. hostnameTmp[0] = (char)0;
  671. }
  672. else {
  673. for (int i = 0; i < (int)sizeof(hostnameTmp); ++i) {
  674. if ((hostnameTmp[i] == '.') || (hostnameTmp[i] == 0)) {
  675. hostnameTmp[i] = (char)0;
  676. break;
  677. }
  678. }
  679. }
  680. const char* controllerId = _myAddressStr.c_str();
  681. const char* publicIdentity = publicId;
  682. const char* hostname = hostnameTmp;
  683. while (_run == 1) {
  684. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  685. auto tracer = provider->GetTracer("cv2");
  686. auto span = tracer->StartSpan("cv2::heartbeat");
  687. auto scope = tracer->WithActiveSpan(span);
  688. auto c = _pool->borrow();
  689. int64_t ts = OSUtils::now();
  690. if (c->c) {
  691. std::string major = std::to_string(ZEROTIER_ONE_VERSION_MAJOR);
  692. std::string minor = std::to_string(ZEROTIER_ONE_VERSION_MINOR);
  693. std::string rev = std::to_string(ZEROTIER_ONE_VERSION_REVISION);
  694. std::string version = major + "." + minor + "." + rev;
  695. std::string versionStr = "v" + version;
  696. try {
  697. pqxx::work w { *c->c };
  698. w.exec_params0(
  699. "INSERT INTO controllers_ctl (id, hostname, last_heartbeat, public_identity, version) VALUES "
  700. "($1, $2, TO_TIMESTAMP($3::double precision/1000), $4, $5) "
  701. "ON CONFLICT (id) DO UPDATE SET hostname = EXCLUDED.hostname, last_heartbeat = "
  702. "EXCLUDED.last_heartbeat, "
  703. "public_identity = EXCLUDED.public_identity, version = EXCLUDED.version",
  704. controllerId, hostname, ts, publicIdentity, versionStr);
  705. w.commit();
  706. }
  707. catch (std::exception& e) {
  708. fprintf(stderr, "ERROR: Error in heartbeat: %s\n", e.what());
  709. span->SetStatus(opentelemetry::trace::StatusCode::kError, e.what());
  710. continue;
  711. }
  712. catch (...) {
  713. fprintf(stderr, "ERROR: Unknown error in heartbeat\n");
  714. span->SetStatus(opentelemetry::trace::StatusCode::kError, "Unknown error in heartbeat");
  715. continue;
  716. }
  717. }
  718. _pool->unborrow(c);
  719. span->End();
  720. std::this_thread::sleep_for(std::chrono::seconds(1));
  721. }
  722. fprintf(stderr, "Exited heartbeat thread\n");
  723. }
  724. void CV2::membersDbWatcher()
  725. {
  726. auto c = _pool->borrow();
  727. std::string stream = "member_" + _myAddressStr;
  728. fprintf(stderr, "Listening to member stream: %s\n", stream.c_str());
  729. MemberNotificationReceiver<CV2> m(this, *c->c, stream);
  730. while (_run == 1) {
  731. c->c->await_notification(5, 0);
  732. }
  733. _pool->unborrow(c);
  734. fprintf(stderr, "Exited membersDbWatcher\n");
  735. }
  736. void CV2::networksDbWatcher()
  737. {
  738. std::string stream = "network_" + _myAddressStr;
  739. fprintf(stderr, "Listening to member stream: %s\n", stream.c_str());
  740. auto c = _pool->borrow();
  741. NetworkNotificationReceiver<CV2> n(this, *c->c, stream);
  742. while (_run == 1) {
  743. c->c->await_notification(5, 0);
  744. }
  745. _pool->unborrow(c);
  746. fprintf(stderr, "Exited networksDbWatcher\n");
  747. }
  748. void CV2::commitThread()
  749. {
  750. fprintf(stderr, "%s: commitThread start\n", _myAddressStr.c_str());
  751. std::pair<nlohmann::json, bool> qitem;
  752. while (_commitQueue.get(qitem) && (_run == 1)) {
  753. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  754. auto tracer = provider->GetTracer("cv2");
  755. auto span = tracer->StartSpan("cv2::commitThread");
  756. auto scope = tracer->WithActiveSpan(span);
  757. // fprintf(stderr, "commitThread tick\n");
  758. if (! qitem.first.is_object()) {
  759. fprintf(stderr, "not an object\n");
  760. continue;
  761. }
  762. std::shared_ptr<PostgresConnection> c;
  763. try {
  764. c = _pool->borrow();
  765. }
  766. catch (std::exception& e) {
  767. fprintf(stderr, "ERROR: %s\n", e.what());
  768. continue;
  769. }
  770. if (! c) {
  771. fprintf(stderr, "Error getting database connection\n");
  772. continue;
  773. }
  774. Metrics::pgsql_commit_ticks++;
  775. try {
  776. nlohmann::json& config = (qitem.first);
  777. const std::string objtype = config["objtype"];
  778. if (objtype == "member") {
  779. auto mspan = tracer->StartSpan("cv2::commitThread::member");
  780. auto mscope = tracer->WithActiveSpan(span);
  781. // fprintf(stderr, "%s: commitThread: member\n", _myAddressStr.c_str());
  782. std::string memberId;
  783. std::string networkId;
  784. try {
  785. pqxx::work w(*c->c);
  786. memberId = config["id"];
  787. networkId = config["nwid"];
  788. std::string target = "NULL";
  789. if (! config["remoteTraceTarget"].is_null()) {
  790. target = config["remoteTraceTarget"];
  791. }
  792. pqxx::row nwrow = w.exec_params1("SELECT COUNT(id) FROM networks_ctl WHERE id = $1", networkId);
  793. int nwcount = nwrow[0].as<int>();
  794. if (nwcount != 1) {
  795. fprintf(stderr, "network %s does not exist. skipping member upsert\n", networkId.c_str());
  796. w.abort();
  797. _pool->unborrow(c);
  798. continue;
  799. }
  800. // only needed for hooks, and no hooks for now
  801. // pqxx::row mrow = w.exec_params1("SELECT COUNT(id) FROM device_networks WHERE device_id = $1 AND
  802. // network_id = $2", memberId, networkId); int membercount = mrow[0].as<int>(); bool isNewMember =
  803. // (membercount == 0);
  804. pqxx::result res = w.exec_params0(
  805. "INSERT INTO network_memberships_ctl (device_id, network_id, authorized, active_bridge, "
  806. "ip_assignments, "
  807. "no_auto_assign_ips, sso_exempt, authentication_expiry_time, capabilities, creation_time, "
  808. "identity, last_authorized_time, last_deauthorized_time, "
  809. "remote_trace_level, remote_trace_target, revision, tags, version_major, version_minor, "
  810. "version_revision, version_protocol) "
  811. "VALUES ($1, $2, $3, $4, $5, $6, $7, TO_TIMESTAMP($8::double precision/1000), $9, "
  812. "TO_TIMESTAMP($10::double precision/1000), $11, TO_TIMESTAMP($12::double precision/1000), "
  813. "TO_TIMESTAMP($13::double precision/1000), $14, $15, $16, $17, $18, $19, $20, $21) "
  814. "ON CONFLICT (device_id, network_id) DO UPDATE SET "
  815. "authorized = EXCLUDED.authorized, active_bridge = EXCLUDED.active_bridge, "
  816. "ip_assignments = EXCLUDED.ip_assignments, no_auto_assign_ips = EXCLUDED.no_auto_assign_ips, "
  817. "sso_exempt = EXCLUDED.sso_exempt, authentication_expiry_time = "
  818. "EXCLUDED.authentication_expiry_time, "
  819. "capabilities = EXCLUDED.capabilities, creation_time = EXCLUDED.creation_time, "
  820. "identity = EXCLUDED.identity, last_authorized_time = EXCLUDED.last_authorized_time, "
  821. "last_deauthorized_time = EXCLUDED.last_deauthorized_time, "
  822. "remote_trace_level = EXCLUDED.remote_trace_level, remote_trace_target = "
  823. "EXCLUDED.remote_trace_target, "
  824. "revision = EXCLUDED.revision, tags = EXCLUDED.tags, version_major = EXCLUDED.version_major, "
  825. "version_minor = EXCLUDED.version_minor, version_revision = EXCLUDED.version_revision, "
  826. "version_protocol = EXCLUDED.version_protocol",
  827. memberId, networkId, (bool)config["authorized"], (bool)config["activeBridge"],
  828. config["ipAssignments"].get<std::vector<std::string> >(), (bool)config["noAutoAssignIps"],
  829. (bool)config["ssoExempt"], (uint64_t)config["authenticationExpiryTime"],
  830. OSUtils::jsonDump(config["capabilities"], -1), (uint64_t)config["creationTime"],
  831. OSUtils::jsonString(config["identity"], ""), (uint64_t)config["lastAuthorizedTime"],
  832. (uint64_t)config["lastDeauthorizedTime"], (int)config["remoteTraceLevel"], target,
  833. (uint64_t)config["revision"], OSUtils::jsonDump(config["tags"], -1), (int)config["vMajor"],
  834. (int)config["vMinor"], (int)config["vRev"], (int)config["vProto"]);
  835. w.commit();
  836. // No hooks for now
  837. // if (_smee != NULL && isNewMember) {
  838. // pqxx::row row = w.exec_params1(
  839. // "SELECT "
  840. // " count(h.hook_id) "
  841. // "FROM "
  842. // " ztc_hook h "
  843. // " INNER JOIN ztc_org o ON o.org_id = h.org_id "
  844. // " INNER JOIN ztc_network n ON n.owner_id = o.owner_id "
  845. // " WHERE "
  846. // "n.id = $1 ",
  847. // networkId
  848. // );
  849. // int64_t hookCount = row[0].as<int64_t>();
  850. // if (hookCount > 0) {
  851. // notifyNewMember(networkId, memberId);
  852. // }
  853. // }
  854. const uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  855. const uint64_t memberidInt = OSUtils::jsonIntHex(config["id"], 0ULL);
  856. if (nwidInt && memberidInt) {
  857. nlohmann::json nwOrig;
  858. nlohmann::json memOrig;
  859. nlohmann::json memNew(config);
  860. get(nwidInt, nwOrig, memberidInt, memOrig);
  861. _memberChanged(memOrig, memNew, qitem.second);
  862. }
  863. else {
  864. fprintf(
  865. stderr, "%s: Can't notify of change. Error parsing nwid or memberid: %llu-%llu\n",
  866. _myAddressStr.c_str(), (unsigned long long)nwidInt, (unsigned long long)memberidInt);
  867. }
  868. }
  869. catch (pqxx::data_exception& e) {
  870. std::string cfgDump = OSUtils::jsonDump(config, 2);
  871. fprintf(stderr, "Member save %s-%s: %s\n", networkId.c_str(), memberId.c_str(), cfgDump.c_str());
  872. const pqxx::sql_error* s = dynamic_cast<const pqxx::sql_error*>(&e);
  873. fprintf(stderr, "%s ERROR: Error updating member: %s\n", _myAddressStr.c_str(), e.what());
  874. if (s) {
  875. fprintf(stderr, "%s ERROR: SQL error: %s\n", _myAddressStr.c_str(), s->query().c_str());
  876. }
  877. mspan->SetStatus(opentelemetry::trace::StatusCode::kError, "pqxx::data_exception");
  878. mspan->SetAttribute("error", e.what());
  879. mspan->SetAttribute("config", cfgDump);
  880. }
  881. catch (std::exception& e) {
  882. std::string cfgDump = OSUtils::jsonDump(config, 2);
  883. fprintf(
  884. stderr, "%s ERROR: Error updating member %s-%s: %s\njsonDump: %s\n", _myAddressStr.c_str(),
  885. networkId.c_str(), memberId.c_str(), e.what(), cfgDump.c_str());
  886. mspan->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  887. mspan->SetAttribute("error", e.what());
  888. mspan->SetAttribute("config", cfgDump);
  889. }
  890. }
  891. else if (objtype == "network") {
  892. auto nspan = tracer->StartSpan("cv2::commitThread::network");
  893. auto nscope = tracer->WithActiveSpan(span);
  894. try {
  895. // fprintf(stderr, "%s: commitThread: network\n", _myAddressStr.c_str());
  896. pqxx::work w(*c->c);
  897. std::string id = config["id"];
  898. // network must already exist
  899. pqxx::result res = w.exec_params0(
  900. "INSERT INTO networks_ctl (id, name, configuration, controller_id, revision) "
  901. "VALUES ($1, $2, $3, $4, $5) "
  902. "ON CONFLICT (id) DO UPDATE SET "
  903. "name = EXCLUDED.name, configuration = EXCLUDED.configuration, revision = EXCLUDED.revision+1",
  904. id, OSUtils::jsonString(config["name"], ""), OSUtils::jsonDump(config, -1), _myAddressStr,
  905. ((uint64_t)config["revision"]));
  906. w.commit();
  907. const uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  908. if (nwidInt) {
  909. nlohmann::json nwOrig;
  910. nlohmann::json nwNew(config);
  911. get(nwidInt, nwOrig);
  912. _networkChanged(nwOrig, nwNew, qitem.second);
  913. }
  914. else {
  915. fprintf(
  916. stderr, "%s: Can't notify network changed: %llu\n", _myAddressStr.c_str(),
  917. (unsigned long long)nwidInt);
  918. }
  919. }
  920. catch (pqxx::data_exception& e) {
  921. const pqxx::sql_error* s = dynamic_cast<const pqxx::sql_error*>(&e);
  922. fprintf(stderr, "%s ERROR: Error updating network: %s\n", _myAddressStr.c_str(), e.what());
  923. if (s) {
  924. fprintf(stderr, "%s ERROR: SQL error: %s\n", _myAddressStr.c_str(), s->query().c_str());
  925. }
  926. nspan->SetStatus(opentelemetry::trace::StatusCode::kError, "pqxx::data_exception");
  927. nspan->SetAttribute("error", e.what());
  928. nspan->SetAttribute("config", OSUtils::jsonDump(config, 2));
  929. }
  930. catch (std::exception& e) {
  931. fprintf(stderr, "%s ERROR: Error updating network: %s\n", _myAddressStr.c_str(), e.what());
  932. nspan->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  933. nspan->SetAttribute("error", e.what());
  934. nspan->SetAttribute("config", OSUtils::jsonDump(config, 2));
  935. }
  936. }
  937. else if (objtype == "_delete_network") {
  938. auto dspan = tracer->StartSpan("cv2::commitThread::delete_network");
  939. auto dscope = tracer->WithActiveSpan(span);
  940. // fprintf(stderr, "%s: commitThread: delete network\n", _myAddressStr.c_str());
  941. try {
  942. pqxx::work w(*c->c);
  943. std::string networkId = config["id"];
  944. fprintf(stderr, "Deleting network %s\n", networkId.c_str());
  945. w.exec_params0("DELETE FROM network_memberships_ctl WHERE network_id = $1", networkId);
  946. w.exec_params0("DELETE FROM networks_ctl WHERE id = $1", networkId);
  947. w.commit();
  948. uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  949. json oldConfig;
  950. get(nwidInt, oldConfig);
  951. json empty;
  952. _networkChanged(oldConfig, empty, qitem.second);
  953. }
  954. catch (std::exception& e) {
  955. fprintf(stderr, "%s ERROR: Error deleting network: %s\n", _myAddressStr.c_str(), e.what());
  956. dspan->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  957. dspan->SetAttribute("error", e.what());
  958. dspan->SetAttribute("config", OSUtils::jsonDump(config, 2));
  959. }
  960. }
  961. else if (objtype == "_delete_member") {
  962. auto dspan = tracer->StartSpan("cv2::commitThread::delete_member");
  963. auto dscope = tracer->WithActiveSpan(span);
  964. // fprintf(stderr, "%s commitThread: delete member\n", _myAddressStr.c_str());
  965. try {
  966. pqxx::work w(*c->c);
  967. std::string memberId = config["id"];
  968. std::string networkId = config["nwid"];
  969. pqxx::result res = w.exec_params0(
  970. "DELETE FROM network_memberships_ctl WHERE device_id = $1 AND network_id = $2", memberId,
  971. networkId);
  972. w.commit();
  973. uint64_t nwidInt = OSUtils::jsonIntHex(config["nwid"], 0ULL);
  974. uint64_t memberidInt = OSUtils::jsonIntHex(config["id"], 0ULL);
  975. nlohmann::json networkConfig;
  976. nlohmann::json oldConfig;
  977. get(nwidInt, networkConfig, memberidInt, oldConfig);
  978. json empty;
  979. _memberChanged(oldConfig, empty, qitem.second);
  980. }
  981. catch (std::exception& e) {
  982. fprintf(stderr, "%s ERROR: Error deleting member: %s\n", _myAddressStr.c_str(), e.what());
  983. dspan->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  984. dspan->SetAttribute("error", e.what());
  985. dspan->SetAttribute("config", OSUtils::jsonDump(config, 2));
  986. }
  987. }
  988. else {
  989. fprintf(stderr, "%s ERROR: unknown objtype\n", _myAddressStr.c_str());
  990. }
  991. }
  992. catch (std::exception& e) {
  993. fprintf(stderr, "%s ERROR: Error getting objtype: %s\n", _myAddressStr.c_str(), e.what());
  994. span->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  995. span->SetAttribute("error", e.what());
  996. }
  997. _pool->unborrow(c);
  998. c.reset();
  999. }
  1000. fprintf(stderr, "%s commitThread finished\n", _myAddressStr.c_str());
  1001. }
  1002. void CV2::onlineNotificationThread()
  1003. {
  1004. waitForReady();
  1005. _connected = 1;
  1006. nlohmann::json jtmp1, jtmp2;
  1007. while (_run == 1) {
  1008. auto provider = opentelemetry::trace::Provider::GetTracerProvider();
  1009. auto tracer = provider->GetTracer("cv2");
  1010. auto span = tracer->StartSpan("cv2::onlineNotificationThread");
  1011. auto scope = tracer->WithActiveSpan(span);
  1012. auto c = _pool->borrow();
  1013. auto c2 = _pool->borrow();
  1014. try {
  1015. fprintf(stderr, "%s onlineNotificationThread\n", _myAddressStr.c_str());
  1016. std::unordered_map<std::pair<uint64_t, uint64_t>, NodeOnlineRecord, _PairHasher> lastOnline;
  1017. {
  1018. std::lock_guard<std::mutex> l(_lastOnline_l);
  1019. lastOnline.swap(_lastOnline);
  1020. }
  1021. pqxx::work w(*c->c);
  1022. pqxx::work w2(*c2->c);
  1023. bool firstRun = true;
  1024. bool memberAdded = false;
  1025. uint64_t updateCount = 0;
  1026. pqxx::pipeline pipe(w);
  1027. for (auto i = lastOnline.begin(); i != lastOnline.end(); ++i) {
  1028. updateCount++;
  1029. uint64_t nwid_i = i->first.first;
  1030. char nwidTmp[64];
  1031. char memTmp[64];
  1032. char ipTmp[64];
  1033. OSUtils::ztsnprintf(nwidTmp, sizeof(nwidTmp), "%.16llx", nwid_i);
  1034. OSUtils::ztsnprintf(memTmp, sizeof(memTmp), "%.10llx", i->first.second);
  1035. if (! get(nwid_i, jtmp1, i->first.second, jtmp2)) {
  1036. continue; // skip non existent networks/members
  1037. }
  1038. std::string networkId(nwidTmp);
  1039. std::string memberId(memTmp);
  1040. try {
  1041. pqxx::row r = w2.exec_params1(
  1042. "SELECT device_id, network_id FROM network_memberships_ctl WHERE network_id = $1 AND device_id "
  1043. "= $2",
  1044. networkId, memberId);
  1045. }
  1046. catch (pqxx::unexpected_rows& e) {
  1047. continue;
  1048. }
  1049. int64_t ts = i->second.lastSeen;
  1050. std::string ipAddr = i->second.physicalAddress.toIpString(ipTmp);
  1051. std::string timestamp = std::to_string(ts);
  1052. std::string osArch = i->second.osArch;
  1053. std::vector<std::string> osArchSplit = split(osArch, '/');
  1054. std::string os = osArchSplit[0];
  1055. std::string arch = osArchSplit[1];
  1056. if (ipAddr.empty()) {
  1057. ipAddr = "relayed";
  1058. }
  1059. json record = {
  1060. { ipAddr, ts },
  1061. };
  1062. std::string device_network_insert =
  1063. "INSERT INTO network_memberships_ctl (device_id, network_id, last_seen, os, arch) "
  1064. "VALUES ('"
  1065. + w2.esc(memberId) + "', '" + w2.esc(networkId) + "', '" + w2.esc(record.dump())
  1066. + "'::JSONB, "
  1067. "'"
  1068. + w2.esc(os) + "', '" + w2.esc(arch)
  1069. + "') "
  1070. "ON CONFLICT (device_id, network_id) DO UPDATE SET os = EXCLUDED.os, arch = EXCLUDED.arch, "
  1071. "last_seen = network_memberships_ctl.last_seen || EXCLUDED.last_seen";
  1072. pipe.insert(device_network_insert);
  1073. Metrics::pgsql_node_checkin++;
  1074. }
  1075. pipe.complete();
  1076. ;
  1077. w2.commit();
  1078. w.commit();
  1079. fprintf(stderr, "%s: Updated online status of %lu members\n", _myAddressStr.c_str(), updateCount);
  1080. }
  1081. catch (std::exception& e) {
  1082. fprintf(stderr, "%s ERROR: Error in onlineNotificationThread: %s\n", _myAddressStr.c_str(), e.what());
  1083. span->SetStatus(opentelemetry::trace::StatusCode::kError, "std::exception");
  1084. span->SetAttribute("error", e.what());
  1085. }
  1086. catch (...) {
  1087. fprintf(stderr, "%s ERROR: Unknown error in onlineNotificationThread\n", _myAddressStr.c_str());
  1088. span->SetStatus(opentelemetry::trace::StatusCode::kError, "unknown");
  1089. }
  1090. _pool->unborrow(c2);
  1091. _pool->unborrow(c);
  1092. span->End();
  1093. std::this_thread::sleep_for(std::chrono::seconds(10));
  1094. }
  1095. fprintf(stderr, "%s: Fell out of run loop in onlineNotificationThread\n", _myAddressStr.c_str());
  1096. if (_run == 1) {
  1097. fprintf(
  1098. stderr, "ERROR: %s onlineNotificationThread should still be running! Exiting Controller.\n",
  1099. _myAddressStr.c_str());
  1100. exit(6);
  1101. }
  1102. }
  1103. #endif // ZT_CONTROLLER_USE_LIBPQ