2 * This file is part of PowerDNS or dnsdist.
3 * Copyright -- PowerDNS.COM B.V. and its contributors
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of version 2 of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * In addition, for the avoidance of any doubt, permission is granted to
10 * link this program with OpenSSL and to (re)distribute the binaries
11 * produced as the result of such linking.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 #include "ext/luawrapper/include/LuaContext.hpp"
32 #include <unordered_map>
34 #include <boost/variant.hpp>
36 #include "bpf-filter.hh"
37 #include "capabilities.hh"
38 #include "circular_buffer.hh"
39 #include "dnscrypt.hh"
40 #include "dnsdist-cache.hh"
41 #include "dnsdist-dynbpf.hh"
44 #include "ednsoptions.hh"
50 #include "tcpiohandler.hh"
51 #include "uuid-utils.hh"
53 void carbonDumpThread();
54 uint64_t uptimeOfProcess(const std::string& str);
56 extern uint16_t g_ECSSourcePrefixV4;
57 extern uint16_t g_ECSSourcePrefixV6;
58 extern bool g_ECSOverride;
60 typedef std::unordered_map<string, string> QTag;
64 DNSQuestion(const DNSName* name, uint16_t type, uint16_t class_, unsigned int consumed_, const ComboAddress* lc, const ComboAddress* rem, struct dnsheader* header, size_t bufferSize, uint16_t queryLen, bool isTcp, const struct timespec* queryTime_):
65 qname(name), local(lc), remote(rem), dh(header), queryTime(queryTime_), size(bufferSize), consumed(consumed_), tempFailureTTL(boost::none), qtype(type), qclass(class_), len(queryLen), ecsPrefixLength(rem->sin4.sin_family == AF_INET ? g_ECSSourcePrefixV4 : g_ECSSourcePrefixV6), tcp(isTcp), ecsOverride(g_ECSOverride) {
66 const uint16_t* flags = getFlagsFromDNSHeader(dh);
69 DNSQuestion(const DNSQuestion&) = delete;
70 DNSQuestion& operator=(const DNSQuestion&) = delete;
71 DNSQuestion(DNSQuestion&&) = default;
74 boost::optional<boost::uuids::uuid> uniqueId;
77 boost::optional<Netmask> subnet;
78 std::string sni; /* Server Name Indication, if any (DoT or DoH) */
80 const DNSName* qname{nullptr};
81 const ComboAddress* local{nullptr};
82 const ComboAddress* remote{nullptr};
83 std::shared_ptr<QTag> qTag{nullptr};
84 std::shared_ptr<std::map<uint16_t, EDNSOptionView> > ednsOptions;
85 std::shared_ptr<DNSCryptQuery> dnsCryptQuery{nullptr};
86 std::shared_ptr<DNSDistPacketCache> packetCache{nullptr};
87 struct dnsheader* dh{nullptr};
88 const struct timespec* queryTime{nullptr};
89 struct DOHUnit* du{nullptr};
91 unsigned int consumed{0};
93 boost::optional<uint32_t> tempFailureTTL;
94 uint32_t cacheKeyNoECS;
97 const uint16_t qclass;
99 uint16_t ecsPrefixLength;
101 uint8_t ednsRCode{0};
103 bool skipCache{false};
108 bool ecsAdded{false};
109 bool ednsAdded{false};
110 bool useZeroScope{false};
111 bool dnssecOK{false};
114 struct DNSResponse : DNSQuestion
116 DNSResponse(const DNSName* name, uint16_t type, uint16_t class_, unsigned int consumed, const ComboAddress* lc, const ComboAddress* rem, struct dnsheader* header, size_t bufferSize, uint16_t responseLen, bool isTcp, const struct timespec* queryTime_):
117 DNSQuestion(name, type, class_, consumed, lc, rem, header, bufferSize, responseLen, isTcp, queryTime_) { }
118 DNSResponse(const DNSResponse&) = delete;
119 DNSResponse& operator=(const DNSResponse&) = delete;
120 DNSResponse(DNSResponse&&) = default;
123 /* so what could you do:
126 provide actual answer,
127 allow & and stop processing,
129 modify header: (servfail|refused|notimp), set TC=1,
135 enum class Action { Drop, Nxdomain, Refused, Spoof, Allow, HeaderModify, Pool, Delay, Truncate, ServFail, None, NoOp, NoRecurse };
136 static std::string typeToString(const Action& action)
141 case Action::Nxdomain:
142 return "Send NXDomain";
143 case Action::Refused:
144 return "Send Refused";
146 return "Spoof an answer";
149 case Action::HeaderModify:
150 return "Modify the header";
152 return "Route to a pool";
155 case Action::Truncate:
156 return "Truncate over UDP";
157 case Action::ServFail:
158 return "Send ServFail";
162 case Action::NoRecurse:
169 virtual Action operator()(DNSQuestion*, string* ruleresult) const =0;
173 virtual string toString() const = 0;
174 virtual std::map<string, double> getStats() const
180 class DNSResponseAction
183 enum class Action { Allow, Delay, Drop, HeaderModify, ServFail, None };
184 virtual Action operator()(DNSResponse*, string* ruleresult) const =0;
185 virtual ~DNSResponseAction()
188 virtual string toString() const = 0;
193 DynBlock(): action(DNSAction::Action::None), warning(false)
197 DynBlock(const std::string& reason_, const struct timespec& until_, const DNSName& domain_, DNSAction::Action action_): reason(reason_), until(until_), domain(domain_), action(action_), warning(false)
201 DynBlock(const DynBlock& rhs): reason(rhs.reason), until(rhs.until), domain(rhs.domain), action(rhs.action), warning(rhs.warning)
203 blocks.store(rhs.blocks);
206 DynBlock& operator=(const DynBlock& rhs)
212 blocks.store(rhs.blocks);
218 struct timespec until;
220 DNSAction::Action action;
221 mutable std::atomic<unsigned int> blocks;
225 extern GlobalStateHolder<NetmaskTree<DynBlock>> g_dynblockNMG;
227 extern vector<pair<struct timeval, std::string> > g_confDelta;
229 extern uint64_t getLatencyCount(const std::string&);
233 using stat_t=std::atomic<uint64_t>; // aww yiss ;-)
235 stat_t servfailResponses{0};
237 stat_t frontendNXDomain{0};
238 stat_t frontendServFail{0};
239 stat_t frontendNoError{0};
240 stat_t nonCompliantQueries{0};
241 stat_t nonCompliantResponses{0};
243 stat_t emptyQueries{0};
245 stat_t dynBlocked{0};
247 stat_t ruleNXDomain{0};
248 stat_t ruleRefused{0};
249 stat_t ruleServFail{0};
250 stat_t selfAnswered{0};
251 stat_t downstreamTimeouts{0};
252 stat_t downstreamSendErrors{0};
256 stat_t cacheMisses{0};
257 stat_t latency0_1{0}, latency1_10{0}, latency10_50{0}, latency50_100{0}, latency100_1000{0}, latencySlow{0}, latencySum{0};
258 stat_t securityStatus{0};
260 double latencyAvg100{0}, latencyAvg1000{0}, latencyAvg10000{0}, latencyAvg1000000{0};
261 typedef std::function<uint64_t(const std::string&)> statfunction_t;
262 typedef boost::variant<stat_t*, double*, statfunction_t> entry_t;
263 std::vector<std::pair<std::string, entry_t>> entries{
264 {"responses", &responses},
265 {"servfail-responses", &servfailResponses},
266 {"queries", &queries},
267 {"frontend-nxdomain", &frontendNXDomain},
268 {"frontend-servfail", &frontendServFail},
269 {"frontend-noerror", &frontendNoError},
270 {"acl-drops", &aclDrops},
271 {"rule-drop", &ruleDrop},
272 {"rule-nxdomain", &ruleNXDomain},
273 {"rule-refused", &ruleRefused},
274 {"rule-servfail", &ruleServFail},
275 {"self-answered", &selfAnswered},
276 {"downstream-timeouts", &downstreamTimeouts},
277 {"downstream-send-errors", &downstreamSendErrors},
278 {"trunc-failures", &truncFail},
279 {"no-policy", &noPolicy},
280 {"latency0-1", &latency0_1},
281 {"latency1-10", &latency1_10},
282 {"latency10-50", &latency10_50},
283 {"latency50-100", &latency50_100},
284 {"latency100-1000", &latency100_1000},
285 {"latency-slow", &latencySlow},
286 {"latency-avg100", &latencyAvg100},
287 {"latency-avg1000", &latencyAvg1000},
288 {"latency-avg10000", &latencyAvg10000},
289 {"latency-avg1000000", &latencyAvg1000000},
290 {"uptime", uptimeOfProcess},
291 {"real-memory-usage", getRealMemoryUsage},
292 {"special-memory-usage", getSpecialMemoryUsage},
293 {"noncompliant-queries", &nonCompliantQueries},
294 {"noncompliant-responses", &nonCompliantResponses},
295 {"rdqueries", &rdQueries},
296 {"empty-queries", &emptyQueries},
297 {"cache-hits", &cacheHits},
298 {"cache-misses", &cacheMisses},
299 {"cpu-user-msec", getCPUTimeUser},
300 {"cpu-sys-msec", getCPUTimeSystem},
301 {"fd-usage", getOpenFileDescriptors},
302 {"dyn-blocked", &dynBlocked},
303 {"dyn-block-nmg-size", [](const std::string&) { return g_dynblockNMG.getLocal()->size(); }},
304 {"security-status", &securityStatus},
306 {"latency-sum", &latencySum},
307 {"latency-count", getLatencyCount},
311 // Metric types for Prometheus
312 enum class PrometheusMetricType: int {
317 // Keeps additional information about metrics
318 struct MetricDefinition {
319 MetricDefinition(PrometheusMetricType _prometheusType, const std::string& _description): description(_description), prometheusType(_prometheusType) {
322 MetricDefinition() = default;
324 // Metric description
325 std::string description;
326 // Metric type for Prometheus
327 PrometheusMetricType prometheusType;
330 struct MetricDefinitionStorage {
331 // Return metric definition by name
332 bool getMetricDetails(std::string metricName, MetricDefinition& metric) {
333 auto metricDetailsIter = metrics.find(metricName);
335 if (metricDetailsIter == metrics.end()) {
339 metric = metricDetailsIter->second;
343 // Return string representation of Prometheus metric type
344 std::string getPrometheusStringMetricType(PrometheusMetricType metricType) {
345 switch (metricType) {
346 case PrometheusMetricType::counter:
349 case PrometheusMetricType::gauge:
358 std::map<std::string, MetricDefinition> metrics = {
359 { "responses", MetricDefinition(PrometheusMetricType::counter, "Number of responses received from backends") },
360 { "servfail-responses", MetricDefinition(PrometheusMetricType::counter, "Number of SERVFAIL answers received from backends") },
361 { "queries", MetricDefinition(PrometheusMetricType::counter, "Number of received queries")},
362 { "frontend-nxdomain", MetricDefinition(PrometheusMetricType::counter, "Number of NXDomain answers sent to clients")},
363 { "frontend-servfail", MetricDefinition(PrometheusMetricType::counter, "Number of SERVFAIL answers sent to clients")},
364 { "frontend-noerror", MetricDefinition(PrometheusMetricType::counter, "Number of NoError answers sent to clients")},
365 { "acl-drops", MetricDefinition(PrometheusMetricType::counter, "Number of packets dropped because of the ACL")},
366 { "rule-drop", MetricDefinition(PrometheusMetricType::counter, "Number of queries dropped because of a rule")},
367 { "rule-nxdomain", MetricDefinition(PrometheusMetricType::counter, "Number of NXDomain answers returned because of a rule")},
368 { "rule-refused", MetricDefinition(PrometheusMetricType::counter, "Number of Refused answers returned because of a rule")},
369 { "rule-servfail", MetricDefinition(PrometheusMetricType::counter, "Number of SERVFAIL answers received because of a rule")},
370 { "self-answered", MetricDefinition(PrometheusMetricType::counter, "Number of self-answered responses")},
371 { "downstream-timeouts", MetricDefinition(PrometheusMetricType::counter, "Number of queries not answered in time by a backend")},
372 { "downstream-send-errors", MetricDefinition(PrometheusMetricType::counter, "Number of errors when sending a query to a backend")},
373 { "trunc-failures", MetricDefinition(PrometheusMetricType::counter, "Number of errors encountered while truncating an answer")},
374 { "no-policy", MetricDefinition(PrometheusMetricType::counter, "Number of queries dropped because no server was available")},
375 { "latency0-1", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in less than 1ms")},
376 { "latency1-10", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in 1-10 ms")},
377 { "latency10-50", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in 10-50 ms")},
378 { "latency50-100", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in 50-100 ms")},
379 { "latency100-1000", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in 100-1000 ms")},
380 { "latency-slow", MetricDefinition(PrometheusMetricType::counter, "Number of queries answered in more than 1 second")},
381 { "latency-avg100", MetricDefinition(PrometheusMetricType::gauge, "Average response latency in microseconds of the last 100 packets")},
382 { "latency-avg1000", MetricDefinition(PrometheusMetricType::gauge, "Average response latency in microseconds of the last 1000 packets")},
383 { "latency-avg10000", MetricDefinition(PrometheusMetricType::gauge, "Average response latency in microseconds of the last 10000 packets")},
384 { "latency-avg1000000", MetricDefinition(PrometheusMetricType::gauge, "Average response latency in microseconds of the last 1000000 packets")},
385 { "uptime", MetricDefinition(PrometheusMetricType::gauge, "Uptime of the dnsdist process in seconds")},
386 { "real-memory-usage", MetricDefinition(PrometheusMetricType::gauge, "Current memory usage in bytes")},
387 { "noncompliant-queries", MetricDefinition(PrometheusMetricType::counter, "Number of queries dropped as non-compliant")},
388 { "noncompliant-responses", MetricDefinition(PrometheusMetricType::counter, "Number of answers from a backend dropped as non-compliant")},
389 { "rdqueries", MetricDefinition(PrometheusMetricType::counter, "Number of received queries with the recursion desired bit set")},
390 { "empty-queries", MetricDefinition(PrometheusMetricType::counter, "Number of empty queries received from clients")},
391 { "cache-hits", MetricDefinition(PrometheusMetricType::counter, "Number of times an answer was retrieved from cache")},
392 { "cache-misses", MetricDefinition(PrometheusMetricType::counter, "Number of times an answer not found in the cache")},
393 { "cpu-user-msec", MetricDefinition(PrometheusMetricType::counter, "Milliseconds spent by dnsdist in the user state")},
394 { "cpu-sys-msec", MetricDefinition(PrometheusMetricType::counter, "Milliseconds spent by dnsdist in the system state")},
395 { "fd-usage", MetricDefinition(PrometheusMetricType::gauge, "Number of currently used file descriptors")},
396 { "dyn-blocked", MetricDefinition(PrometheusMetricType::counter, "Number of queries dropped because of a dynamic block")},
397 { "dyn-block-nmg-size", MetricDefinition(PrometheusMetricType::gauge, "Number of dynamic blocks entries") },
398 { "security-status", MetricDefinition(PrometheusMetricType::gauge, "Security status of this software. 0=unknown, 1=OK, 2=upgrade recommended, 3=upgrade mandatory") },
402 extern MetricDefinitionStorage g_metricDefinitions;
403 extern struct DNSDistStats g_stats;
404 void doLatencyStats(double udiff);
409 StopWatch(bool realTime=false): d_needRealTime(realTime)
412 struct timespec d_start{0,0};
413 bool d_needRealTime{false};
416 if(gettime(&d_start, d_needRealTime) < 0)
417 unixDie("Getting timestamp");
421 void set(const struct timespec& from) {
425 double udiff() const {
427 if(gettime(&now, d_needRealTime) < 0)
428 unixDie("Getting timestamp");
430 return 1000000.0*(now.tv_sec - d_start.tv_sec) + (now.tv_nsec - d_start.tv_nsec)/1000.0;
433 double udiffAndSet() {
435 if(gettime(&now, d_needRealTime) < 0)
436 unixDie("Getting timestamp");
438 auto ret= 1000000.0*(now.tv_sec - d_start.tv_sec) + (now.tv_nsec - d_start.tv_nsec)/1000.0;
445 class BasicQPSLimiter
452 BasicQPSLimiter(unsigned int burst): d_tokens(burst)
457 bool check(unsigned int rate, unsigned int burst) const // this is not quite fair
459 auto delta = d_prev.udiffAndSet();
461 if(delta > 0.0) // time, frequently, does go backwards..
462 d_tokens += 1.0 * rate * (delta/1000000.0);
464 if(d_tokens > burst) {
469 if(d_tokens >= 1.0) { // we need this because burst=1 is weird otherwise
477 bool seenSince(const struct timespec& cutOff) const
479 return cutOff < d_prev.d_start;
483 mutable StopWatch d_prev;
484 mutable double d_tokens;
487 class QPSLimiter : public BasicQPSLimiter
490 QPSLimiter(): BasicQPSLimiter()
494 QPSLimiter(unsigned int rate, unsigned int burst): BasicQPSLimiter(burst), d_rate(rate), d_burst(burst), d_passthrough(false)
499 unsigned int getRate() const
501 return d_passthrough ? 0 : d_rate;
504 int getPassed() const
509 int getBlocked() const
514 bool check() const // this is not quite fair
520 bool ret = BasicQPSLimiter::check(d_rate, d_burst);
531 mutable unsigned int d_passed{0};
532 mutable unsigned int d_blocked{0};
534 unsigned int d_burst;
535 bool d_passthrough{true};
542 IDState(): sentTime(true), delayMsec(0), tempFailureTTL(boost::none) { origDest.sin4.sin_family = 0;}
543 IDState(const IDState& orig): origRemote(orig.origRemote), origDest(orig.origDest), age(orig.age)
545 usageIndicator.store(orig.usageIndicator.load());
546 origFD = orig.origFD;
547 origID = orig.origID;
548 delayMsec = orig.delayMsec;
549 tempFailureTTL = orig.tempFailureTTL;
552 static const int64_t unusedIndicator = -1;
554 static bool isInUse(int64_t usageIndicator)
556 return usageIndicator != unusedIndicator;
561 return usageIndicator != unusedIndicator;
564 /* return true if the value has been successfully replaced meaning that
565 no-one updated the usage indicator in the meantime */
566 bool tryMarkUnused(int64_t expectedUsageIndicator)
568 return usageIndicator.compare_exchange_strong(expectedUsageIndicator, unusedIndicator);
571 /* mark as unused no matter what, return true if the state was in use before */
574 auto currentGeneration = generation++;
575 return markAsUsed(currentGeneration);
578 /* mark as unused no matter what, return true if the state was in use before */
579 bool markAsUsed(int64_t currentGeneration)
581 int64_t oldUsage = usageIndicator.exchange(currentGeneration);
582 return oldUsage != unusedIndicator;
585 /* We use this value to detect whether this state is in use.
586 For performance reasons we don't want to use a lock here, but that means
587 we need to be very careful when modifying this value. Modifications happen
589 - one of the UDP or DoH 'client' threads receiving a query, selecting a backend
590 then picking one of the states associated to this backend (via the idOffset).
591 Most of the time this state should not be in use and usageIndicator is -1, but we
592 might not yet have received a response for the query previously associated to this
593 state, meaning that we will 'reuse' this state and erase the existing state.
594 If we ever receive a response for this state, it will be discarded. This is
595 mostly fine for UDP except that we still need to be careful in order to miss
596 the 'outstanding' counters, which should only be increased when we are picking
597 an empty state, and not when reusing ;
598 For DoH, though, we have dynamically allocated a DOHUnit object that needs to
599 be freed, as well as internal objects internals to libh2o.
600 - one of the UDP receiver threads receiving a response from a backend, picking
601 the corresponding state and sending the response to the client ;
602 - the 'healthcheck' thread scanning the states to actively discover timeouts,
603 mostly to keep some counters like the 'outstanding' one sane.
604 We previously based that logic on the origFD (FD on which the query was received,
605 and therefore from where the response should be sent) but this suffered from an
606 ABA problem since it was quite likely that a UDP 'client thread' would reset it to the
607 same value since we only have so much incoming sockets:
608 - 1/ 'client' thread gets a query and set origFD to its FD, say 5 ;
609 - 2/ 'receiver' thread gets a response, read the value of origFD to 5, check that the qname,
610 qtype and qclass match
611 - 3/ during that time the 'client' thread reuses the state, setting again origFD to 5 ;
612 - 4/ the 'receiver' thread uses compare_exchange_strong() to only replace the value if it's still
613 5, except it's not the same 5 anymore and it overrides a fresh state.
614 We now use a 32-bit unsigned counter instead, which is incremented every time the state is set,
615 wrapping around if necessary, and we set an atomic signed 64-bit value, so that we still have -1
616 when the state is unused and the value of our counter otherwise.
618 std::atomic<int64_t> usageIndicator{unusedIndicator}; // set to unusedIndicator to indicate this state is empty // 8
619 std::atomic<uint32_t> generation{0}; // increased every time a state is used, to be able to detect an ABA issue // 4
620 ComboAddress origRemote; // 28
621 ComboAddress origDest; // 28
622 StopWatch sentTime; // 16
624 std::shared_ptr<DNSCryptQuery> dnsCryptQuery{nullptr};
626 boost::optional<boost::uuids::uuid> uniqueId;
628 boost::optional<Netmask> subnet{boost::none};
629 std::shared_ptr<DNSDistPacketCache> packetCache{nullptr};
630 std::shared_ptr<QTag> qTag{nullptr};
631 const ClientState* cs{nullptr};
632 DOHUnit* du{nullptr};
633 uint32_t cacheKey; // 4
634 uint32_t cacheKeyNoECS; // 4
637 uint16_t qclass; // 2
638 uint16_t origID; // 2
639 uint16_t origFlags; // 2
642 boost::optional<uint32_t> tempFailureTTL;
643 bool ednsAdded{false};
644 bool ecsAdded{false};
645 bool skipCache{false};
646 bool destHarvested{false}; // if true, origDest holds the original dest addr, otherwise the listening addr
647 bool dnssecOK{false};
651 typedef std::unordered_map<string, unsigned int> QueryCountRecords;
652 typedef std::function<std::tuple<bool, string>(const DNSQuestion* dq)> QueryCountFilter;
656 pthread_rwlock_init(&queryLock, nullptr);
658 QueryCountRecords records;
659 QueryCountFilter filter;
660 pthread_rwlock_t queryLock;
664 extern QueryCount g_qcount;
668 ClientState(const ComboAddress& local_, bool isTCP, bool doReusePort, int fastOpenQueue, const std::string& itfName, const std::set<int>& cpus_): cpus(cpus_), local(local_), interface(itfName), fastOpenQueueSize(fastOpenQueue), tcp(isTCP), reuseport(doReusePort)
674 std::shared_ptr<DNSCryptContext> dnscryptCtx{nullptr};
675 std::shared_ptr<TLSFrontend> tlsFrontend{nullptr};
676 std::shared_ptr<DOHFrontend> dohFrontend{nullptr};
677 std::string interface;
678 std::atomic<uint64_t> queries{0};
679 mutable std::atomic<uint64_t> responses{0};
680 std::atomic<uint64_t> tcpDiedReadingQuery{0};
681 std::atomic<uint64_t> tcpDiedSendingResponse{0};
682 std::atomic<uint64_t> tcpGaveUp{0};
683 std::atomic<uint64_t> tcpClientTimeouts{0};
684 std::atomic<uint64_t> tcpDownstreamTimeouts{0};
685 std::atomic<uint64_t> tcpCurrentConnections{0};
686 std::atomic<uint64_t> tlsNewSessions{0}; // A new TLS session has been negotiated, no resumption
687 std::atomic<uint64_t> tlsResumptions{0}; // A TLS session has been resumed, either via session id or via a TLS ticket
689 std::atomic<double> tcpAvgQueriesPerConnection{0.0};
691 std::atomic<double> tcpAvgConnectionDuration{0.0};
694 int fastOpenQueueSize{0};
700 int getSocket() const
702 return udpFD != -1 ? udpFD : tcpFD;
715 std::string getType() const
717 std::string result = udpFD != -1 ? "UDP" : "TCP";
720 result += " (DNS over HTTPS)";
722 else if (tlsFrontend) {
723 result += " (DNS over TLS)";
725 else if (dnscryptCtx) {
726 result += " (DNSCrypt)";
733 shared_ptr<BPFFilter> d_filter;
738 d_filter->removeSocket(getSocket());
743 void attachFilter(shared_ptr<BPFFilter> bpf)
747 bpf->addSocket(getSocket());
750 #endif /* HAVE_EBPF */
752 void updateTCPMetrics(size_t queries, uint64_t durationMs)
754 tcpAvgQueriesPerConnection = (99.0 * tcpAvgQueriesPerConnection / 100.0) + (queries / 100.0);
755 tcpAvgConnectionDuration = (99.0 * tcpAvgConnectionDuration / 100.0) + (durationMs / 100.0);
759 class TCPClientCollection {
760 std::vector<int> d_tcpclientthreads;
761 std::atomic<uint64_t> d_numthreads{0};
762 std::atomic<uint64_t> d_pos{0};
763 std::atomic<uint64_t> d_queued{0};
764 const uint64_t d_maxthreads{0};
767 const bool d_useSinglePipe;
770 TCPClientCollection(size_t maxThreads, bool useSinglePipe=false): d_maxthreads(maxThreads), d_singlePipe{-1,-1}, d_useSinglePipe(useSinglePipe)
773 d_tcpclientthreads.reserve(maxThreads);
775 if (d_useSinglePipe) {
776 if (pipe(d_singlePipe) < 0) {
778 throw std::runtime_error("Error creating the TCP single communication pipe: " + stringerror(err));
781 if (!setNonBlocking(d_singlePipe[0])) {
783 close(d_singlePipe[0]);
784 close(d_singlePipe[1]);
785 throw std::runtime_error("Error setting the TCP single communication pipe non-blocking: " + stringerror(err));
788 if (!setNonBlocking(d_singlePipe[1])) {
790 close(d_singlePipe[0]);
791 close(d_singlePipe[1]);
792 throw std::runtime_error("Error setting the TCP single communication pipe non-blocking: " + stringerror(err));
798 uint64_t pos = d_pos++;
800 return d_tcpclientthreads[pos % d_numthreads];
802 bool hasReachedMaxThreads() const
804 return d_numthreads >= d_maxthreads;
806 uint64_t getThreadsCount() const
810 uint64_t getQueuedCount() const
814 void decrementQueuedCount()
818 void addTCPClientThread();
821 extern std::unique_ptr<TCPClientCollection> g_tcpclientthreads;
823 struct DownstreamState
825 typedef std::function<std::tuple<DNSName, uint16_t, uint16_t>(const DNSName&, uint16_t, uint16_t, dnsheader*)> checkfunc_t;
827 DownstreamState(const ComboAddress& remote_, const ComboAddress& sourceAddr_, unsigned int sourceItf, size_t numberOfSockets);
828 DownstreamState(const ComboAddress& remote_): DownstreamState(remote_, ComboAddress(), 0, 1) {}
831 for (auto& fd : sockets) {
838 boost::uuids::uuid id;
839 std::set<unsigned int> hashes;
840 mutable pthread_rwlock_t d_lock;
841 std::vector<int> sockets;
842 std::mutex socketsLock;
843 std::mutex connectLock;
844 std::unique_ptr<FDMultiplexer> mplexer{nullptr};
846 const ComboAddress remote;
848 vector<IDState> idStates;
849 const ComboAddress sourceAddr;
850 checkfunc_t checkFunction;
851 DNSName checkName{"a.root-servers.net."};
852 QType checkType{QType::A};
853 uint16_t checkClass{QClass::IN};
854 std::atomic<uint64_t> idOffset{0};
855 std::atomic<uint64_t> sendErrors{0};
856 std::atomic<uint64_t> outstanding{0};
857 std::atomic<uint64_t> reuseds{0};
858 std::atomic<uint64_t> queries{0};
859 std::atomic<uint64_t> responses{0};
861 std::atomic<uint64_t> sendErrors{0};
862 std::atomic<uint64_t> reuseds{0};
863 std::atomic<uint64_t> queries{0};
865 std::atomic<uint64_t> tcpDiedSendingQuery{0};
866 std::atomic<uint64_t> tcpDiedReadingResponse{0};
867 std::atomic<uint64_t> tcpGaveUp{0};
868 std::atomic<uint64_t> tcpReadTimeouts{0};
869 std::atomic<uint64_t> tcpWriteTimeouts{0};
870 std::atomic<uint64_t> tcpCurrentConnections{0};
871 std::atomic<double> tcpAvgQueriesPerConnection{0.0};
873 std::atomic<double> tcpAvgConnectionDuration{0.0};
875 size_t socketsOffset{0};
876 double queryLoad{0.0};
877 double dropRate{0.0};
878 double latencyUsec{0.0};
881 int tcpConnectTimeout{5};
882 int tcpRecvTimeout{30};
883 int tcpSendTimeout{30};
884 unsigned int checkInterval{1};
885 unsigned int lastCheck{0};
886 const unsigned int sourceItf{0};
888 uint16_t xpfRRCode{0};
889 uint16_t checkTimeout{1000}; /* in milliseconds */
890 uint8_t currentCheckFailures{0};
891 uint8_t consecutiveSuccessfulChecks{0};
892 uint8_t maxCheckFailures{1};
893 uint8_t minRiseSuccesses{1};
896 enum class Availability { Up, Down, Auto} availability{Availability::Auto};
897 bool mustResolve{false};
898 bool upStatus{false};
901 bool disableZeroScope{false};
902 std::atomic<bool> connected{false};
903 std::atomic_flag threadStarted;
904 bool tcpFastOpen{false};
905 bool ipBindAddrNoPort{true};
909 if(availability == Availability::Down)
911 if(availability == Availability::Up)
915 void setUp() { availability = Availability::Up; }
916 void setDown() { availability = Availability::Down; }
917 void setAuto() { availability = Availability::Auto; }
918 string getName() const {
920 return remote.toStringWithPort();
924 string getNameWithAddr() const {
926 return remote.toStringWithPort();
928 return name + " (" + remote.toStringWithPort()+ ")";
930 string getStatus() const
933 if(availability == DownstreamState::Availability::Up)
935 else if(availability == DownstreamState::Availability::Down)
938 status = (upStatus ? "up" : "down");
943 void setId(const boost::uuids::uuid& newId);
944 void setWeight(int newWeight);
946 void updateTCPMetrics(size_t queries, uint64_t durationMs)
948 tcpAvgQueriesPerConnection = (99.0 * tcpAvgQueriesPerConnection / 100.0) + (queries / 100.0);
949 tcpAvgConnectionDuration = (99.0 * tcpAvgConnectionDuration / 100.0) + (durationMs / 100.0);
952 using servers_t =vector<std::shared_ptr<DownstreamState>>;
954 template <class T> using NumberedVector = std::vector<std::pair<unsigned int, T> >;
956 void responderThread(std::shared_ptr<DownstreamState> state);
957 extern std::mutex g_luamutex;
958 extern LuaContext g_lua;
959 extern std::string g_outputBuffer; // locking for this is ok, as locked by g_luamutex
967 virtual bool matches(const DNSQuestion* dq) const =0;
968 virtual string toString() const = 0;
969 mutable std::atomic<uint64_t> d_matches{0};
972 using NumberedServerVector = NumberedVector<shared_ptr<DownstreamState>>;
973 typedef std::function<shared_ptr<DownstreamState>(const NumberedServerVector& servers, const DNSQuestion*)> policyfunc_t;
980 std::string toString() const {
981 return string("ServerPolicy") + (isLua ? " (Lua)" : "") + " \"" + name + "\"";
989 pthread_rwlock_init(&d_lock, nullptr);
992 const std::shared_ptr<DNSDistPacketCache> getCache() const { return packetCache; };
999 void setECS(bool useECS)
1004 std::shared_ptr<DNSDistPacketCache> packetCache{nullptr};
1005 std::shared_ptr<ServerPolicy> policy{nullptr};
1007 size_t countServers(bool upOnly)
1010 ReadLock rl(&d_lock);
1011 for (const auto& server : d_servers) {
1012 if (!upOnly || std::get<1>(server)->isUp() ) {
1019 NumberedVector<shared_ptr<DownstreamState>> getServers()
1021 NumberedVector<shared_ptr<DownstreamState>> result;
1023 ReadLock rl(&d_lock);
1029 void addServer(shared_ptr<DownstreamState>& server)
1031 WriteLock wl(&d_lock);
1032 unsigned int count = (unsigned int) d_servers.size();
1033 d_servers.push_back(make_pair(++count, server));
1034 /* we need to reorder based on the server 'order' */
1035 std::stable_sort(d_servers.begin(), d_servers.end(), [](const std::pair<unsigned int,std::shared_ptr<DownstreamState> >& a, const std::pair<unsigned int,std::shared_ptr<DownstreamState> >& b) {
1036 return a.second->order < b.second->order;
1038 /* and now we need to renumber for Lua (custom policies) */
1040 for (auto& serv : d_servers) {
1045 void removeServer(shared_ptr<DownstreamState>& server)
1047 WriteLock wl(&d_lock);
1050 for (auto it = d_servers.begin(); it != d_servers.end();) {
1052 /* we need to renumber the servers placed
1053 after the removed one, for Lua (custom policies) */
1057 else if (it->second == server) {
1058 it = d_servers.erase(it);
1068 NumberedVector<shared_ptr<DownstreamState>> d_servers;
1069 pthread_rwlock_t d_lock;
1070 bool d_useECS{false};
1072 using pools_t=map<std::string,std::shared_ptr<ServerPool>>;
1073 void setPoolPolicy(pools_t& pools, const string& poolName, std::shared_ptr<ServerPolicy> policy);
1074 void addServerToPool(pools_t& pools, const string& poolName, std::shared_ptr<DownstreamState> server);
1075 void removeServerFromPool(pools_t& pools, const string& poolName, std::shared_ptr<DownstreamState> server);
1079 ComboAddress server;
1080 std::string namespace_name;
1081 std::string ourname;
1082 std::string instance_name;
1083 unsigned int interval;
1086 enum ednsHeaderFlags {
1087 EDNS_HEADER_FLAG_NONE = 0,
1088 EDNS_HEADER_FLAG_DO = 32768
1091 struct DNSDistRuleAction
1093 std::shared_ptr<DNSRule> d_rule;
1094 std::shared_ptr<DNSAction> d_action;
1095 boost::uuids::uuid d_id;
1096 uint64_t d_creationOrder;
1099 struct DNSDistResponseRuleAction
1101 std::shared_ptr<DNSRule> d_rule;
1102 std::shared_ptr<DNSResponseAction> d_action;
1103 boost::uuids::uuid d_id;
1104 uint64_t d_creationOrder;
1107 extern GlobalStateHolder<SuffixMatchTree<DynBlock>> g_dynblockSMT;
1108 extern DNSAction::Action g_dynBlockAction;
1110 extern GlobalStateHolder<vector<CarbonConfig> > g_carbon;
1111 extern GlobalStateHolder<ServerPolicy> g_policy;
1112 extern GlobalStateHolder<servers_t> g_dstates;
1113 extern GlobalStateHolder<pools_t> g_pools;
1114 extern GlobalStateHolder<vector<DNSDistRuleAction> > g_rulactions;
1115 extern GlobalStateHolder<vector<DNSDistResponseRuleAction> > g_resprulactions;
1116 extern GlobalStateHolder<vector<DNSDistResponseRuleAction> > g_cachehitresprulactions;
1117 extern GlobalStateHolder<vector<DNSDistResponseRuleAction> > g_selfansweredresprulactions;
1118 extern GlobalStateHolder<NetmaskGroup> g_ACL;
1120 extern ComboAddress g_serverControl; // not changed during runtime
1122 extern std::vector<std::tuple<ComboAddress, bool, bool, int, std::string, std::set<int>>> g_locals; // not changed at runtime (we hope XXX)
1123 extern std::vector<shared_ptr<TLSFrontend>> g_tlslocals;
1124 extern std::vector<shared_ptr<DOHFrontend>> g_dohlocals;
1125 extern std::vector<std::unique_ptr<ClientState>> g_frontends;
1126 extern bool g_truncateTC;
1127 extern bool g_fixupCase;
1128 extern int g_tcpRecvTimeout;
1129 extern int g_tcpSendTimeout;
1130 extern int g_udpTimeout;
1131 extern uint16_t g_maxOutstanding;
1132 extern std::atomic<bool> g_configurationDone;
1133 extern uint64_t g_maxTCPClientThreads;
1134 extern uint64_t g_maxTCPQueuedConnections;
1135 extern size_t g_maxTCPQueriesPerConn;
1136 extern size_t g_maxTCPConnectionDuration;
1137 extern size_t g_maxTCPConnectionsPerClient;
1138 extern std::atomic<uint16_t> g_cacheCleaningDelay;
1139 extern std::atomic<uint16_t> g_cacheCleaningPercentage;
1140 extern bool g_verboseHealthChecks;
1141 extern uint32_t g_staleCacheEntriesTTL;
1142 extern bool g_apiReadWrite;
1143 extern std::string g_apiConfigDirectory;
1144 extern bool g_servFailOnNoPolicy;
1145 extern uint32_t g_hashperturb;
1146 extern bool g_useTCPSinglePipe;
1147 extern uint16_t g_downstreamTCPCleanupInterval;
1148 extern size_t g_udpVectorSize;
1149 extern bool g_preserveTrailingData;
1150 extern bool g_allowEmptyResponse;
1151 extern bool g_roundrobinFailOnNoServer;
1154 extern shared_ptr<BPFFilter> g_defaultBPFFilter;
1155 extern std::vector<std::shared_ptr<DynBPFFilter> > g_dynBPFFilters;
1156 #endif /* HAVE_EBPF */
1160 LocalHolders(): acl(g_ACL.getLocal()), policy(g_policy.getLocal()), rulactions(g_rulactions.getLocal()), cacheHitRespRulactions(g_cachehitresprulactions.getLocal()), selfAnsweredRespRulactions(g_selfansweredresprulactions.getLocal()), servers(g_dstates.getLocal()), dynNMGBlock(g_dynblockNMG.getLocal()), dynSMTBlock(g_dynblockSMT.getLocal()), pools(g_pools.getLocal())
1164 LocalStateHolder<NetmaskGroup> acl;
1165 LocalStateHolder<ServerPolicy> policy;
1166 LocalStateHolder<vector<DNSDistRuleAction> > rulactions;
1167 LocalStateHolder<vector<DNSDistResponseRuleAction> > cacheHitRespRulactions;
1168 LocalStateHolder<vector<DNSDistResponseRuleAction> > selfAnsweredRespRulactions;
1169 LocalStateHolder<servers_t> servers;
1170 LocalStateHolder<NetmaskTree<DynBlock> > dynNMGBlock;
1171 LocalStateHolder<SuffixMatchTree<DynBlock> > dynSMTBlock;
1172 LocalStateHolder<pools_t> pools;
1177 void controlThread(int fd, ComboAddress local);
1178 vector<std::function<void(void)>> setupLua(bool client, const std::string& config);
1179 std::shared_ptr<ServerPool> getPool(const pools_t& pools, const std::string& poolName);
1180 std::shared_ptr<ServerPool> createPoolIfNotExists(pools_t& pools, const string& poolName);
1181 NumberedServerVector getDownstreamCandidates(const pools_t& pools, const std::string& poolName);
1183 std::shared_ptr<DownstreamState> firstAvailable(const NumberedServerVector& servers, const DNSQuestion* dq);
1185 std::shared_ptr<DownstreamState> leastOutstanding(const NumberedServerVector& servers, const DNSQuestion* dq);
1186 std::shared_ptr<DownstreamState> wrandom(const NumberedServerVector& servers, const DNSQuestion* dq);
1187 std::shared_ptr<DownstreamState> whashed(const NumberedServerVector& servers, const DNSQuestion* dq);
1188 std::shared_ptr<DownstreamState> chashed(const NumberedServerVector& servers, const DNSQuestion* dq);
1189 std::shared_ptr<DownstreamState> roundrobin(const NumberedServerVector& servers, const DNSQuestion* dq);
1191 struct WebserverConfig
1193 std::string password;
1195 boost::optional<std::map<std::string, std::string> > customHeaders;
1199 void setWebserverAPIKey(const boost::optional<std::string> apiKey);
1200 void setWebserverPassword(const std::string& password);
1201 void setWebserverCustomHeaders(const boost::optional<std::map<std::string, std::string> > customHeaders);
1203 void dnsdistWebserverThread(int sock, const ComboAddress& local);
1204 void tcpAcceptorThread(void* p);
1205 #ifdef HAVE_DNS_OVER_HTTPS
1206 void dohThread(ClientState* cs);
1207 #endif /* HAVE_DNS_OVER_HTTPS */
1209 void setLuaNoSideEffect(); // if nothing has been declared, set that there are no side effects
1210 void setLuaSideEffect(); // set to report a side effect, cancelling all _no_ side effect calls
1211 bool getLuaNoSideEffect(); // set if there were only explicit declarations of _no_ side effect
1212 void resetLuaSideEffect(); // reset to indeterminate state
1214 bool responseContentMatches(const char* response, const uint16_t responseLen, const DNSName& qname, const uint16_t qtype, const uint16_t qclass, const ComboAddress& remote, unsigned int& consumed);
1215 bool processResponse(char** response, uint16_t* responseLen, size_t* responseSize, LocalStateHolder<vector<DNSDistResponseRuleAction> >& localRespRulactions, DNSResponse& dr, size_t addRoom, std::vector<uint8_t>& rewrittenResponse, bool muted);
1216 bool processRulesResult(const DNSAction::Action& action, DNSQuestion& dq, std::string& ruleresult, bool& drop);
1218 bool checkQueryHeaders(const struct dnsheader* dh);
1220 extern std::vector<std::shared_ptr<DNSCryptContext>> g_dnsCryptLocals;
1221 int handleDNSCryptQuery(char* packet, uint16_t len, std::shared_ptr<DNSCryptQuery> query, uint16_t* decryptedQueryLen, bool tcp, time_t now, std::vector<uint8_t>& response);
1222 boost::optional<std::vector<uint8_t>> checkDNSCryptQuery(const ClientState& cs, const char* query, uint16_t& len, std::shared_ptr<DNSCryptQuery>& dnsCryptQuery, time_t now, bool tcp);
1224 bool addXPF(DNSQuestion& dq, uint16_t optionCode);
1226 uint16_t getRandomDNSID();
1228 #include "dnsdist-snmp.hh"
1230 extern bool g_snmpEnabled;
1231 extern bool g_snmpTrapsEnabled;
1232 extern DNSDistSNMPAgent* g_snmpAgent;
1233 extern bool g_addEDNSToSelfGeneratedResponses;
1235 static const size_t s_udpIncomingBufferSize{1500};
1237 enum class ProcessQueryResult { Drop, SendAnswer, PassToBackend };
1238 ProcessQueryResult processQuery(DNSQuestion& dq, ClientState& cs, LocalHolders& holders, std::shared_ptr<DownstreamState>& selectedBackend);
1240 DNSResponse makeDNSResponseFromIDState(IDState& ids, struct dnsheader* dh, size_t bufferSize, uint16_t responseLen, bool isTCP);
1241 void setIDStateFromDNSQuestion(IDState& ids, DNSQuestion& dq, DNSName&& qname);
1243 int pickBackendSocketForSending(std::shared_ptr<DownstreamState>& state);
1244 ssize_t udpClientSendRequestToBackend(const std::shared_ptr<DownstreamState>& ss, const int sd, const char* request, const size_t requestLen, bool healthCheck=false);