#include <sys/stat.h>
#include <unistd.h>
+#include "recpacketcache.hh"
#include "ws-recursor.hh"
#include <pthread.h>
-#include "recpacketcache.hh"
#include "utility.hh"
#include "dns_random.hh"
#ifdef HAVE_LIBSODIUM
typedef map<ComboAddress, uint32_t, ComboAddress::addressOnlyLessThan> tcpClientCounts_t;
static thread_local std::shared_ptr<RecursorLua4> t_pdl;
-static thread_local int t_id;
+static thread_local int t_id = -1;
static thread_local std::shared_ptr<Regex> t_traceRegex;
static thread_local std::unique_ptr<tcpClientCounts_t> t_tcpClientCounts;
int readToThread;
int writeFromThread;
int readFromThread;
+ int writeQueriesToThread; // this one is non-blocking
+ int readQueriesToThread;
};
/* the TID of the thread handling the web server, carbon, statistics and the control channel */
to the other threads */
static const int s_distributorThreadID = 0;
-typedef vector<int> tcpListenSockets_t;
+typedef std::map<int, std::set<int>> tcpListenSockets_t;
typedef map<int, ComboAddress> listenSocketsAddresses_t; // is shared across all threads right now
+
typedef vector<pair<int, function< void(int, any&) > > > deferredAdd_t;
static const ComboAddress g_local4("0.0.0.0"), g_local6("::");
static std::shared_ptr<SyncRes::domainmap_t> g_initialDomainMap; // new threads needs this to be setup
static std::shared_ptr<NetmaskGroup> g_initialAllowFrom; // new thread needs to be setup with this
static size_t g_tcpMaxQueriesPerConn;
+static size_t s_maxUDPQueriesPerRound;
static uint64_t g_latencyStatSize;
static uint32_t g_disthashseed;
static unsigned int g_maxTCPPerClient;
return theArg;
}
-unsigned int getRecursorThreadId()
+int getRecursorThreadId()
{
- return static_cast<unsigned int>(t_id);
+ return t_id;
}
int getMTaskerTID()
Netmask requestorNM(remote, remote.sin4.sin_family == AF_INET ? maskV4 : maskV6);
const ComboAddress& requestor = requestorNM.getMaskedNetwork();
RecProtoBufMessage message(DNSProtoBufMessage::Query, uniqueId, &requestor, &local, qname, qtype, qclass, id, tcp, len);
+ message.setServerIdentity(SyncRes::s_serverID);
message.setEDNSSubnet(ednssubnet, ednssubnet.isIpv4() ? maskV4 : maskV6);
message.setRequestorId(requestorId);
message.setDeviceId(deviceId);
if (luaconfsLocal->protobufServer) {
Netmask requestorNM(dc->d_remote, dc->d_remote.sin4.sin_family == AF_INET ? luaconfsLocal->protobufMaskV4 : luaconfsLocal->protobufMaskV6);
const ComboAddress& requestor = requestorNM.getMaskedNetwork();
+ pbMessage.setServerIdentity(SyncRes::s_serverID);
pbMessage.update(dc->d_uuid, &requestor, &dc->d_local, dc->d_tcp, dc->d_mdp.d_header.id);
pbMessage.setEDNSSubnet(dc->d_ednssubnet.source, dc->d_ednssubnet.source.isIpv4() ? luaconfsLocal->protobufMaskV4 : luaconfsLocal->protobufMaskV6);
pbMessage.setQuestion(dc->d_mdp.d_qname, dc->d_mdp.d_qtype, dc->d_mdp.d_qclass);
}
- sr.d_outqueries ? t_RC->cacheMisses++ : t_RC->cacheHits++;
+ if (sr.d_outqueries || sr.d_authzonequeries) {
+ t_RC->cacheMisses++;
+ }
+ else {
+ t_RC->cacheHits++;
+ }
if(spent < 0.001)
g_stats.answers0_1++;
}
catch(...) {
L<<Logger::Error<<"Any other exception in a resolver context "<< makeLoginfo(dc) <<endl;
+ delete dc;
}
g_stats.maxMThreadStackUsage = max(MT->getMaxStackUsage(), g_stats.maxMThreadStackUsage);
dc->d_tcp=true;
dc->setRemote(&conn->d_remote);
ComboAddress dest;
- memset(&dest, 0, sizeof(dest));
+ dest.reset();
dest.sin4.sin_family = conn->d_remote.sin4.sin_family;
socklen_t len = dest.getSocklen();
getsockname(conn->getFD(), (sockaddr*)&dest, &len); // if this fails, we're ok with it
bool cacheHit = false;
RecProtoBufMessage pbMessage(DNSProtoBufMessage::DNSProtoBufMessageType::Response);
#ifdef HAVE_PROTOBUF
+ pbMessage.setServerIdentity(SyncRes::s_serverID);
if(luaconfsLocal->protobufServer) {
if (!luaconfsLocal->protobufTaggedOnly || !policyTags.empty()) {
protobufLogQuery(luaconfsLocal->protobufServer, luaconfsLocal->protobufMaskV4, luaconfsLocal->protobufMaskV6, uniqueId, fromaddr, destaddr, ednssubnet.source, false, dh->id, question.size(), qname, qtype, qclass, policyTags, requestorId, deviceId);
fromaddr.sin6.sin6_family=AF_INET6; // this makes sure fromaddr is big enough
fillMSGHdr(&msgh, &iov, cbuf, sizeof(cbuf), data, sizeof(data), &fromaddr);
- for(;;)
+ for(size_t counter = 0; counter < s_maxUDPQueriesPerRound; counter++)
if((len=recvmsg(fd, &msgh, 0)) >= 0) {
firstQuery = false;
struct timeval tv={0,0};
HarvestTimestamp(&msgh, &tv);
ComboAddress dest;
- memset(&dest, 0, sizeof(dest)); // this makes sure we ignore this address if not returned by recvmsg above
+ dest.reset(); // this makes sure we ignore this address if not returned by recvmsg above
auto loc = rplookup(g_listenSocketsAddresses, fd);
if(HarvestDestinationAddress(&msgh, &dest)) {
// but.. need to get port too
ComboAddress sin;
- memset((char *)&sin,0, sizeof(sin));
+ sin.reset();
sin.sin4.sin_family = AF_INET;
if(!IpToU32(st.host, (uint32_t*)&sin.sin4.sin_addr.s_addr)) {
sin.sin6.sin6_family = AF_INET6;
setSocketSendBuffer(fd, 65000);
listen(fd, 128);
deferredAdds[threadId].push_back(make_pair(fd, handleNewTCPQuestion));
- g_tcpListenSockets.push_back(fd);
+ g_tcpListenSockets[threadId].insert(fd);
// we don't need to update g_listenSocketsAddresses since it doesn't work for TCP/IP:
// - fd is not that which we know here, but returned from accept()
if(sin.sin4.sin_family == AF_INET)
ComboAddress sin;
- memset(&sin, 0, sizeof(sin));
+ sin.reset();
sin.sin4.sin_family = AF_INET;
if(!IpToU32(st.host.c_str() , (uint32_t*)&sin.sin4.sin_addr.s_addr)) {
sin.sin6.sin6_family = AF_INET6;
static void houseKeeping(void *)
{
- static thread_local time_t last_stat, last_rootupdate, last_prune, last_secpoll;
+ static thread_local time_t last_rootupdate, last_prune, last_secpoll;
static thread_local int cleanCounter=0;
static thread_local bool s_running; // houseKeeping can get suspended in secpoll, and be restarted, which makes us do duplicate work
try {
last_rootupdate=now.tv_sec;
}
- if (t_id == s_handlerThreadID) {
- if(g_statisticsInterval > 0 && now.tv_sec - last_stat >= g_statisticsInterval) {
- doStats();
- last_stat=time(0);
- }
- }
- else if(t_id == s_distributorThreadID) {
+ if(t_id == s_distributorThreadID) {
if(now.tv_sec - last_secpoll >= 3600) {
try {
tps.readFromThread = fd[0];
tps.writeFromThread = fd[1];
+ if(pipe(fd) < 0)
+ unixDie("Creating pipe for inter-thread communications");
+ tps.readQueriesToThread = fd[0];
+ tps.writeQueriesToThread = fd[1];
+
+ if (!setNonBlocking(tps.writeQueriesToThread)) {
+ unixDie("Making pipe for inter-thread communications non-blocking");
+ }
+
g_pipes.push_back(tps);
}
}
void broadcastFunction(const pipefunc_t& func)
{
- /* This function might be called by the worker with t_id 0 during startup
- for the initialization of ACLs and domain maps */
- if (t_id != s_handlerThreadID && t_id != s_distributorThreadID) {
+ /* This function might be called before t_id are set during startup
+ for the initialization of ACLs and domain maps, but the default is the same
+ than the handler thread */
+ if (t_id != s_handlerThreadID) {
L<<Logger::Error<<"broadcastFunction() has been called by a worker ("<<t_id<<")"<<endl;
exit(1);
}
- if (t_id == s_handlerThreadID) {
- /* the distributor will call itself below, but if we are the handler thread,
- call the function ourselves to update the ACL or domain maps for example */
- func();
- }
+ /* the distributor will call itself below, but if we are the handler thread,
+ call the function ourselves to update the ACL or domain maps for example */
+ func();
int n = 0;
for(ThreadPipeSet& tps : g_pipes)
tmsg->func = func;
tmsg->wantAnswer = false;
- if(write(tps.writeToThread, &tmsg, sizeof(tmsg)) != sizeof(tmsg)) {
+ ssize_t written = write(tps.writeQueriesToThread, &tmsg, sizeof(tmsg));
+ if (written > 0) {
+ if (static_cast<size_t>(written) != sizeof(tmsg)) {
+ delete tmsg;
+ unixDie("write to thread pipe returned wrong size or error");
+ }
+ }
+ else {
+ int error = errno;
delete tmsg;
- unixDie("write to thread pipe returned wrong size or error");
+ if (error == EAGAIN || error == EWOULDBLOCK) {
+ g_stats.queryPipeFullDrops++;
+ } else {
+ unixDie("write to thread pipe returned wrong size or error:" + std::to_string(error));
+ }
}
}
{
ThreadMSG* tmsg = nullptr;
- if(read(fd, &tmsg, sizeof(tmsg)) != sizeof(tmsg)) { // fd == readToThread
+ if(read(fd, &tmsg, sizeof(tmsg)) != sizeof(tmsg)) { // fd == readToThread || fd == readQueriesToThread
unixDie("read from thread pipe returned wrong size or error");
}
/*
This function should only be called by the handler to gather metrics, wipe the cache,
- reload the Lua script (not the Lua config) or change the current trace regex */
+ reload the Lua script (not the Lua config) or change the current trace regex,
+ and by the SNMP thread to gather metrics. */
template<class T> T broadcastAccFunction(const boost::function<T*()>& func)
{
+ /* the SNMP thread uses id -1 too */
if (t_id != s_handlerThreadID) {
- L<<Logger::Error<<"broadcastFunction has been called by a worker ("<<t_id<<")"<<endl;
+ L<<Logger::Error<<"broadcastAccFunction has been called by a worker ("<<t_id<<")"<<endl;
exit(1);
-
}
T ret=T();
g_maxCacheEntries = ::arg().asNum("max-cache-entries");
g_maxPacketCacheEntries = ::arg().asNum("max-packetcache-entries");
-
+
+ luaConfigDelayedThreads delayedLuaThreads;
try {
- loadRecursorLuaConfig(::arg()["lua-config-file"], ::arg().mustDo("daemon"));
+ loadRecursorLuaConfig(::arg()["lua-config-file"], delayedLuaThreads);
}
catch (PDNSException &e) {
L<<Logger::Error<<"Cannot load Lua configuration: "<<e.reason<<endl;
else {
makeUDPServerSockets(0);
makeTCPServerSockets(0);
+
+ if (!g_weDistributeQueries) {
+ /* we are not distributing queries and we don't have reuseport,
+ so every thread will be listening on all the TCP sockets */
+ for (unsigned int threadId = 1; threadId < g_numWorkerThreads; threadId++) {
+ g_tcpListenSockets[threadId] = g_tcpListenSockets[0];
+ }
+ }
}
SyncRes::parseEDNSSubnetWhitelist(::arg()["edns-subnet-whitelist"]);
L<<Logger::Warning<<"Calling daemonize, going to background"<<endl;
L.toConsole(Logger::Critical);
daemonize();
- loadRecursorLuaConfig(::arg()["lua-config-file"], false);
}
signal(SIGUSR1,usr1Handler);
signal(SIGUSR2,usr2Handler);
Utility::dropUserPrivs(newuid);
+ startLuaConfigDelayedThreads(delayedLuaThreads);
+
makeThreadPipes();
g_tcpTimeout=::arg().asNum("client-tcp-timeout");
g_maxTCPPerClient=::arg().asNum("max-tcp-per-client");
g_tcpMaxQueriesPerConn=::arg().asNum("max-tcp-queries-per-connection");
+ s_maxUDPQueriesPerRound=::arg().asNum("max-udp-queries-per-round");
if (::arg().mustDo("snmp-agent")) {
g_snmpAgent = std::make_shared<RecursorSNMPAgent>("recursor", ::arg()["snmp-master-socket"]);
#endif
L<<Logger::Warning<<"Done priming cache with root hints"<<endl;
- try {
- if(!::arg()["lua-dns-script"].empty()) {
- t_pdl = std::make_shared<RecursorLua4>(::arg()["lua-dns-script"]);
- L<<Logger::Warning<<"Loaded 'lua' script from '"<<::arg()["lua-dns-script"]<<"'"<<endl;
+ if(worker && (!g_weDistributeQueries || t_id != s_distributorThreadID)) {
+ try {
+ if(!::arg()["lua-dns-script"].empty()) {
+ t_pdl = std::make_shared<RecursorLua4>(::arg()["lua-dns-script"]);
+ L<<Logger::Warning<<"Loaded 'lua' script from '"<<::arg()["lua-dns-script"]<<"'"<<endl;
+ }
+ }
+ catch(std::exception &e) {
+ L<<Logger::Error<<"Failed to load 'lua' script from '"<<::arg()["lua-dns-script"]<<"': "<<e.what()<<endl;
+ _exit(99);
}
- }
- catch(std::exception &e) {
- L<<Logger::Error<<"Failed to load 'lua' script from '"<<::arg()["lua-dns-script"]<<"': "<<e.what()<<endl;
- _exit(99);
}
unsigned int ringsize=::arg().asNum("stats-ringbuffer-entries") / g_numWorkerThreads;
}
else {
t_fdm->addReadFD(g_pipes[t_id].readToThread, handlePipeRequest);
+ t_fdm->addReadFD(g_pipes[t_id].readQueriesToThread, handlePipeRequest);
if(g_useOneSocketPerThread) {
for(deferredAdd_t::const_iterator i = deferredAdds[t_id].cbegin(); i != deferredAdds[t_id].cend(); ++i) {
bool listenOnTCP(true);
+ time_t last_stat = 0;
time_t last_carbon=0;
time_t carbonInterval=::arg().asNum("carbon-interval");
counter.store(0); // used to periodically execute certain tasks
counter++;
- if(!worker && statsWanted) {
- doStats();
- }
+ if(!worker) {
+ if(statsWanted || (g_statisticsInterval > 0 && (g_now.tv_sec - last_stat) >= g_statisticsInterval)) {
+ doStats();
+ last_stat = g_now.tv_sec;
+ }
- Utility::gettimeofday(&g_now, 0);
+ Utility::gettimeofday(&g_now, 0);
- if(!worker && (g_now.tv_sec - last_carbon >= carbonInterval)) {
- MT->makeThread(doCarbonDump, 0);
- last_carbon = g_now.tv_sec;
+ if((g_now.tv_sec - last_carbon) >= carbonInterval) {
+ MT->makeThread(doCarbonDump, 0);
+ last_carbon = g_now.tv_sec;
+ }
}
t_fdm->run(&g_now);
if(worker && (!g_weDistributeQueries || t_id == s_distributorThreadID)) { // if pdns distributes queries, only tid 0 should do this
if(listenOnTCP) {
if(TCPConnection::getCurrentConnections() > maxTcpClients) { // shutdown, too many connections
- for(tcpListenSockets_t::iterator i=g_tcpListenSockets.begin(); i != g_tcpListenSockets.end(); ++i)
- t_fdm->removeReadFD(*i);
+ for(const auto fd : g_tcpListenSockets[t_id]) {
+ t_fdm->removeReadFD(fd);
+ }
listenOnTCP=false;
}
}
else {
if(TCPConnection::getCurrentConnections() <= maxTcpClients) { // reenable
- for(tcpListenSockets_t::iterator i=g_tcpListenSockets.begin(); i != g_tcpListenSockets.end(); ++i)
- t_fdm->addReadFD(*i, handleNewTCPQuestion);
+ for(const auto fd : g_tcpListenSockets[t_id]) {
+ t_fdm->addReadFD(fd, handleNewTCPQuestion);
+ }
listenOnTCP=true;
}
}
::arg().set("max-qperq", "Maximum outgoing queries per query")="50";
::arg().set("max-total-msec", "Maximum total wall-clock time per query in milliseconds, 0 for unlimited")="7000";
::arg().set("max-recursion-depth", "Maximum number of internal recursion calls per query, 0 for unlimited")="40";
+ ::arg().set("max-udp-queries-per-round", "Maximum number of UDP queries processed per recvmsg() round, before returning back to normal processing")="10000";
::arg().set("include-dir","Include *.conf files from this directory")="";
::arg().set("security-poll-suffix","Domain name from which to query security update notifications")="secpoll.powerdns.com.";
projects / thirdparty / pdns.git / blobdiff