root/trunk/pdns/pdns/pdns_recursor.cc @ 633

Revision 633, 31.2 KB (checked in by ahu, 7 years ago)

make rec_control clean up its socket properly. Remove some silly dependencies. Silence signedness warning.

  • Property svn:eol-style set to native
  • Property svn:keywords set to author date id revision
Line 
1/*
2    PowerDNS Versatile Database Driven Nameserver
3    Copyright (C) 2003 - 2006  PowerDNS.COM BV
4
5    This program is free software; you can redistribute it and/or modify
6    it under the terms of the GNU General Public License version 2
7    as published by the Free Software Foundation
8
9    This program is distributed in the hope that it will be useful,
10    but WITHOUT ANY WARRANTY; without even the implied warranty of
11    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12    GNU General Public License for more details.
13
14    You should have received a copy of the GNU General Public License
15    along with this program; if not, write to the Free Software
16    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17*/
18
19#include "utility.hh"
20#include <iostream>
21#include <errno.h>
22#include <map>
23#include <set>
24#ifndef WIN32
25#include <netdb.h>
26#endif // WIN32
27#include "recursor_cache.hh"
28#include <stdio.h>
29#include <signal.h>
30#include <stdlib.h>
31#include <unistd.h>
32#include "mtasker.hh"
33#include <utility>
34#include "arguments.hh"
35#include "syncres.hh"
36#include <fcntl.h>
37#include <fstream>
38#include "sstuff.hh"
39#include <boost/tuple/tuple.hpp>
40#include <boost/tuple/tuple_comparison.hpp>
41#include <boost/shared_array.hpp>
42#include <boost/lexical_cast.hpp>
43#include "dnsparser.hh"
44#include "dnswriter.hh"
45#include "dnsrecords.hh"
46#include "zoneparser-tng.hh"
47#include "rec_channel.hh"
48#include "logger.hh"
49
50#ifndef RECURSOR
51#include "statbag.hh"
52StatBag S;
53#endif
54
55
56using namespace boost;
57
58#ifdef __FreeBSD__           // see cvstrac ticket #26
59#include <pthread.h>
60#include <semaphore.h>
61#endif
62
63MemRecursorCache RC;
64RecursorStats g_stats;
65bool g_quiet;
66string s_programname="pdns_recursor";
67
68struct DNSComboWriter {
69  DNSComboWriter(const char* data, uint16_t len, const struct timeval& now) : d_mdp(data, len), d_now(now), d_tcp(false), d_socket(-1)
70  {}
71  MOADNSParser d_mdp;
72  void setRemote(struct sockaddr* sa, socklen_t len)
73  {
74    memcpy((void *)d_remote, (void *)sa, len);
75    d_socklen=len;
76  }
77
78  void setSocket(int sock)
79  {
80    d_socket=sock;
81  }
82
83  string getRemote() const
84  {
85    return sockAddrToString((struct sockaddr_in *)d_remote, d_socklen);
86  }
87
88  struct timeval d_now;
89  char d_remote[sizeof(sockaddr_in6)];
90  socklen_t d_socklen;
91  bool d_tcp;
92  int d_socket;
93};
94
95
96#ifndef WIN32
97#ifndef __FreeBSD__
98extern "C" {
99  int sem_init(sem_t*, int, unsigned int){return 0;}
100  int sem_wait(sem_t*){return 0;}
101  int sem_trywait(sem_t*){return 0;}
102  int sem_post(sem_t*){return 0;}
103  int sem_getvalue(sem_t*, int*){return 0;}
104  pthread_t pthread_self(void){return (pthread_t) 0;}
105  int pthread_mutex_init(pthread_mutex_t *mutex, const pthread_mutexattr_t *mutexattr){ return 0; }
106  int pthread_mutex_lock(pthread_mutex_t *mutex){ return 0; }
107  int pthread_mutex_unlock(pthread_mutex_t *mutex) { return 0; }
108  int pthread_mutex_destroy(pthread_mutex_t *mutex) { return 0; }
109}
110#endif // __FreeBSD__
111#endif // WIN32
112
113ArgvMap &arg()
114{
115  static ArgvMap theArg;
116  return theArg;
117}
118static int d_clientsock;
119static vector<int> d_udpserversocks;
120
121typedef vector<int> tcpserversocks_t;
122static tcpserversocks_t s_tcpserversocks;
123
124static map<int,PacketID> d_tcpclientreadsocks, d_tcpclientwritesocks;
125
126MTasker<PacketID,string>* MT;
127
128int asendtcp(const string& data, Socket* sock) 
129{
130  PacketID pident;
131  pident.sock=sock;
132  pident.outMSG=data;
133  string packet;
134
135  d_tcpclientwritesocks[sock->getHandle()]=pident;
136
137  int ret=MT->waitEvent(pident,&packet,1);
138  if(!ret || ret==-1) { // timeout
139    d_tcpclientwritesocks.erase(sock->getHandle());
140  }
141  return ret;
142}
143
144// -1 is error, 0 is timeout, 1 is success
145int arecvtcp(string& data, int len, Socket* sock) 
146{
147  data="";
148  PacketID pident;
149  pident.sock=sock;
150  pident.inNeeded=len;
151
152  d_tcpclientreadsocks[sock->getHandle()]=pident;
153
154  int ret=MT->waitEvent(pident,&data,1);
155  if(!ret || ret==-1) { // timeout
156    d_tcpclientreadsocks.erase(sock->getHandle());
157  }
158  return ret;
159}
160
161
162/* these two functions are used by LWRes */
163// -1 is error, > 1 is success
164int asendto(const char *data, int len, int flags, struct sockaddr *toaddr, int addrlen, int id) 
165{
166  return sendto(d_clientsock, data, len, flags, toaddr, addrlen);
167}
168
169// -1 is error, 0 is timeout, 1 is success
170int arecvfrom(char *data, int len, int flags, struct sockaddr *toaddr, Utility::socklen_t *addrlen, int *d_len, int id)
171{
172  PacketID pident;
173  pident.id=id;
174  memcpy(&pident.remote, toaddr, sizeof(pident.remote));
175
176  string packet;
177  int ret=MT->waitEvent(pident, &packet, 1);
178  if(ret > 0) {
179    *d_len=packet.size();
180    memcpy(data,packet.c_str(),min(len,*d_len));
181  }
182  return ret;
183}
184
185void setReceiveBuffer(int fd, uint32_t size)
186{
187  uint32_t psize;
188  socklen_t len=sizeof(psize);
189  getsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char*)&psize, &len);
190  if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, (char*)&size, sizeof(size)) < 0 )
191    L<<Logger::Error<<"Warning: unable to raise socket buffer size to "<<size<<": "<<strerror(errno)<<"\n";
192}
193
194
195static void writePid(void)
196{
197  string fname=::arg()["socket-dir"]+"/"+s_programname+".pid";
198  ofstream of(fname.c_str());
199  if(of)
200    of<< getpid() <<endl;
201  else
202    L<<Logger::Error<<"Requested to write pid for "<<getpid()<<" to "<<fname<<" failed: "<<strerror(errno)<<endl;
203}
204
205void primeHints(void)
206{
207  // prime root cache
208  set<DNSResourceRecord>nsset;
209
210  if(::arg()["hint-file"].empty()) {
211    static char*ips[]={"198.41.0.4", "192.228.79.201", "192.33.4.12", "128.8.10.90", "192.203.230.10", "192.5.5.241", "192.112.36.4", "128.63.2.53", 
212                       "192.36.148.17","192.58.128.30", "193.0.14.129", "198.32.64.12", "202.12.27.33"};
213    DNSResourceRecord arr, nsrr;
214    arr.qtype=QType::A;
215    arr.ttl=time(0)+3600000;
216    nsrr.qtype=QType::NS;
217    nsrr.ttl=time(0)+3600000;
218   
219    for(char c='A';c<='M';++c) {
220      static char templ[40];
221      strncpy(templ,"A.ROOT-SERVERS.NET", sizeof(templ) - 1);
222      *templ=c;
223      arr.qname=nsrr.content=templ;
224      arr.content=ips[c-'A'];
225      set<DNSResourceRecord> aset;
226      aset.insert(arr);
227      RC.replace(string(templ), QType(QType::A), aset);
228     
229      nsset.insert(nsrr);
230    }
231  }
232  else {
233    ZoneParserTNG zpt(::arg()["hint-file"]);
234    DNSResourceRecord rr;
235    set<DNSResourceRecord> aset;
236
237    while(zpt.get(rr)) {
238      rr.ttl+=time(0);
239      if(rr.qtype.getCode()==QType::A) {
240        set<DNSResourceRecord> aset;
241        aset.insert(rr);
242        RC.replace(rr.qname, QType(QType::A), aset);
243      }
244      if(rr.qtype.getCode()==QType::NS) {
245        nsset.insert(rr);
246      }
247    }
248  }
249  RC.replace("", QType(QType::NS), nsset); // and stuff in the cache
250}
251
252void startDoResolve(void *p)
253{
254  try {
255    DNSComboWriter* dc=(DNSComboWriter *)p;
256
257    uint16_t maxudpsize=512;
258    MOADNSParser::EDNSOpts edo;
259    if(dc->d_mdp.getEDNSOpts(&edo)) {
260      maxudpsize=edo.d_packetsize;
261    }
262
263    vector<DNSResourceRecord> ret;
264   
265    vector<uint8_t> packet;
266    DNSPacketWriter pw(packet, dc->d_mdp.d_qname, dc->d_mdp.d_qtype, dc->d_mdp.d_qclass);
267
268    pw.getHeader()->aa=0;
269    pw.getHeader()->ra=1;
270    pw.getHeader()->qr=1;
271    pw.getHeader()->id=dc->d_mdp.d_header.id;
272    pw.getHeader()->rd=dc->d_mdp.d_header.rd;
273
274    //    MT->setTitle("udp question for "+P.qdomain+"|"+P.qtype.getName());
275    SyncRes sr(dc->d_now);
276    if(!g_quiet)
277      L<<Logger::Error<<"["<<MT->getTid()<<"] " << (dc->d_tcp ? "TCP " : "") << "question for '"<<dc->d_mdp.d_qname<<"|"
278       <<DNSRecordContent::NumberToType(dc->d_mdp.d_qtype)<<"' from "<<dc->getRemote()<<endl;
279
280    sr.setId(MT->getTid());
281    if(!dc->d_mdp.d_header.rd)
282      sr.setCacheOnly();
283
284    int res=sr.beginResolve(dc->d_mdp.d_qname, QType(dc->d_mdp.d_qtype), ret);
285    if(res<0) {
286      pw.getHeader()->rcode=RCode::ServFail;
287      g_stats.servFails++;
288    }
289    else {
290      pw.getHeader()->rcode=res;
291      switch(res) {
292      case RCode::ServFail:
293        g_stats.servFails++;
294        break;
295      case RCode::NXDomain:
296        g_stats.nxDomains++;
297        break;
298      case RCode::NoError:
299        g_stats.noErrors++;
300        break;
301      }
302     
303      if(ret.size()) {
304        shuffle(ret);
305        for(vector<DNSResourceRecord>::const_iterator i=ret.begin();i!=ret.end();++i) {
306          pw.startRecord(i->qname, i->qtype.getCode(), i->ttl, 1, (DNSPacketWriter::Place)i->d_place);
307          shared_ptr<DNSRecordContent> drc(DNSRecordContent::mastermake(i->qtype.getCode(), 1, i->content)); 
308          drc->toPacket(pw);
309          if(!dc->d_tcp && pw.size() > maxudpsize) {
310            pw.rollback();
311            if(i->d_place==DNSResourceRecord::ANSWER)  // only truncate if we actually omitted parts of the answer
312              pw.getHeader()->tc=1;
313            goto sendit; // need to jump over pw.commit
314          }
315        }
316        pw.commit();
317      }
318    }
319  sendit:;
320    if(!dc->d_tcp) {
321      sendto(dc->d_socket, &*packet.begin(), packet.size(), 0, (struct sockaddr *)(dc->d_remote), dc->d_socklen);
322    }
323    else {
324      char buf[2];
325      buf[0]=packet.size()/256;
326      buf[1]=packet.size()%256;
327
328      struct iovec iov[2];
329
330      iov[0].iov_base=(void*)buf;              iov[0].iov_len=2;
331      iov[1].iov_base=(void*)&*packet.begin(); iov[1].iov_len = packet.size();
332
333      int ret=writev(dc->d_socket, iov, 2);
334
335      if(ret <= 0 ) 
336        L<<Logger::Error<<"Error writing TCP answer to "<<dc->getRemote()<<": "<< (ret ? strerror(errno) : "EOF") <<endl;
337      else if((unsigned int)ret != 2 + packet.size())
338        L<<Logger::Error<<"Oops, partial answer sent to "<<dc->getRemote()<<" - probably would have trouble receiving our answer anyhow (size="<<packet.size()<<")"<<endl;
339    }
340
341    //    MT->setTitle("DONE! udp question for "+P.qdomain+"|"+P.qtype.getName());
342    if(!g_quiet) {
343      L<<Logger::Error<<"["<<MT->getTid()<<"] answer to "<<(dc->d_mdp.d_header.rd?"":"non-rd ")<<"question '"<<dc->d_mdp.d_qname<<"|"<<DNSRecordContent::NumberToType(dc->d_mdp.d_qtype);
344      L<<"': "<<ntohs(pw.getHeader()->ancount)<<" answers, "<<ntohs(pw.getHeader()->arcount)<<" additional, took "<<sr.d_outqueries<<" packets, "<<
345        sr.d_throttledqueries<<" throttled, "<<sr.d_timeouts<<" timeouts, "<<sr.d_tcpoutqueries<<" tcp connections, rcode="<<res<<endl;
346    }
347   
348    sr.d_outqueries ? RC.cacheMisses++ : RC.cacheHits++; 
349    float spent=makeFloat(sr.d_now-dc->d_now);
350    if(spent < 0.001)
351      g_stats.answers0_1++;
352    else if(spent < 0.010)
353      g_stats.answers1_10++;
354    else if(spent < 0.1)
355      g_stats.answers10_100++;
356    else if(spent < 1.0)
357      g_stats.answers100_1000++;
358    else
359      g_stats.answersSlow++;
360
361    uint64_t newLat=(uint64_t)(spent*1000000);
362    if(newLat < 1000000)  // outliers of several minutes exist..
363      g_stats.avgLatencyUsec=(uint64_t)((1-0.0001)*g_stats.avgLatencyUsec + 0.0001*newLat);
364    delete dc;
365  }
366  catch(AhuException &ae) {
367    L<<Logger::Error<<"startDoResolve problem: "<<ae.reason<<endl;
368  }
369  catch(exception& e) {
370    L<<Logger::Error<<"STL error: "<<e.what()<<endl;
371  }
372  catch(...) {
373    L<<Logger::Error<<"Any other exception in a resolver context"<<endl;
374  }
375}
376
377RecursorControlChannel s_rcc;
378
379void makeControlChannelSocket()
380{
381  s_rcc.listen(::arg()["socket-dir"]+"/pdns_recursor.controlsocket");
382}
383
384void makeClientSocket()
385{
386  d_clientsock=socket(AF_INET, SOCK_DGRAM,0);
387  if(d_clientsock<0) 
388    throw AhuException("Making a socket for resolver: "+stringerror());
389  setReceiveBuffer(d_clientsock, 250000); 
390  struct sockaddr_in sin;
391  memset((char *)&sin,0, sizeof(sin));
392 
393  sin.sin_family = AF_INET;
394
395  if(!IpToU32(::arg()["query-local-address"], &sin.sin_addr.s_addr))
396    throw AhuException("Unable to resolve local address '"+ ::arg()["query-local-address"] +"'"); 
397
398  int tries=10;
399  while(--tries) {
400    uint16_t port=10000+Utility::random()%10000;
401    sin.sin_port = htons(port); 
402   
403    if (::bind(d_clientsock, (struct sockaddr *)&sin, sizeof(sin)) >= 0) 
404      break;
405   
406  }
407  if(!tries)
408    throw AhuException("Resolver binding to local socket: "+stringerror());
409
410  Utility::setNonBlocking(d_clientsock);
411
412  L<<Logger::Error<<"Sending UDP queries from "<<inet_ntoa(sin.sin_addr)<<":"<< ntohs(sin.sin_port)  <<endl;
413}
414
415void makeTCPServerSockets()
416{
417  vector<string>locals;
418  stringtok(locals,::arg()["local-address"]," ,");
419
420  if(locals.empty())
421    throw AhuException("No local address specified");
422 
423  for(vector<string>::const_iterator i=locals.begin();i!=locals.end();++i) {
424    int fd=socket(AF_INET, SOCK_STREAM,0);
425    if(fd<0) 
426      throw AhuException("Making a server socket for resolver: "+stringerror());
427 
428    struct sockaddr_in sin;
429    memset((char *)&sin,0, sizeof(sin));
430   
431    sin.sin_family = AF_INET;
432    if(!IpToU32(*i, &sin.sin_addr.s_addr))
433      throw AhuException("Unable to resolve local address '"+ *i +"'"); 
434
435    int tmp=1;
436    if(setsockopt(fd,SOL_SOCKET,SO_REUSEADDR,(char*)&tmp,sizeof tmp)<0) {
437      L<<Logger::Error<<"Setsockopt failed for TCP listening socket"<<endl;
438      exit(1); 
439    }
440   
441    sin.sin_port = htons(::arg().asNum("local-port")); 
442   
443    if (::bind(fd, (struct sockaddr *)&sin, sizeof(sin))<0) 
444      throw AhuException("Binding TCP server socket for "+*i+": "+stringerror());
445   
446    Utility::setNonBlocking(fd);
447    listen(fd, 128);
448    s_tcpserversocks.push_back(fd);
449    L<<Logger::Error<<"Listening for TCP queries on "<<inet_ntoa(sin.sin_addr)<<":"<<::arg().asNum("local-port")<<endl;
450  }
451}
452
453void makeUDPServerSockets()
454{
455  vector<string>locals;
456  stringtok(locals,::arg()["local-address"]," ,");
457
458  if(locals.empty())
459    throw AhuException("No local address specified");
460 
461  if(::arg()["local-address"]=="0.0.0.0") {
462    L<<Logger::Warning<<"It is advised to bind to explicit addresses with the --local-address option"<<endl;
463  }
464
465  for(vector<string>::const_iterator i=locals.begin();i!=locals.end();++i) {
466    int fd=socket(AF_INET, SOCK_DGRAM,0);
467    if(fd<0) 
468      throw AhuException("Making a server socket for resolver: "+stringerror());
469    setReceiveBuffer(fd, 250000);
470    struct sockaddr_in sin;
471    memset((char *)&sin,0, sizeof(sin));
472   
473    sin.sin_family = AF_INET;
474    if(!IpToU32(*i, &sin.sin_addr.s_addr))
475      throw AhuException("Unable to resolve local address '"+ *i +"'"); 
476   
477    sin.sin_port = htons(::arg().asNum("local-port")); 
478   
479    if (::bind(fd, (struct sockaddr *)&sin, sizeof(sin))<0) 
480      throw AhuException("Resolver binding to server socket for "+*i+": "+stringerror());
481   
482    Utility::setNonBlocking(fd);
483    d_udpserversocks.push_back(fd);
484    L<<Logger::Error<<"Listening for UDP queries on "<<inet_ntoa(sin.sin_addr)<<":"<<::arg().asNum("local-port")<<endl;
485  }
486}
487
488
489#ifndef WIN32
490void daemonize(void)
491{
492  if(fork())
493    exit(0); // bye bye
494 
495  setsid(); 
496
497  // cleanup open fds, but skip sockets
498  close(0);
499  close(1);
500  close(2);
501}
502#endif
503
504uint64_t counter;
505bool statsWanted;
506
507
508void usr1Handler(int)
509{
510  statsWanted=true;
511}
512
513
514
515void usr2Handler(int)
516{
517  SyncRes::setLog(true);
518  g_quiet=false;
519  ::arg().set("quiet")="no";
520
521}
522
523void doStats(void)
524{
525  if(g_stats.qcounter) {
526    L<<Logger::Error<<"stats: "<<g_stats.qcounter<<" questions, "<<RC.size()<<" cache entries, "<<SyncRes::s_negcache.size()<<" negative entries, "
527     <<(int)((RC.cacheHits*100.0)/(RC.cacheHits+RC.cacheMisses))<<"% cache hits"<<endl;
528    L<<Logger::Error<<"stats: throttle map: "<<SyncRes::s_throttle.size()<<", ns speeds: "
529     <<SyncRes::s_nsSpeeds.size()<<endl; // ", bytes: "<<RC.bytes()<<endl;
530    L<<Logger::Error<<"stats: outpacket/query ratio "<<(int)(SyncRes::s_outqueries*100.0/SyncRes::s_queries)<<"%";
531    L<<Logger::Error<<", "<<(int)(SyncRes::s_throttledqueries*100.0/(SyncRes::s_outqueries+SyncRes::s_throttledqueries))<<"% throttled, "
532     <<SyncRes::s_nodelegated<<" no-delegation drops"<<endl;
533    L<<Logger::Error<<"stats: "<<SyncRes::s_tcpoutqueries<<" outgoing tcp connections, "<<MT->numProcesses()<<" queries running, "<<SyncRes::s_outgoingtimeouts<<" outgoing timeouts"<<endl;
534  }
535  else if(statsWanted) 
536    L<<Logger::Error<<"stats: no stats yet!"<<endl;
537
538  statsWanted=false;
539}
540
541static void houseKeeping(void *)
542{
543  static time_t last_stat, last_rootupdate, last_prune;
544  struct timeval now;
545  gettimeofday(&now, 0);
546
547  if(now.tv_sec - last_prune > 60) { 
548    DTime dt;
549    dt.setTimeval(now);
550    RC.doPrune();
551   
552    typedef SyncRes::negcache_t::nth_index<1>::type negcache_by_ttd_index_t;
553    negcache_by_ttd_index_t& ttdindex=boost::multi_index::get<1>(SyncRes::s_negcache);
554
555    negcache_by_ttd_index_t::iterator i=ttdindex.lower_bound(now.tv_sec);
556    ttdindex.erase(ttdindex.begin(), i);
557
558    time_t limit=now.tv_sec-300;
559    for(SyncRes::nsspeeds_t::iterator i = SyncRes::s_nsSpeeds.begin() ; i!= SyncRes::s_nsSpeeds.end(); )
560      if(i->second.stale(limit))
561        SyncRes::s_nsSpeeds.erase(i++);
562      else
563        ++i;
564
565    //    cerr<<"Pruned "<<pruned<<" records, left "<<SyncRes::s_negcache.size()<<"\n";
566//    cout<<"Prune took "<<dt.udiff()<<"usec\n";
567    last_prune=time(0);
568  }
569  if(now.tv_sec - last_stat>1800) { 
570    doStats();
571    last_stat=time(0);
572  }
573  if(now.tv_sec -last_rootupdate>7200) {
574    SyncRes sr(now);
575    vector<DNSResourceRecord> ret;
576
577    sr.setNoCache();
578    int res=sr.beginResolve("", QType(QType::NS), ret);
579    if(!res) {
580      L<<Logger::Error<<"Refreshed . records"<<endl;
581      last_rootupdate=now.tv_sec;
582    }
583    else
584      L<<Logger::Error<<"Failed to update . records, RCODE="<<res<<endl;
585  }
586}
587
588struct TCPConnection
589{
590  int fd;
591  enum {BYTE0, BYTE1, GETQUESTION} state;
592  int qlen;
593  int bytesread;
594  struct sockaddr_in remote;
595  char data[65535];
596  time_t startTime;
597};
598
599#if 0
600#include <execinfo.h>
601
602  multimap<uint32_t,string> rev;
603  for(map<string,uint32_t>::const_iterator i=casesptr->begin(); i!=casesptr->end(); ++i) {
604    rev.insert(make_pair(i->second,i->first));
605  }
606  for(multimap<uint32_t,string>::const_iterator i=rev.begin(); i!= rev.end(); ++i)
607    cout<<i->first<<" times: \n"<<i->second<<"\n";
608
609  cout.flush();
610
611map<string,uint32_t>* casesptr;
612static string maketrace()
613{
614  void *array[20]; //only care about last 17 functions (3 taken with tracing support)
615  size_t size;
616  char **strings;
617  size_t i;
618
619  size = backtrace (array, 5);
620  strings = backtrace_symbols (array, size); //Need -rdynamic gcc (linker) flag for this to work
621
622  string ret;
623
624  for (i = 0; i < size; i++) //skip useless functions
625    ret+=string(strings[i])+"\n";
626  return ret;
627}
628
629extern "C" {
630
631int gettimeofday (struct timeval *__restrict __tv,
632                  __timezone_ptr_t __tz)
633{
634  static map<string, uint32_t> s_cases;
635  casesptr=&s_cases;
636  s_cases[maketrace()]++;
637  __tv->tv_sec=time(0);
638  return 0;
639}
640
641}
642#endif
643
644int main(int argc, char **argv) 
645{
646  reportBasicTypes();
647
648  int ret = EXIT_SUCCESS;
649#ifdef WIN32
650    WSADATA wsaData;
651    WSAStartup( MAKEWORD( 2, 0 ), &wsaData );
652#endif // WIN32
653
654  try {
655    Utility::srandom(time(0));
656    ::arg().set("soa-minimum-ttl","Don't change")="0";
657    ::arg().set("soa-serial-offset","Don't change")="0";
658    ::arg().set("no-shuffle","Don't change")="off";
659    ::arg().set("aaaa-additional-processing","turn on to do AAAA additional processing (slow)")="off";
660    ::arg().set("local-port","port to listen on")="53";
661    ::arg().set("local-address","IP addresses to listen on, separated by spaces or commas")="0.0.0.0";
662    ::arg().set("trace","if we should output heaps of logging")="off";
663    ::arg().set("daemon","Operate as a daemon")="yes";
664    ::arg().set("chroot","switch to chroot jail")="";
665    ::arg().set("setgid","If set, change group id to this gid for more security")="";
666    ::arg().set("setuid","If set, change user id to this uid for more security")="";
667    ::arg().set("quiet","Suppress logging of questions and answers")="true";
668    ::arg().set("config-dir","Location of configuration directory (recursor.conf)")=SYSCONFDIR;
669    ::arg().set("socket-dir","Where the controlsocket will live")=LOCALSTATEDIR;
670    ::arg().set("delegation-only","Which domains we only accept delegations from")="";
671    ::arg().set("query-local-address","Source IP address for sending queries")="0.0.0.0";
672    ::arg().set("client-tcp-timeout","Timeout in seconds when talking to TCP clients")="2";
673    ::arg().set("max-tcp-clients","Maximum number of simultaneous TCP clients")="128";
674    ::arg().set("hint-file", "If set, load root hints from this file")="";
675
676    ::arg().setCmd("help","Provide a helpful message");
677    L.toConsole(Logger::Warning);
678    ::arg().laxParse(argc,argv); // do a lax parse
679
680    string configname=::arg()["config-dir"]+"/recursor.conf";
681    cleanSlashes(configname);
682
683    if(!::arg().file(configname.c_str())) 
684      L<<Logger::Warning<<"Unable to parse configuration file '"<<configname<<"'"<<endl;
685
686    ::arg().parse(argc,argv);
687
688    ::arg().set("delegation-only")=toLower(::arg()["delegation-only"]);
689
690    if(::arg().mustDo("help")) {
691      cerr<<"syntax:"<<endl<<endl;
692      cerr<<::arg().helpstring(::arg()["help"])<<endl;
693      exit(99);
694    }
695
696    L.setName("pdns_recursor");
697
698    L<<Logger::Warning<<"PowerDNS recursor "<<VERSION<<" (C) 2001-2006 PowerDNS.COM BV ("<<__DATE__", "__TIME__;
699#ifdef __GNUC__
700    L<<", gcc "__VERSION__;
701#endif // add other compilers here
702    L<<") starting up"<<endl;
703
704    L<<Logger::Warning<<"Operating in "<<(sizeof(unsigned long)*8) <<" bits mode"<<endl;
705  L<<Logger::Warning<<"PowerDNS comes with ABSOLUTELY NO WARRANTY. "
706    "This is free software, and you are welcome to redistribute it "
707    "according to the terms of the GPL version 2."<<endl;
708
709
710  g_quiet=::arg().mustDo("quiet");
711  if(::arg().mustDo("trace")) {
712      SyncRes::setLog(true);
713      ::arg().set("quiet")="no";
714      g_quiet=false;
715  }
716
717    makeClientSocket();
718    makeUDPServerSockets();
719    makeTCPServerSockets();
720    makeControlChannelSocket();
721   
722    MT=new MTasker<PacketID,string>(100000);
723
724    char data[1500];
725    struct sockaddr_in fromaddr;
726   
727    PacketID pident;
728    primeHints();   
729    L<<Logger::Warning<<"Done priming cache with root hints"<<endl;
730#ifndef WIN32
731    if(::arg().mustDo("daemon")) {
732      L.toConsole(Logger::Critical);
733      daemonize();
734    }
735    signal(SIGUSR1,usr1Handler);
736    signal(SIGUSR2,usr2Handler);
737    signal(SIGPIPE,SIG_IGN);
738
739    writePid();
740#endif
741
742    int newgid=0;
743    if(!::arg()["setgid"].empty())
744      newgid=Utility::makeGidNumeric(::arg()["setgid"]);
745    int newuid=0;
746    if(!::arg()["setuid"].empty())
747      newuid=Utility::makeUidNumeric(::arg()["setuid"]);
748
749
750    if (!::arg()["chroot"].empty()) {
751        if (chroot(::arg()["chroot"].c_str())<0) {
752            L<<Logger::Error<<"Unable to chroot to '"+::arg()["chroot"]+"': "<<strerror (errno)<<", exiting"<<endl;
753            exit(1);
754        }
755    }
756
757    Utility::dropPrivs(newuid, newgid);
758
759    vector<TCPConnection> tcpconnections;
760    counter=0;
761    struct timeval now;
762    unsigned int maxTcpClients=::arg().asNum("max-tcp-clients");
763    int tcpLimit=::arg().asNum("client-tcp-timeout");
764    for(;;) {
765      while(MT->schedule()); // housekeeping, let threads do their thing
766     
767      if(!((counter++)%500)) 
768        MT->makeThread(houseKeeping,0,"housekeeping");
769      if(statsWanted) {
770        doStats();
771      }
772
773      Utility::socklen_t addrlen=sizeof(fromaddr);
774      int d_len;
775     
776      struct timeval tv;
777      tv.tv_sec=0;
778      tv.tv_usec=500000;
779     
780      fd_set readfds, writefds;
781      FD_ZERO( &readfds );
782      FD_ZERO( &writefds );
783      FD_SET( d_clientsock, &readfds );
784      FD_SET( s_rcc.d_fd, &readfds);
785      int fdmax=max(d_clientsock, s_rcc.d_fd);
786
787      if(!tcpconnections.empty())
788        gettimeofday(&now, 0);
789
790      vector<TCPConnection> sweeped;
791
792      for(vector<TCPConnection>::iterator i=tcpconnections.begin();i!=tcpconnections.end();++i) {
793        if(now.tv_sec < i->startTime + tcpLimit) {
794          FD_SET(i->fd, &readfds);
795          fdmax=max(fdmax,i->fd);
796          sweeped.push_back(*i);
797        }
798        else {
799          L<<Logger::Error<<"TCP timeout from client "<<inet_ntoa(i->remote.sin_addr)<<endl;
800          close(i->fd);
801        }
802      }
803      sweeped.swap(tcpconnections);
804
805      for(vector<int>::const_iterator i=d_udpserversocks.begin(); i!=d_udpserversocks.end(); ++i) {
806        FD_SET( *i, &readfds );
807        fdmax=max(fdmax,*i);
808      }
809      if(tcpconnections.size() < maxTcpClients) 
810        for(tcpserversocks_t::const_iterator i=s_tcpserversocks.begin(); i!=s_tcpserversocks.end(); ++i) {
811          FD_SET(*i, &readfds );
812          fdmax=max(fdmax,*i);
813        }
814
815      for(map<int,PacketID>::const_iterator i=d_tcpclientreadsocks.begin(); i!=d_tcpclientreadsocks.end(); ++i) {
816        // cerr<<"Adding TCP socket "<<i->first<<" to read select set"<<endl;
817        FD_SET( i->first, &readfds );
818        fdmax=max(fdmax,i->first);
819      }
820
821      for(map<int,PacketID>::const_iterator i=d_tcpclientwritesocks.begin(); i!=d_tcpclientwritesocks.end(); ++i) {
822        // cerr<<"Adding TCP socket "<<i->first<<" to write select set"<<endl;
823        FD_SET( i->first, &writefds );
824        fdmax=max(fdmax,i->first);
825      }
826
827      int selret = select(  fdmax + 1, &readfds, &writefds, NULL, &tv );
828      gettimeofday(&now, 0);
829      if(selret<=0) 
830        if (selret == -1 && errno!=EINTR) 
831          throw AhuException("Select returned: "+stringerror());
832        else
833          continue;
834
835      if(FD_ISSET(s_rcc.d_fd, &readfds)) {
836        string remote;
837        string msg=s_rcc.recv(&remote);
838        RecursorControlParser rcp;
839        RecursorControlParser::func_t* command;
840        string answer=rcp.getAnswer(msg, &command);
841        s_rcc.send(answer, &remote);
842        command();
843      }
844
845      if(FD_ISSET(d_clientsock,&readfds)) { // do we have a UDP question response?
846        d_len=recvfrom(d_clientsock, data, sizeof(data), 0, (sockaddr *)&fromaddr, &addrlen);   
847        if(d_len<0) 
848          continue;
849
850        try {
851          DNSComboWriter dc(data, d_len, now);
852          dc.setRemote((struct sockaddr *)&fromaddr, addrlen);
853
854          if(dc.d_mdp.d_header.qr) {
855            pident.remote=fromaddr;
856            pident.id=dc.d_mdp.d_header.id;
857            string packet;
858            packet.assign(data, d_len);
859            MT->sendEvent(pident, &packet);
860          }
861          else 
862            L<<Logger::Warning<<"Ignoring question on outgoing socket from "<<dc.getRemote()<<endl;
863        }
864        catch(MOADNSException& mde) {
865          L<<Logger::Error<<"Unparseable packet from remote server "<< sockAddrToString((struct sockaddr_in*) &fromaddr, addrlen) <<": "<<mde.what()<<endl;
866        }
867      }
868     
869      for(vector<int>::const_iterator i=d_udpserversocks.begin(); i!=d_udpserversocks.end(); ++i) {
870        if(FD_ISSET(*i,&readfds)) { // do we have a new question on udp?
871          d_len=recvfrom(*i, data, sizeof(data), 0, (sockaddr *)&fromaddr, &addrlen);   
872          if(d_len<0) 
873            continue;
874
875          g_stats.queryrate.pulse(now);
876
877          try {
878            DNSComboWriter* dc = new DNSComboWriter(data, d_len, now);
879
880            dc->setRemote((struct sockaddr *)&fromaddr, addrlen);
881
882            if(dc->d_mdp.d_header.qr)
883              L<<Logger::Error<<"Ignoring answer on server socket!"<<endl;
884            else {
885              ++g_stats.qcounter;
886              dc->setSocket(*i);
887              dc->d_tcp=false;
888              MT->makeThread(startDoResolve, (void*) dc, "udp");
889            }
890          }
891          catch(MOADNSException& mde) {
892            L<<Logger::Error<<"Unparseable packet from remote server "<< sockAddrToString((struct sockaddr_in*) &fromaddr, addrlen) <<": "<<mde.what()<<endl;
893          }
894        }
895      }
896
897      for(tcpserversocks_t::const_iterator i=s_tcpserversocks.begin(); i!=s_tcpserversocks.end(); ++i) { 
898        if(FD_ISSET(*i ,&readfds)) { // do we have a new TCP connection?
899          struct sockaddr_in addr;
900          socklen_t addrlen=sizeof(addr);
901          int newsock=accept(*i, (struct sockaddr*)&addr, &addrlen);
902         
903          if(newsock>0) {
904            Utility::setNonBlocking(newsock);
905            TCPConnection tc;
906            tc.fd=newsock;
907            tc.state=TCPConnection::BYTE0;
908            tc.remote=addr;
909            tc.startTime=now.tv_sec;
910            tcpconnections.push_back(tc);
911          }
912        }
913      }
914
915      // have any question answers come in over TCP?
916      for(map<int,PacketID>::iterator i=d_tcpclientreadsocks.begin(); i!=d_tcpclientreadsocks.end();) { 
917        bool haveErased=false;
918        if(FD_ISSET(i->first, &readfds)) { // can we receive
919          shared_array<char> buffer(new char[i->second.inNeeded]);
920
921          int ret=read(i->first, buffer.get(), min(i->second.inNeeded,200));
922          // cerr<<"Read returned "<<ret<<endl;
923          if(ret > 0) {
924            i->second.inMSG.append(&buffer[0], &buffer[ret]);
925            i->second.inNeeded-=ret;
926            if(!i->second.inNeeded) {
927              // cerr<<"Got entire load of "<<i->second.inMSG.size()<<" bytes"<<endl;
928              PacketID pid=i->second;
929              string msg=i->second.inMSG;
930             
931              d_tcpclientreadsocks.erase((i++));
932              haveErased=true;
933              MT->sendEvent(pid, &msg);   // XXX DODGY
934            }
935            else {
936              // cerr<<"Still have "<<i->second.inNeeded<<" left to go"<<endl;
937            }
938          }
939          else {
940            //      cerr<<"when reading ret="<<ret<<endl;
941            // XXX FIXME I think some stuff needs to happen here - like send an EOF event
942          }
943        }
944        if(!haveErased)
945          ++i;
946      }
947     
948      // is there data we can send to remote nameservers over TCP?
949      for(map<int,PacketID>::iterator i=d_tcpclientwritesocks.begin(); i!=d_tcpclientwritesocks.end(); ) { 
950        bool haveErased=false;
951        if(FD_ISSET(i->first, &writefds)) { // can we send over TCP
952          // cerr<<"Socket "<<i->first<<" available for writing"<<endl;
953          int ret=write(i->first, i->second.outMSG.c_str(), i->second.outMSG.size() - i->second.outPos);
954          if(ret > 0) {
955            i->second.outPos+=ret;
956            if(i->second.outPos==i->second.outMSG.size()) {
957              // cerr<<"Sent out entire load of "<<i->second.outMSG.size()<<" bytes"<<endl;
958              PacketID pid=i->second;
959              d_tcpclientwritesocks.erase(i++);   // erase!
960              haveErased=true;
961              MT->sendEvent(pid, 0);
962            }
963
964          }
965          else { 
966            //      cerr<<"ret="<<ret<<" when writing"<<endl;
967            // XXX FIXME I think some stuff needs to happen here - like send an EOF event
968          }
969        }
970        if(!haveErased)
971          ++i;
972      }
973     
974      // very braindead TCP incoming question parser
975      for(vector<TCPConnection>::iterator i=tcpconnections.begin();i!=tcpconnections.end();++i) {
976        if(FD_ISSET(i->fd, &readfds)) {
977          if(i->state==TCPConnection::BYTE0) {
978            int bytes=read(i->fd,i->data,2);
979            if(bytes==1)
980              i->state=TCPConnection::BYTE1;
981            if(bytes==2) { 
982              i->qlen=(i->data[0]<<8)+i->data[1];
983              i->bytesread=0;
984              i->state=TCPConnection::GETQUESTION;
985            }
986            if(!bytes || bytes < 0) {
987              close(i->fd);
988              tcpconnections.erase(i);
989              break;
990            }
991          }
992          else if(i->state==TCPConnection::BYTE1) {
993            int bytes=read(i->fd,i->data+1,1);
994            if(bytes==1) {
995              i->state=TCPConnection::GETQUESTION;
996              i->qlen=(i->data[0]<<8)+i->data[1];
997              i->bytesread=0;
998            }
999            if(!bytes || bytes < 0) {
1000              L<<Logger::Error<<"TCP Remote "<<sockAddrToString(&i->remote,sizeof(i->remote))<<" disconnected after first byte"<<endl;
1001              close(i->fd);
1002              tcpconnections.erase(i);
1003              break;
1004            }
1005           
1006          }
1007          else if(i->state==TCPConnection::GETQUESTION) {
1008            int bytes=read(i->fd,i->data + i->bytesread,i->qlen - i->bytesread);
1009            if(!bytes || bytes < 0) {
1010              L<<Logger::Error<<"TCP Remote "<<sockAddrToString(&i->remote,sizeof(i->remote))<<" disconnected while reading question body"<<endl;
1011              close(i->fd);
1012              tcpconnections.erase(i);
1013              break;
1014            }
1015            i->bytesread+=bytes;
1016            if(i->bytesread==i->qlen) {
1017              i->state=TCPConnection::BYTE0;
1018              DNSComboWriter* dc=0;
1019              try {
1020                dc=new DNSComboWriter(i->data, i->qlen, now);
1021              }
1022              catch(MOADNSException &mde) {
1023                L<<Logger::Error<<"Unparseable packet from remote client "<<sockAddrToString(&i->remote,sizeof(i->remote))<<endl;
1024                close(i->fd);
1025                tcpconnections.erase(i);
1026                break;
1027              }
1028
1029              dc->setSocket(i->fd);
1030              dc->d_tcp=true;
1031              dc->setRemote((struct sockaddr *)&i->remote,sizeof(i->remote));
1032              if(dc->d_mdp.d_header.qr)
1033                L<<Logger::Error<<"Ignoring answer on server socket!"<<endl;
1034              else {
1035                ++g_stats.qcounter;
1036                ++g_stats.tcpqcounter;
1037                MT->makeThread(startDoResolve, dc, "tcp");
1038              }
1039            }
1040          }
1041        }
1042      }
1043    }
1044  }
1045  catch(AhuException &ae) {
1046    L<<Logger::Error<<"Exception: "<<ae.reason<<endl;
1047    ret=EXIT_FAILURE;
1048  }
1049  catch(exception &e) {
1050    L<<Logger::Error<<"STL Exception: "<<e.what()<<endl;
1051    ret=EXIT_FAILURE;
1052  }
1053  catch(...) {
1054    L<<Logger::Error<<"any other exception in main: "<<endl;
1055    ret=EXIT_FAILURE;
1056  }
1057 
1058#ifdef WIN32
1059  WSACleanup();
1060#endif // WIN32
1061
1062  return ret;
1063}
Note: See TracBrowser for help on using the browser.