pdns/nameserver.cc

   1 /*
   2  * This file is part of PowerDNS or dnsdist.
   3  * Copyright -- PowerDNS.COM B.V. and its contributors
   4  *
   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.
   8  *
   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.
  12  *
  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.
  17  *
  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.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include "config.h"
  24 #endif
  25 #include "utility.hh"
  26 #include <cstdio>
  27 #include <cstring>
  28 #include <cstdlib>
  29 #include <cerrno>
  30 #include <iostream>
  31 #include <string>
  32 #include <sys/types.h>
  33 #include "responsestats.hh"
  34
  35 #include "dns.hh"
  36 #include "dnsbackend.hh"
  37 #include "dnspacket.hh"
  38 #include "nameserver.hh"
  39 #include "distributor.hh"
  40 #include "logger.hh"
  41 #include "arguments.hh"
  42 #include "statbag.hh"
  43
  44 #include "namespaces.hh"
  45
  46 extern StatBag S;
  47
  48 /** \mainpage
  49     PowerDNS is a very versatile nameserver that can answer questions from different backends. To implement your
  50     own backend, see the documentation for the DNSBackend class.
  51
  52     \section copyright Copyright and License
  53     PowerDNS is (C) 2001-2008 PowerDNS.COM BV. It is distributed according to the terms of the General Public License version 2.
  54
  55     \section overview High level overview
  56
  57     The Distributor contains a configurable number of PacketHandler instances, each in its own thread, for connection pooling.
  58     PacketHandler instances are recycled of they let escape an PDNSException.
  59
  60     The PacketHandler implements the RFC1034 algorithm and converts question packets into DNSBackend queries.
  61
  62     A DNSBackend is an entity that returns DNSResourceRecord objects in return to explicit questions for domains with a specified QType
  63
  64     PowerDNS uses the UeberBackend, which hosts DNSBackends. By default it has no DNSBackends within itself, those are loaded
  65     by setting --load=<list of backends>. This way DNSBackend implementations can be kept completely separate, but most aren't.
  66
  67     If one or more DNSBackends are loaded, the UeberBackend fields the queries to all of them until one answers.
  68
  69     \section TCP TCP Operations
  70
  71     The TCP operation runs within a single thread called tcpreceiver(), that also queries the PacketHandler.
  72
  73     \section Cache Caching
  74
  75     On its own, this setup is not suitable for high performance operations. A single DNS query can turn into many DNSBackend questions,
  76     each taking many milliseconds to complete. This is why the qthread() first checks the PacketCache to see if an answer is known to a packet
  77     asking this question. If so, the entire Distributor is shunted, and the answer is sent back *directly*, within a few microseconds.
  78
  79     \section misc Miscellaneous
  80     Configuration details are available via the ArgvMap instance arg. Statistics are created by making calls to the StatBag object called S.
  81     These statistics are made available via the UeberBackend on the same socket that is used for dynamic module commands.
  82
  83     \section Main Main
  84     The main() of PowerDNS can be found in receiver.cc - start reading there for further insights into the operation of the nameserver
  85 */
  86
  87 vector<ComboAddress> g_localaddresses; // not static, our unit tests need to poke this
  88
  89 void UDPNameserver::bindIPv4()
  90 {
  91   vector<string>locals;
  92   stringtok(locals,::arg()["local-address"]," ,");
  93   int one = 1;
  94
  95   if(locals.empty())
  96     throw PDNSException("No local address specified");
  97
  98   int s;
  99   for(vector<string>::const_iterator i=locals.begin();i!=locals.end();++i) {
 100     string localname(*i);
 101     ComboAddress locala;
 102
 103     s=socket(AF_INET,SOCK_DGRAM,0);
 104
 105     if(s<0) {
 106       g_log<<Logger::Error<<"Unable to acquire UDP socket: "+string(strerror(errno)) << endl;
 107       throw PDNSException("Unable to acquire a UDP socket: "+string(strerror(errno)));
 108     }
 109
 110     setCloseOnExec(s);
 111
 112     if(!setNonBlocking(s))
 113       throw PDNSException("Unable to set UDP socket to non-blocking: "+stringerror());
 114
 115     locala.reset();
 116     locala.sin4.sin_family=AF_INET;
 117
 118     if(localname=="0.0.0.0")
 119       setsockopt(s, IPPROTO_IP, GEN_IP_PKTINFO, &one, sizeof(one));
 120
 121     if (!setSocketTimestamps(s))
 122       g_log<<Logger::Warning<<"Unable to enable timestamp reporting for socket"<<endl;
 123
 124 #ifdef SO_REUSEPORT
 125     if( d_can_reuseport )
 126         if( setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one)) )
 127           d_can_reuseport = false;
 128 #endif
 129
 130     if( ::arg().mustDo("non-local-bind") )
 131         Utility::setBindAny(AF_INET, s);
 132
 133     locala=ComboAddress(localname, ::arg().asNum("local-port"));
 134     if(locala.sin4.sin_family != AF_INET)
 135       throw PDNSException("Attempting to bind IPv4 socket to IPv6 address");
 136
 137     if( !d_additional_socket )
 138         g_localaddresses.push_back(locala);
 139
 140     if(::bind(s, (sockaddr*)&locala, locala.getSocklen()) < 0) {
 141       string binderror = strerror(errno);
 142       close(s);
 143       if( errno == EADDRNOTAVAIL && ! ::arg().mustDo("local-address-nonexist-fail") ) {
 144         g_log<<Logger::Error<<"IPv4 Address " << localname << " does not exist on this server - skipping UDP bind" << endl;
 145         continue;
 146       } else {
 147         g_log<<Logger::Error<<"Unable to bind UDP socket to '"+locala.toStringWithPort()+"': "<<binderror<<endl;
 148         throw PDNSException("Unable to bind to UDP socket");
 149       }
 150     }
 151     d_sockets.push_back(s);
 152     g_log<<Logger::Error<<"UDP server bound to "<<locala.toStringWithPort()<<endl;
 153     struct pollfd pfd;
 154     pfd.fd = s;
 155     pfd.events = POLLIN;
 156     pfd.revents = 0;
 157     d_rfds.push_back(pfd);
 158   }
 159 }
 160
 161 bool AddressIsUs(const ComboAddress& remote)
 162 {
 163   for(const ComboAddress& us :  g_localaddresses) {
 164     if(remote == us)
 165       return true;
 166     if(IsAnyAddress(us)) {
 167       int s = socket(remote.sin4.sin_family, SOCK_DGRAM, 0);
 168       if(s < 0)
 169         continue;
 170
 171       if(connect(s, (struct sockaddr*)&remote, remote.getSocklen()) < 0) {
 172         close(s);
 173         continue;
 174       }
 175
 176       ComboAddress actualLocal;
 177       actualLocal.sin4.sin_family = remote.sin4.sin_family;
 178       socklen_t socklen = actualLocal.getSocklen();
 179
 180       if(getsockname(s, (struct sockaddr*) &actualLocal, &socklen) < 0) {
 181         close(s);
 182         continue;
 183       }
 184       close(s);
 185       actualLocal.sin4.sin_port = us.sin4.sin_port;
 186       if(actualLocal == remote)
 187         return true;
 188     }
 189   }
 190   return false;
 191 }
 192
 193
 194 void UDPNameserver::bindIPv6()
 195 {
 196   vector<string> locals;
 197   stringtok(locals,::arg()["local-ipv6"]," ,");
 198   int one=1;
 199
 200   if(locals.empty())
 201     return;
 202
 203   int s;
 204   for(vector<string>::const_iterator i=locals.begin();i!=locals.end();++i) {
 205     string localname(*i);
 206
 207     s=socket(AF_INET6,SOCK_DGRAM,0);
 208     if(s<0) {
 209       if( errno == EAFNOSUPPORT ) {
 210         g_log<<Logger::Error<<"IPv6 Address Family is not supported - skipping UDPv6 bind" << endl;
 211         return;
 212       } else {
 213         g_log<<Logger::Error<<"Unable to acquire a UDPv6 socket: "+string(strerror(errno)) << endl;
 214         throw PDNSException("Unable to acquire a UDPv6 socket: "+string(strerror(errno)));
 215       }
 216     }
 217
 218     setCloseOnExec(s);
 219     if(!setNonBlocking(s))
 220       throw PDNSException("Unable to set UDPv6 socket to non-blocking: "+stringerror());
 221
 222     ComboAddress locala(localname, ::arg().asNum("local-port"));
 223
 224     if(IsAnyAddress(locala)) {
 225       setsockopt(s, IPPROTO_IP, GEN_IP_PKTINFO, &one, sizeof(one));     // linux supports this, so why not - might fail on other systems
 226 #ifdef IPV6_RECVPKTINFO
 227       setsockopt(s, IPPROTO_IPV6, IPV6_RECVPKTINFO, &one, sizeof(one));
 228 #endif
 229       setsockopt(s, IPPROTO_IPV6, IPV6_V6ONLY, &one, sizeof(one));      // if this fails, we report an error in tcpreceiver too
 230     }
 231
 232     if (!setSocketTimestamps(s))
 233       g_log<<Logger::Warning<<"Unable to enable timestamp reporting for socket"<<endl;
 234
 235 #ifdef SO_REUSEPORT
 236     if( d_can_reuseport )
 237         if( setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one)) )
 238           d_can_reuseport = false;
 239 #endif
 240
 241     if( ::arg().mustDo("non-local-bind") )
 242         Utility::setBindAny(AF_INET6, s);
 243
 244     if( !d_additional_socket )
 245         g_localaddresses.push_back(locala);
 246     if(::bind(s, (sockaddr*)&locala, sizeof(locala))<0) {
 247       close(s);
 248       if( errno == EADDRNOTAVAIL && ! ::arg().mustDo("local-ipv6-nonexist-fail") ) {
 249         g_log<<Logger::Error<<"IPv6 Address " << localname << " does not exist on this server - skipping UDP bind" << endl;
 250         continue;
 251       } else {
 252         g_log<<Logger::Error<<"Unable to bind to UDPv6 socket "<< localname <<": "<<strerror(errno)<<endl;
 253         throw PDNSException("Unable to bind to UDPv6 socket");
 254       }
 255     }
 256     d_sockets.push_back(s);
 257     struct pollfd pfd;
 258     pfd.fd = s;
 259     pfd.events = POLLIN;
 260     pfd.revents = 0;
 261     d_rfds.push_back(pfd);
 262     g_log<<Logger::Error<<"UDPv6 server bound to "<<locala.toStringWithPort()<<endl;
 263   }
 264 }
 265
 266 UDPNameserver::UDPNameserver( bool additional_socket )
 267 {
 268 #ifdef SO_REUSEPORT
 269   d_can_reuseport = ::arg().mustDo("reuseport");
 270 #endif
 271   // Are we the main socket (false) or a rebinding using SO_REUSEPORT ?
 272   d_additional_socket = additional_socket;
 273
 274   if(!::arg()["local-address"].empty())
 275     bindIPv4();
 276   if(!::arg()["local-ipv6"].empty())
 277     bindIPv6();
 278
 279   if(::arg()["local-address"].empty() && ::arg()["local-ipv6"].empty())
 280     g_log<<Logger::Critical<<"PDNS is deaf and mute! Not listening on any interfaces"<<endl;
 281 }
 282
 283 void UDPNameserver::send(DNSPacket *p)
 284 {
 285   string buffer=p->getString();
 286   g_rs.submitResponse(*p, true);
 287
 288   struct msghdr msgh;
 289   struct iovec iov;
 290   char cbuf[256];
 291
 292   fillMSGHdr(&msgh, &iov, cbuf, 0, (char*)buffer.c_str(), buffer.length(), &p->d_remote);
 293
 294   msgh.msg_control=NULL;
 295   if(p->d_anyLocal) {
 296     addCMsgSrcAddr(&msgh, cbuf, p->d_anyLocal.get_ptr(), 0);
 297   }
 298   DLOG(g_log<<Logger::Notice<<"Sending a packet to "<< p->getRemote() <<" ("<< buffer.length()<<" octets)"<<endl);
 299   if(buffer.length() > p->getMaxReplyLen()) {
 300     g_log<<Logger::Error<<"Weird, trying to send a message that needs truncation, "<< buffer.length()<<" > "<<p->getMaxReplyLen()<<". Question was for "<<p->qdomain<<"|"<<p->qtype.getName()<<endl;
 301   }
 302   if(sendmsg(p->getSocket(), &msgh, 0) < 0)
 303     g_log<<Logger::Error<<"Error sending reply with sendmsg (socket="<<p->getSocket()<<", dest="<<p->d_remote.toStringWithPort()<<"): "<<strerror(errno)<<endl;
 304 }
 305
 306 DNSPacket *UDPNameserver::receive(DNSPacket *prefilled, std::string& buffer)
 307 {
 308   ComboAddress remote;
 309   extern StatBag S;
 310   ssize_t len=-1;
 311   Utility::sock_t sock=-1;
 312
 313   struct msghdr msgh;
 314   struct iovec iov;
 315   char cbuf[256];
 316
 317   remote.sin6.sin6_family=AF_INET6; // make sure it is big enough
 318   fillMSGHdr(&msgh, &iov, cbuf, sizeof(cbuf), &buffer.at(0), buffer.size(), &remote);
 319
 320   int err;
 321   vector<struct pollfd> rfds= d_rfds;
 322
 323   for(auto &pfd :  rfds) {
 324     pfd.events = POLLIN;
 325     pfd.revents = 0;
 326   }
 327
 328   retry:;
 329
 330   err = poll(&rfds[0], rfds.size(), -1);
 331   if(err < 0) {
 332     if(errno==EINTR)
 333       goto retry;
 334     unixDie("Unable to poll for new UDP events");
 335   }
 336
 337   for(auto &pfd :  rfds) {
 338     if(pfd.revents & POLLIN) {
 339       sock=pfd.fd;
 340       if((len=recvmsg(sock, &msgh, 0)) < 0 ) {
 341         if(errno != EAGAIN)
 342           g_log<<Logger::Error<<"recvfrom gave error, ignoring: "<<strerror(errno)<<endl;
 343         return 0;
 344       }
 345       break;
 346     }
 347   }
 348   if(sock==-1)
 349     throw PDNSException("poll betrayed us! (should not happen)");
 350
 351   DLOG(g_log<<"Received a packet " << len <<" bytes long from "<< remote.toString()<<endl);
 352
 353   BOOST_STATIC_ASSERT(offsetof(sockaddr_in, sin_port) == offsetof(sockaddr_in6, sin6_port));
 354
 355   if(remote.sin4.sin_port == 0) // would generate error on responding. sin4 also works for ipv6
 356     return 0;
 357
 358   DNSPacket *packet;
 359   if(prefilled)  // they gave us a preallocated packet
 360     packet=prefilled;
 361   else
 362     packet=new DNSPacket(true); // don't forget to free it!
 363
 364   packet->setSocket(sock);
 365   packet->setRemote(&remote);
 366
 367   ComboAddress dest;
 368   if(HarvestDestinationAddress(&msgh, &dest)) {
 369 //    cerr<<"Setting d_anyLocal to '"<<dest.toString()<<"'"<<endl;
 370     packet->d_anyLocal = dest;
 371   }
 372
 373   struct timeval recvtv;
 374   if(HarvestTimestamp(&msgh, &recvtv)) {
 375     packet->d_dt.setTimeval(recvtv);
 376   }
 377   else
 378     packet->d_dt.set(); // timing
 379
 380   if(packet->parse(&buffer.at(0), (size_t) len)<0) {
 381     S.inc("corrupt-packets");
 382     S.ringAccount("remotes-corrupt", packet->d_remote);
 383
 384     if(!prefilled)
 385       delete packet;
 386     return 0; // unable to parse
 387   }
 388
 389   return packet;
 390 }