added logic to deal with dns replies to old dns requests --> no more consistency...
[pingcheck] / src / dns / dnsresolver.cpp
CommitLineData
36ad976b
CH
1/*
2 The software in this package is distributed under the GNU General
3 Public License version 2 (with a special exception described below).
4
5 A copy of GNU General Public License (GPL) is included in this distribution,
6 in the file COPYING.GPL.
7
8 As a special exception, if other files instantiate templates or use macros
9 or inline functions from this file, or you compile this file and link it
10 with other works to produce a work based on this file, this file
11 does not by itself cause the resulting work to be covered
12 by the GNU General Public License.
13
14 However the source code for this file must still be made available
15 in accordance with section (3) of the GNU General Public License.
16
17 This exception does not invalidate any other reasons why a work based
18 on this file might be covered by the GNU General Public License.
19
20 Christian Herdtweck, Intra2net AG 2015
4e7b6ff9
CH
21
22 with code copied from boost::net::dns::resolve.hpp
23 by Andreas Haberstroh (andreas at ibusy dot com)
24 from https://github.com/softwareace/Boost.DNS
36ad976b
CH
25 */
26
c5b4902d
CH
27#include "dns/dnsresolver.h"
28
29#include <stdint.h>
e91538f0 30#include <sstream>
36ad976b 31
4e7b6ff9
CH
32#include <boost/foreach.hpp>
33#include <boost/bind.hpp>
34#include <boost/function.hpp>
35#include <boost/net/dns.hpp>
36#include <boost/date_time/posix_time/posix_time.hpp>
ad83004d
CH
37#include <boost/uuid/uuid.hpp>
38#include <boost/uuid/uuid_io.hpp>
4e7b6ff9 39
36ad976b 40#include <logfunc.hpp>
4e7b6ff9 41
36ad976b 42using I2n::Logger::GlobalLogger;
4e7b6ff9 43using boost::posix_time::seconds;
36ad976b
CH
44
45namespace Config
46{
fd62d09f
CH
47 const int ResolveTimeoutSeconds = 2;
48 const int PauseBeforeRetrySeconds = 1;
49 const int StaleDataLongtermSeconds = 5*60;
36ad976b 50 const int DNS_PORT = 53;
36ad976b
CH
51}
52
4e7b6ff9
CH
53DnsResolver::DnsResolver(IoServiceItem &io_serv,
54 const std::string &hostname,
923626c0 55 const DnsIpProtocol &protocol,
96779587 56 const DnsCacheItem cache,
96779587 57 const boost::asio::ip::address &name_server)
4e7b6ff9 58 : ResolverBase( io_serv, hostname, cache )
e91538f0
CH
59 , Socket( *io_serv, ip::udp::endpoint(ip::udp::v4(), 0))
60 // just connect to anything, will specify sender/receiver later
4e7b6ff9 61 , ReceiveBuffer()
ad83004d 62 , RequestBuffer()
923626c0 63 , Protocol( protocol )
4e7b6ff9 64 , NameServer( name_server, Config::DNS_PORT )
36ad976b
CH
65 , ResolveTimeoutTimer( *io_serv )
66 , PauseBeforeRetryTimer( *io_serv )
67 , StaleDataLongtermTimer( *io_serv )
4e7b6ff9 68 , NextIpIndex( 0 )
36ad976b
CH
69 , RetryCount( 0 )
70 , IsResolving( false )
e91538f0 71 , LogPrefix( "DnsResolver" )
ad83004d
CH
72 , RandomIdGenerator()
73 , RequestId( 0 )
e18c1337 74 , OperationCancelled( false )
fd62d09f 75 , LongtermTimerIsActive( false )
e91538f0
CH
76{
77 std::stringstream temp;
78 temp << "Dns(" << ResolverBase::Hostname << "): ";
79 LogPrefix = temp.str();
ad83004d 80
e91538f0 81}
96779587 82
c5b4902d
CH
83DnsResolver::~DnsResolver()
84{
85 boost::system::error_code error;
e18c1337
CH
86 //Socket.shutdown(boost::asio::ip::udp::socket::shutdown_both, error);
87 //if ( error )
49b82a1d 88 // GlobalLogger.info() << LogPrefix << "Received error " << error
e18c1337
CH
89 // << " when shutting down socket for DNS";
90 // in IcmpPinger always gave an error system:9 (EBADF: Bad file descriptor)
91 // Here gives error system:107 ENOTCONN: Transport endpoint is not connected
c5b4902d
CH
92
93 Socket.close(error);
94 if ( error )
49b82a1d
CH
95 GlobalLogger.info() << LogPrefix << "Received error " << error
96 << " when closing socket for DNS";
c5b4902d
CH
97}
98
99
96779587
CH
100
101//==============================================================================
102// ASYNC RESOLVE
103//==============================================================================
36ad976b
CH
104
105/**
c5b4902d 106 * copied here code from boost::net::dns::resolve.hpp, since want async
36ad976b 107 * operation and that is used only internally, there
36ad976b 108 */
cd71d095 109void DnsResolver::do_resolve(const int recursion_count)
36ad976b 110{
36ad976b 111 // check if resolving already
4e7b6ff9 112 if (IsResolving)
36ad976b 113 {
e91538f0 114 GlobalLogger.info() << LogPrefix
4e7b6ff9 115 << "Call to do_resolve ignored since resolving already";
36ad976b
CH
116 return;
117 }
ad83004d 118 IsResolving = true;
e18c1337 119 OperationCancelled = false;
36ad976b 120
e18c1337
CH
121 GlobalLogger.info() << LogPrefix << "start resolving for IPs of type "
122 << to_string(Protocol) << " using name server " << NameServer;
e91538f0 123
36ad976b
CH
124 // just to be sure: cancel timers
125 ResolveTimeoutTimer.cancel();
126 PauseBeforeRetryTimer.cancel();
127 StaleDataLongtermTimer.cancel();
fd62d09f 128 LongtermTimerIsActive = false;
36ad976b
CH
129
130 // create DNS request
e18c1337 131 boost::net::dns::message dns_message( ResolverBase::Hostname, Protocol );
23f51766 132 dns_message.recursive(true);
4e7b6ff9
CH
133 dns_message.action(boost::net::dns::message::query);
134 dns_message.opcode(boost::net::dns::message::squery);
ad83004d
CH
135
136 // create random ID for message
137 boost::uuids::uuid message_id = RandomIdGenerator();
138 memcpy( &RequestId, message_id.data, sizeof(RequestId) );
139 dns_message.id( RequestId );
140 GlobalLogger.debug() << LogPrefix << "Request has ID "
cd71d095 141 << std::showbase << std::hex << dns_message.id();
36ad976b
CH
142
143 // setup receipt of reply
144 Socket.async_receive_from(
145 boost::asio::buffer(ReceiveBuffer.get_array()),
146 NameServer,
147 boost::bind( &DnsResolver::handle_dns_result, this,
cd71d095 148 recursion_count,
36ad976b
CH
149 boost::asio::placeholders::error,
150 boost::asio::placeholders::bytes_transferred)
4e7b6ff9 151 );
36ad976b
CH
152
153 // schedule timeout
4e7b6ff9
CH
154 (void) ResolveTimeoutTimer.expires_from_now(
155 seconds(Config::ResolveTimeoutSeconds));
156 ResolveTimeoutTimer.async_wait( boost::bind(
157 &DnsResolver::handle_resolve_timeout,
cd71d095
CH
158 this, recursion_count,
159 boost::asio::placeholders::error) );
36ad976b
CH
160
161 // send dns request
ad83004d 162 dns_message.encode(RequestBuffer);
e91538f0
CH
163 size_t bytes_sent;
164 try
165 {
166 bytes_sent = Socket.send_to(
ad83004d 167 boost::asio::buffer(RequestBuffer.get_array()),
e91538f0
CH
168 NameServer );
169 }
170 catch (boost::system::system_error &err)
171 {
49b82a1d 172 GlobalLogger.info() << LogPrefix
e91538f0
CH
173 << "Sending of DNS request message failed: "
174 << err.what();
cd71d095 175 schedule_retry(recursion_count);
e91538f0
CH
176 return;
177 }
178
179 if ( bytes_sent == 0 )
180 {
49b82a1d 181 GlobalLogger.info() << LogPrefix << "Empty DNS request sent!";
cd71d095 182 schedule_retry(recursion_count);
e91538f0
CH
183 return;
184 }
36ad976b
CH
185}
186
187
cd71d095
CH
188void DnsResolver::handle_dns_result(const int recursion_count,
189 const boost::system::error_code &error,
ad83004d 190 const std::size_t bytes_transferred)
36ad976b 191{
e18c1337 192 if (error)
36ad976b 193 {
e91538f0 194 GlobalLogger.info() << LogPrefix << "DNS resolve resulted in error "
fd62d09f 195 << error << " --> request retry";
cd71d095 196 schedule_retry(recursion_count);
e18c1337
CH
197 return;
198 }
199 else if ( OperationCancelled )
200 { // async_resolve was cancelled --> callbacks already called
201 GlobalLogger.info() << LogPrefix
202 << "Ignoring DNS results since we were cancelled";
36ad976b
CH
203 return;
204 }
205
ad83004d
CH
206 GlobalLogger.debug() << LogPrefix << "Handling DNS result ("
207 << bytes_transferred << " bytes transferred)";
208
36ad976b
CH
209 // next 3(+1) lines copied from boost/net/dns/resolver.hpp:
210 // clamp the recvBuffer with the number of bytes transferred or decode buffr
211 ReceiveBuffer.length(bytes_transferred);
4e7b6ff9 212 boost::net::dns::message result_message;
36ad976b
CH
213 result_message.decode( ReceiveBuffer );
214
ad83004d 215 // check ID
3f489614
CH
216 if (RequestId == 0)
217 { // system DNS or firewall might have longer timeout than me
218 // --> might receive replies for old requests whose time-out has expired
219 // --> have already called callbacks etc, so nothing to do with result
220 // TODO: did receive probably same data several times --> reset buffer?
221 GlobalLogger.info() << LogPrefix << "Ignoring reply to old DNS request "
222 << "(reply has ID " << std::showbase << std::hex
223 << result_message.id() << " or buffer not reset)";
224 return;
225 }
226 else if (RequestId != result_message.id())
227 {
49b82a1d 228 GlobalLogger.info() << LogPrefix << "Received answer for request ID "
ad83004d 229 << std::showbase << std::hex << result_message.id()
3f489614
CH
230 << " but expected ID " << RequestId << " --> ignore and retry";
231 schedule_retry(recursion_count);
232 return;
233 }
ad83004d
CH
234 else
235 GlobalLogger.debug() << LogPrefix << "Result has correct ID "
236 << std::showbase << std::hex << RequestId;
e18c1337 237 RequestId = 0;
ad83004d
CH
238
239 // loop over answers, remembering ips and cnames
4e7b6ff9
CH
240 // work with a regular pointer to list of answers since result_message is
241 // owner of data and that exists until end of function
242 // Items in answers list are shared_ptr to resource_base_t
e18c1337 243 std::vector<host_addr_pair> result_ips;
dbe986b9
CH
244 std::vector<src_cname_pair> result_cnames;
245 std::vector<string_pair> result_name_servers;
36ad976b 246
c1abff61 247 GlobalLogger.debug() << LogPrefix <<"Checking ANSWERS section of dns reply";
e18c1337 248 gather_results(result_message.answers(), &result_ips, &result_cnames,
dbe986b9 249 &result_name_servers);
ad83004d 250
e18c1337 251 // remember cname list (if there were any)
cd71d095
CH
252 // results should have the logical order
253 // Hostname [ --> cname1 --> cname2 --> ... --> cnameN ] [ --> ips ];
254 // otherwise just have unneccessary cnames in cache
dbe986b9 255 BOOST_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
e18c1337 256 ResolverBase::update_cache(host_and_cname.first, host_and_cname.second);
ad83004d 257
e18c1337 258 if ( !result_ips.empty() )
cd71d095 259 handle_ips( recursion_count, result_ips );
ad83004d 260 else if ( !result_cnames.empty() )
e18c1337
CH
261 // no IPs but at least one cname --> find the "last" cname and
262 // re-start resolving with that
cd71d095 263 handle_cname(recursion_count, result_cnames);
ad83004d 264 else
23f51766 265 { // no answers --> cannot proceed
49b82a1d 266 GlobalLogger.info() << LogPrefix << "No IP nor CNAME received! "
fd62d09f 267 << "--> request retry";
cd71d095 268 schedule_retry(recursion_count);
ad83004d 269 }
ad83004d
CH
270}
271
23f51766
CH
272/**
273 * gather IPs, CNAMEs and name servers from list of resource records;
274 *
275 * can be run on anwers(), autorities() and additional() sections of dns reply
276 * messages
cd71d095 277 *
23f51766
CH
278 * @param rr_list: input list of resource records
279 * @param result_ips: output vector of ips
280 * @param result_cnames: output vector of cnames
281 * @param result_name_servers: output vector of name servers
282 */
283void DnsResolver::gather_results(const boost::net::dns::rr_list_t *rr_list,
e18c1337 284 std::vector<host_addr_pair> *result_ips,
dbe986b9
CH
285 std::vector<src_cname_pair> *result_cnames,
286 std::vector<string_pair> *result_name_servers)
ad83004d
CH
287 const
288{
4e7b6ff9 289 using boost::net::dns::resource_base_t;
dbe986b9 290 boost::posix_time::ptime now =boost::posix_time::second_clock::local_time();
23f51766 291 BOOST_FOREACH( boost::shared_ptr<resource_base_t> rr_item, *rr_list )
36ad976b 292 {
4e7b6ff9 293 boost::net::dns::type_t rr_type = rr_item->rtype();
ad83004d
CH
294 uint32_t ttl = rr_item->ttl();
295 std::string domain = rr_item->domain();
dbe986b9
CH
296 std::string expiry =
297 boost::posix_time::to_simple_string(now + seconds(ttl));
36ad976b
CH
298
299 if (rr_type == boost::net::dns::type_a)
300 { // 'A' resource records carry IPv4 addresses
923626c0
CH
301 if (Protocol == DNS_IPv6)
302 {
e91538f0
CH
303 GlobalLogger.info() << LogPrefix << "Ignoring IPv4 address "
304 << "because resolver was configured to only use IPv6.";
923626c0
CH
305 continue;
306 }
4e7b6ff9
CH
307 boost::asio::ip::address_v4 ip =
308 ( dynamic_cast<boost::net::dns::a_resource *> (rr_item.get()) )
309 ->address();
e18c1337 310 result_ips->push_back(host_addr_pair(domain, HostAddress(ip, ttl)));
dbe986b9
CH
311 GlobalLogger.debug() << LogPrefix << domain << ": IPv4 " << ip
312 << " with TTL " << ttl << "s (until "
313 << expiry << ")";
36ad976b
CH
314 }
315 else if (rr_type == boost::net::dns::type_a6)
316 { // 'AAAA' resource records carry IPv6 addresses
923626c0
CH
317 if (Protocol == DNS_IPv4)
318 {
e91538f0
CH
319 GlobalLogger.info() << LogPrefix << "Ignoring IPv6 address "
320 << "because resolver was configured to only use IPv4.";
923626c0
CH
321 continue;
322 }
4e7b6ff9
CH
323 boost::asio::ip::address_v6 ip =
324 ( dynamic_cast<boost::net::dns::a6_resource *> (rr_item.get()) )
325 ->address();
e18c1337 326 result_ips->push_back(host_addr_pair(domain, HostAddress(ip, ttl)));
dbe986b9
CH
327 GlobalLogger.debug() << LogPrefix << domain << ": IPv6 " << ip
328 << " with TTL " << ttl << "s (until "
329 << expiry << ")";
36ad976b
CH
330 }
331 else if (rr_type == boost::net::dns::type_cname)
332 { // 'CNAME' resource records that carry aliases
4e7b6ff9
CH
333 std::string cname =
334 (dynamic_cast<boost::net::dns::cname_resource *>(rr_item.get()))
335 ->canonicalname();
dbe986b9
CH
336 result_cnames->push_back( src_cname_pair(domain,
337 Cname(cname, ttl)) );
338 GlobalLogger.debug() << LogPrefix << domain << ": CNAME to "
339 << cname << " with TTL " << ttl << "s (until "
340 << expiry << ")";
36ad976b
CH
341 }
342 else if (rr_type == boost::net::dns::type_ns)
dbe986b9
CH
343 { // NS (name_server) resource records
344 std::string name_server =
ad83004d
CH
345 (dynamic_cast<boost::net::dns::ns_resource *>(rr_item.get()))
346 ->nameserver();
dbe986b9
CH
347 result_name_servers->push_back( string_pair(domain, name_server) );
348 GlobalLogger.debug() << LogPrefix << "NameServer " << name_server
349 << " for " << domain << " with TTL " << ttl
350 << "s (until " << expiry << ")";
ad83004d 351 }
36ad976b 352 else if (rr_type == boost::net::dns::type_soa)
e91538f0 353 GlobalLogger.debug() << LogPrefix << "SOA resource";
36ad976b 354 else if (rr_type == boost::net::dns::type_ptr)
e91538f0 355 GlobalLogger.debug() << LogPrefix << "ptr resource";
36ad976b 356 else if (rr_type == boost::net::dns::type_hinfo)
e91538f0 357 GlobalLogger.debug() << LogPrefix << "hinfo resource";
36ad976b 358 else if (rr_type == boost::net::dns::type_mx)
e91538f0 359 GlobalLogger.debug() << LogPrefix << "mx resource";
36ad976b 360 else if (rr_type == boost::net::dns::type_txt)
e91538f0 361 GlobalLogger.debug() << LogPrefix << "txt resource";
36ad976b 362 else if (rr_type == boost::net::dns::type_srv)
e91538f0 363 GlobalLogger.debug() << LogPrefix << "srv resource";
36ad976b 364 else if (rr_type == boost::net::dns::type_axfr)
e91538f0 365 GlobalLogger.debug() << LogPrefix << "axfr resource";
36ad976b 366 else
ad83004d
CH
367 GlobalLogger.debug() << LogPrefix << "unknown resource type: "
368 << std::showbase << std::hex
369 << static_cast<unsigned>(rr_item->rtype());
36ad976b 370 }
36ad976b
CH
371}
372
373
cd71d095 374void DnsResolver::handle_unavailable(const int recursion_count)
36ad976b
CH
375{
376 // schedule new attempt in quite a while
4e7b6ff9 377 StaleDataLongtermTimer.expires_from_now(
fd62d09f 378 seconds(Config::StaleDataLongtermSeconds));
36ad976b 379 StaleDataLongtermTimer.async_wait(
cd71d095
CH
380 boost::bind( &DnsResolver::wait_timer_timeout_handler, this,
381 recursion_count,
382 boost::asio::placeholders::error
36ad976b
CH
383 )
384 );
fd62d09f 385 LongtermTimerIsActive = true;
36ad976b
CH
386
387 // for now, admit failure
3f489614 388 RequestId = 0; // do not accept answers from old requests
36ad976b 389 bool was_success = false;
cd71d095 390 finalize_resolve(was_success, recursion_count);
36ad976b
CH
391}
392
e18c1337 393
cd71d095
CH
394void DnsResolver::handle_ips(const int recursion_count,
395 const std::vector<host_addr_pair> &result_ips)
36ad976b 396{
cd71d095 397 // received at least one IP which could be for the queried host name
e18c1337
CH
398 // or the cname at the "end" of the cname list;
399 // but all IPs should be for the same
400 HostAddressVec addr_list;
401 std::string only_host_for_ips = result_ips[0].first;
402 BOOST_FOREACH( const host_addr_pair &host_and_addr, result_ips)
403 {
404 if ( host_and_addr.first != only_host_for_ips )
49b82a1d 405 GlobalLogger.info() << LogPrefix
e18c1337
CH
406 << "Received IPs for different hosts " << only_host_for_ips
407 << " and " << host_and_addr.first << " in one DNS result! "
408 << "--> ignore second";
409 else
410 {
49b82a1d 411 GlobalLogger.info() << LogPrefix << "Found IP "
e18c1337
CH
412 << host_and_addr.second.get_ip() << " with TTL "
413 << host_and_addr.second.get_ttl().get_value() << "s";
414 addr_list.push_back(host_and_addr.second);
415 }
416 }
417 ResolverBase::update_cache( only_host_for_ips, addr_list );
36ad976b 418
e18c1337
CH
419 // clean up
420 bool was_success = true;
cd71d095 421 finalize_resolve(was_success, recursion_count);
e18c1337
CH
422}
423
424
cd71d095
CH
425void DnsResolver::handle_cname(const int recursion_count,
426 const std::vector<src_cname_pair> &result_cnames)
e18c1337
CH
427{
428 // find the "last" cname in the list
429 // Hostname --> cname1 --> cname2 --> ... --> cnameN
430 // We assume here that this list might not be in order but that all cnames
cd71d095
CH
431 // form a single list (form one connected list and not several independent
432 // lists)
dbe986b9 433
e18c1337
CH
434 std::string last_cname = "";
435 bool could_be_last;
dbe986b9 436 BOOST_REVERSE_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
e18c1337
CH
437 {
438 could_be_last = true;
dbe986b9 439 BOOST_REVERSE_FOREACH( const src_cname_pair &other, result_cnames )
e18c1337 440 {
23f51766 441 if (other.first == host_and_cname.second.Host)
e18c1337
CH
442 { // found cname for current cname
443 could_be_last = false;
444 break;
445 }
446 }
447 if (could_be_last)
448 {
23f51766 449 last_cname = host_and_cname.second.Host;
e18c1337
CH
450 break;
451 }
452 }
453
454 if (last_cname.empty())
455 {
49b82a1d 456 GlobalLogger.info() << LogPrefix
e18c1337
CH
457 << "Could not identify \"last\" CNAME to handle -- "
458 << "maybe we encountered a CNAME loop? Anyway, cannot proceed!";
459 GlobalLogger.info() << LogPrefix << "Result CNAMEs were:";
dbe986b9 460 BOOST_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
e18c1337 461 GlobalLogger.info() << LogPrefix << host_and_cname.first << " --> "
23f51766 462 << host_and_cname.second.Host;
cd71d095 463 handle_unavailable(recursion_count);
e18c1337
CH
464 }
465 else
466 { // check cache for IP for this cname
467 bool check_up_to_date = true;
468 HostAddressVec cached_data = Cache->get_ips_recursive(last_cname,
469 check_up_to_date);
470 if ( !cached_data.empty() )
471 {
472 bool was_success = true;
cd71d095 473 finalize_resolve(was_success, recursion_count+1);
e18c1337
CH
474 }
475 else
476 { // get resolver for canonical name
e18c1337
CH
477 ResolverItem resolver = DnsMaster::get_instance()
478 ->get_resolver_for(last_cname, Protocol);
479 callback_type callback = boost::bind(
480 &DnsResolver::cname_resolve_callback,
481 this, _1, _2 );
cd71d095 482 resolver->async_resolve( callback, recursion_count+1 );
e18c1337
CH
483
484 // treat a CNAME as a partial result: not enough to run callbacks
485 // from finalize_resolve, but enough to stop timers and reset
486 // RetryCount --> name resolution can take longer
72be9e7d 487 stop_trying(true);
e18c1337
CH
488 }
489 }
36ad976b
CH
490}
491
492
cd71d095
CH
493/**
494 * the recursion_count here is really the one from the recursion, not the one
495 * forwarded from async_resolve!
496 */
e18c1337 497void DnsResolver::cname_resolve_callback(const bool was_success,
cd71d095 498 const int recursion_count)
36ad976b 499{
e18c1337
CH
500 if ( OperationCancelled )
501 { // async_resolve was cancelled --> callbacks already called
502 GlobalLogger.info() << LogPrefix
503 << "Ignoring CNAME results since we were cancelled";
504 return;
505 }
506 else if (was_success)
72be9e7d 507 {
cd71d095
CH
508 GlobalLogger.debug() << LogPrefix << "CNAME resolution succeeded after "
509 << recursion_count << " recursions";
510 finalize_resolve(was_success, recursion_count);
72be9e7d 511 }
96779587 512 else
72be9e7d 513 {
cd71d095
CH
514 GlobalLogger.info() << LogPrefix << "CNAME resolution failed after "
515 << recursion_count << " recursions";
e18c1337 516 // no use to schedule retry in this case since cname resolver must have
23f51766 517 // failed several times and we can only re-start the same procedure with
72be9e7d 518 // the same information. But can re-try later
cd71d095 519 handle_unavailable(recursion_count);
72be9e7d 520 }
dbe986b9
CH
521}
522
523
cd71d095
CH
524/**
525 * @brief always called at end of resolving process
526 *
527 * runs callbacks, resets timers and checks state consistency; only thing that
528 * is "left alive" is the long-term timer that might cause a re-start of
529 * resolution after a while
530 *
531 * @param was_success: indicates whether resolution was successfull
532 * @param recursion_count number of recursions or (if not successfull) negative
533 * value indicating who called this function
534 */
96779587 535void DnsResolver::finalize_resolve(const bool was_success,
cd71d095 536 const int recursion_count)
96779587 537{
e18c1337
CH
538 // some consistency checks; failure might indicate a situation I had not
539 // anticipated during programming but might not be harmfull yet
540 if ( !IsResolving )
541 GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
542 << "not resolving any more!";
543 if ( OperationCancelled )
544 GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
545 << " was cancelled!";
546 if ( ResolverBase::CallbackList.empty() )
547 GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
548 << "no callbacks!";
549 if ( RequestId != 0 )
550 GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
551 << "waiting for DNS reply!";
552
96779587 553 // stop timers
72be9e7d 554 stop_trying(was_success);
96779587
CH
555
556 // schedule callbacks, clearing callback list
cd71d095 557 ResolverBase::schedule_callbacks(was_success, recursion_count);
96779587
CH
558
559 // finalize
49b82a1d 560 GlobalLogger.info() << LogPrefix << "finalized resolve"
96779587 561 << " with success = " << was_success
cd71d095 562 << " and recursion_count = " << recursion_count;
96779587
CH
563 IsResolving = false;
564}
565
e18c1337 566
72be9e7d
CH
567/**
568 * arg was_success determines if stop trying forever or just for the moment
569 * --> determines if we cancel StaleDataLongtermTimer or not
570 */
571void DnsResolver::stop_trying(bool was_success)
96779587
CH
572{
573 // cancel timers
e91538f0 574 GlobalLogger.debug() << LogPrefix << "Cancelling timers";
96779587
CH
575 ResolveTimeoutTimer.cancel();
576 PauseBeforeRetryTimer.cancel();
72be9e7d
CH
577
578 if (was_success)
579 {
580 StaleDataLongtermTimer.cancel();
fd62d09f 581 LongtermTimerIsActive = false;
72be9e7d 582 }
96779587
CH
583
584 // clean up
585 RetryCount = 0;
586}
587
e18c1337 588
72be9e7d
CH
589/**
590 * return true if resolver is currently resolving
591 *
592 * Is true from call to async_resolve until callbacks
593 * --> returns true if waiting for result or (short-term) retry
594 *
cd71d095 595 * However, does NOT tell you if the (long-term) stale timeout is active!
72be9e7d
CH
596 * That timer has no effect on result, need to check is_waiting_to_resolve
597 * for that
598 */
599bool DnsResolver::is_resolving() const
e18c1337
CH
600{
601 return IsResolving;
602}
603
604
dbe986b9 605/**
72be9e7d 606 * returns true if either is_resolving or the long-term timer is active
cd71d095 607 *
72be9e7d
CH
608 * is_resolving returns true if the short-term retry timer is active
609 */
610bool DnsResolver::is_waiting_to_resolve() const
611{
fd62d09f 612 return IsResolving || LongtermTimerIsActive;
72be9e7d
CH
613}
614
615
616/**
dbe986b9
CH
617 * cancel a earlier call to async_resolve
618 *
619 * callbacks will be called with was_success=false; all internal operations
620 * will be cancelled and internal callbacks (timers, dns results) have no
72be9e7d 621 * effect any more; cancels also the long-term stale-data timer
dbe986b9 622 */
e18c1337
CH
623void DnsResolver::cancel_resolve()
624{
fd62d09f 625 if ( !IsResolving && !LongtermTimerIsActive)
e18c1337 626 {
fd62d09f
CH
627 GlobalLogger.info() << LogPrefix << "Cancel called on non-resolving, "
628 << "non-waiting resolver -- ignore";
e18c1337
CH
629 return;
630 }
631 else if (OperationCancelled)
632 {
633 GlobalLogger.info() << LogPrefix
634 << "Cancel called on cancelled resolver -- ignore";
635 return;
636 }
fd62d09f 637 GlobalLogger.info() << LogPrefix << "Cancel resolver";
e18c1337 638
e18c1337 639 // set before finalize_resolve so can check in finalize_resolve that ID is
dbe986b9 640 // always 0; ID is not used any more since handle_dns_result stops if
e18c1337
CH
641 // OperationCancelled is true
642 RequestId = 0;
643
fd62d09f
CH
644 if ( IsResolving )
645 {
646 bool was_success = false;
cd71d095
CH
647 int recursion_count = -1;
648 finalize_resolve(was_success, recursion_count);
fd62d09f 649 }
e18c1337 650
72be9e7d
CH
651 // also cancel the long-term timer
652 StaleDataLongtermTimer.cancel();
fd62d09f 653 LongtermTimerIsActive = false;
72be9e7d 654
cd71d095 655 // set after finalize_resolve, so can check in finalize_resolve that
dbe986b9 656 // OperationCancelled is never true
e18c1337
CH
657 OperationCancelled = true;
658
659}
660
661
cd71d095
CH
662void DnsResolver::handle_resolve_timeout(const int recursion_count,
663 const boost::system::error_code &error)
96779587
CH
664{
665 if ( error == boost::asio::error::operation_aborted ) // cancelled
666 {
fd62d09f
CH
667 GlobalLogger.debug() << LogPrefix
668 << "Resolve timeout timer was cancelled!";
96779587
CH
669 return;
670 }
671 else if (error)
672 {
49b82a1d 673 GlobalLogger.info() << LogPrefix
e91538f0 674 << "resolve timeout handler received error "
fd62d09f 675 << error << " --> request retry";
cd71d095 676 schedule_retry(recursion_count);
96779587 677 }
e18c1337
CH
678 else if ( OperationCancelled )
679 { // async_resolve was cancelled --> callbacks already called
680 GlobalLogger.info() << LogPrefix
dbe986b9 681 << "Ignoring DNS timeout since we were cancelled";
e18c1337
CH
682 return;
683 }
684 else
685 {
49b82a1d 686 GlobalLogger.info() << LogPrefix << "DNS resolving timed out";
cd71d095 687 schedule_retry(recursion_count);
e18c1337
CH
688 }
689}
690
96779587 691
cd71d095 692void DnsResolver::schedule_retry(const int recursion_count)
e18c1337 693{
23f51766 694 // cancel timers
e18c1337
CH
695 ResolveTimeoutTimer.cancel();
696 PauseBeforeRetryTimer.cancel();
36ad976b 697
96779587
CH
698 // increment timer
699 ++RetryCount;
700
923626c0
CH
701 if ( RetryCount > DnsMaster::get_instance()
702 ->get_max_address_resolution_attempts() )
e18c1337
CH
703 { // too many re-tries
704 GlobalLogger.info() << LogPrefix << "Not scheduling a retry since "
705 << "RetryCount " << RetryCount << " too high";
cd71d095
CH
706 handle_unavailable(recursion_count); // will call stop_trying
707 } // --> reset RetryCount
96779587
CH
708 else
709 { // schedule retry
e18c1337
CH
710 GlobalLogger.info() << LogPrefix << "Scheduling a retry (RetryCount="
711 << RetryCount << ")";
96779587 712 PauseBeforeRetryTimer.expires_from_now(
4e7b6ff9 713 seconds(Config::PauseBeforeRetrySeconds));
96779587 714 PauseBeforeRetryTimer.async_wait(
4e7b6ff9 715 boost::bind( &DnsResolver::wait_timer_timeout_handler,
cd71d095
CH
716 this, recursion_count,
717 boost::asio::placeholders::error) );
96779587
CH
718 }
719}
720
cd71d095 721void DnsResolver::wait_timer_timeout_handler( const int recursion_count,
96779587
CH
722 const boost::system::error_code &error)
723{
724 if ( error == boost::asio::error::operation_aborted ) // cancelled
e18c1337
CH
725 { // assume that our code cancelled this timer, so callbacks will be
726 // taken care of!
fd62d09f
CH
727 GlobalLogger.debug() << LogPrefix
728 << "Resolve wait timer was cancelled! ";
e18c1337 729 }
96779587 730 else if (error)
e18c1337
CH
731 { // not sure what to do here, but callers waiting forever for a callback
732 // is probably the worst thing to happen, so call finalize_resolve
49b82a1d 733 GlobalLogger.info() << LogPrefix
e18c1337
CH
734 << "resolve wait handler received error "
735 << error << "! Try to finalize resolve";
736 bool was_success = false;
cd71d095 737 finalize_resolve(was_success, recursion_count);
e18c1337
CH
738 }
739 else if ( OperationCancelled )
740 { // async_resolve was cancelled --> callbacks already called
741 GlobalLogger.info() << LogPrefix
742 << "Ignoring waiting timeout since we were cancelled";
743 return;
744 }
96779587
CH
745 else
746 {
e91538f0 747 GlobalLogger.info() << LogPrefix << "Done waiting --> re-try resolve";
fd62d09f 748 IsResolving = false; // will be set to true immediately in do_resolve
cd71d095 749 do_resolve(recursion_count);
96779587 750 }
36ad976b
CH
751}
752
753
96779587
CH
754//==============================================================================
755// RETRIEVAL
756//==============================================================================
757
26b0f687 758HostAddress DnsResolver::get_next_ip(bool check_up_to_date)
36ad976b
CH
759{
760 // get cached data
26b0f687 761 // (do not use arg check_up_to_date here in order to give NextIpIndex
fd62d09f 762 // a chance to stay above number of outdated IPs)
e18c1337 763 HostAddressVec cached_data = ResolverBase::get_cached_ips_recursively();
96779587
CH
764
765 // if no results cached, return default-constructed HostAddress (0.0.0.0)
26b0f687 766 HostAddress return_candidate;
96779587 767 if ( cached_data.empty() )
4e7b6ff9 768 {
26b0f687
CH
769 GlobalLogger.debug() << LogPrefix << "Get next IP: nothing cached";
770 return return_candidate;
4e7b6ff9 771 }
36ad976b 772
fd62d09f 773 std::size_t n_iter = 0;
26b0f687
CH
774 std::size_t n_ips = cached_data.size();
775 uint32_t ttl_thresh = static_cast<uint32_t>( DnsMaster::get_instance()
776 ->get_resolved_ip_ttl_threshold() );
36ad976b 777
26b0f687
CH
778 GlobalLogger.info() << LogPrefix << "Get next IP from cached result of "
779 << n_ips << " IPs; first index to consider is " << NextIpIndex
72be9e7d 780 << "; TTL thresh=" << ttl_thresh << "s is used: " << check_up_to_date;
26b0f687
CH
781
782 // loop until we have found a cached result (that is up to date)
783 // or until we have tried all cached IPs
784 while (true)
785 {
786 // check index since cache size may have changed since last call
787 if (NextIpIndex >= n_ips)
72be9e7d
CH
788 {
789 GlobalLogger.debug() << LogPrefix << "Reset NextIpIndex";
26b0f687 790 NextIpIndex = 0;
72be9e7d 791 }
26b0f687 792 else if ( n_iter >= n_ips)
72be9e7d
CH
793 {
794 GlobalLogger.debug() << LogPrefix << "No IP found";
26b0f687 795 return HostAddress(); // have checked all candidates
72be9e7d 796 }
26b0f687
CH
797 else
798 { // there are candidates left to consider
fd62d09f
CH
799 GlobalLogger.debug() << LogPrefix << "Check IP candidate at index "
800 << NextIpIndex;
26b0f687
CH
801 return_candidate = cached_data[NextIpIndex++];
802 if (!check_up_to_date)
fd62d09f
CH
803 {
804 GlobalLogger.debug() << LogPrefix << "not checking ttl, accept";
26b0f687 805 return return_candidate;
fd62d09f
CH
806 }
807 else if (return_candidate.get_ttl().get_updated_value()
26b0f687 808 > ttl_thresh)
fd62d09f
CH
809 {
810 GlobalLogger.debug() << LogPrefix << "is up to date, accept";
72be9e7d 811 return return_candidate;
fd62d09f 812 }
26b0f687 813 else
fd62d09f
CH
814 {
815 GlobalLogger.debug() << LogPrefix << "is out of date ("
816 << return_candidate.get_ttl().get_updated_value()
817 << "s <= " << ttl_thresh << "s), continue";
26b0f687 818 ++n_iter;
fd62d09f 819 }
26b0f687
CH
820 }
821 }
36ad976b
CH
822}
823
923626c0
CH
824bool DnsResolver::have_up_to_date_ip()
825{
fd62d09f 826 return get_resolved_ip_count(true) > 0;
923626c0
CH
827}
828
fd62d09f 829int DnsResolver::get_resolved_ip_count(const bool check_up_to_date)
923626c0 830{
fd62d09f
CH
831 // run with empty hostname --> uses internal var Hostname
832 return ResolverBase::get_cached_ips_recursively("",check_up_to_date).size();
923626c0
CH
833}
834