[pingcheck] / src / dns / dnsresolver.cpp

/*
 The software in this package is distributed under the GNU General
 Public License version 2 (with a special exception described below).

 A copy of GNU General Public License (GPL) is included in this distribution,
 in the file COPYING.GPL.

 As a special exception, if other files instantiate templates or use macros
 or inline functions from this file, or you compile this file and link it
 with other works to produce a work based on this file, this file
 does not by itself cause the resulting work to be covered
 by the GNU General Public License.

 However the source code for this file must still be made available
 in accordance with section (3) of the GNU General Public License.

 This exception does not invalidate any other reasons why a work based
 on this file might be covered by the GNU General Public License.

 Christian Herdtweck, Intra2net AG 2015

 with code copied from boost::net::dns::resolve.hpp
 by Andreas Haberstroh (andreas at ibusy dot com)
    from https://github.com/softwareace/Boost.DNS
 */

#include "dns/dnsresolver.h"

#include <stdint.h>
#include <sstream>

#include <boost/foreach.hpp>
#include <boost/bind.hpp>
#include <boost/function.hpp>
#include <boost/net/dns.hpp>
#include <boost/date_time/posix_time/posix_time.hpp>
#include <boost/uuid/uuid.hpp>
#include <boost/uuid/uuid_io.hpp>

#include <logfunc.hpp>

using I2n::Logger::GlobalLogger;
using boost::posix_time::seconds;
using boost::posix_time::minutes;

namespace Config
{
    const int ResolveTimeoutSeconds = 3;
    const int PauseBeforeRetrySeconds = 10;
    const int StaleDataLongtermMinutes = 15;
    const int DNS_PORT = 53;
}

DnsResolver::DnsResolver(IoServiceItem &io_serv,
                         const std::string &hostname,
                         const DnsIpProtocol &protocol,
                         const DnsCacheItem cache,
                         const boost::asio::ip::address &name_server)
    : ResolverBase( io_serv, hostname, cache )
    , Socket( *io_serv, ip::udp::endpoint(ip::udp::v4(), 0))
        // just connect to anything, will specify sender/receiver later
    , ReceiveBuffer()
    , RequestBuffer()
    , Protocol( protocol )
    , NameServer( name_server, Config::DNS_PORT )
    , ResolveTimeoutTimer( *io_serv )
    , PauseBeforeRetryTimer( *io_serv )
    , StaleDataLongtermTimer( *io_serv )
    , NextIpIndex( 0 )
    , RetryCount( 0 )
    , IsResolving( false )
    , LogPrefix( "DnsResolver" )
    , RandomIdGenerator()
    , RequestId( 0 )
    , OperationCancelled( false )
{
    std::stringstream temp;
    temp << "Dns(" << ResolverBase::Hostname << "): ";
    LogPrefix = temp.str();

}

DnsResolver::~DnsResolver()
{
    boost::system::error_code error;
    //Socket.shutdown(boost::asio::ip::udp::socket::shutdown_both, error);
    //if ( error )
    //    GlobalLogger.warning() << LogPrefix << "Received error " << error
    //                           << " when shutting down socket for DNS";
    // in IcmpPinger always gave an error system:9 (EBADF: Bad file descriptor)
    // Here gives error system:107 ENOTCONN: Transport endpoint is not connected

    Socket.close(error);
    if ( error )
        GlobalLogger.warning() << LogPrefix << "Received error " << error
                               << " when closing socket for DNS";
}


//==============================================================================
// ASYNC RESOLVE
//==============================================================================

/**
 * copied here code from boost::net::dns::resolve.hpp, since want async
 * operation and that is used only internally, there
 */
void DnsResolver::do_resolve()
{
    // check if resolving already
    if (IsResolving)
    {
        GlobalLogger.info() << LogPrefix
            << "Call to do_resolve ignored since resolving already";
        return;
    }
    IsResolving = true;
    OperationCancelled = false;

    GlobalLogger.info() << LogPrefix << "start resolving for IPs of type "
        << to_string(Protocol) << " using name server " << NameServer;

    // just to be sure: cancel timers
    ResolveTimeoutTimer.cancel();
    PauseBeforeRetryTimer.cancel();
    StaleDataLongtermTimer.cancel();

    // create DNS request
    boost::net::dns::message dns_message( ResolverBase::Hostname, Protocol );
    dns_message.recursive(true);
    dns_message.action(boost::net::dns::message::query);
    dns_message.opcode(boost::net::dns::message::squery);

    // create random ID for message
    boost::uuids::uuid message_id = RandomIdGenerator();
    memcpy( &RequestId, message_id.data, sizeof(RequestId) );
    dns_message.id( RequestId );
    GlobalLogger.debug() << LogPrefix << "Request has ID "
                               << std::showbase << std::hex << dns_message.id();

    // setup receipt of reply
    Socket.async_receive_from(
            boost::asio::buffer(ReceiveBuffer.get_array()),
            NameServer,
            boost::bind( &DnsResolver::handle_dns_result, this,
                         boost::asio::placeholders::error,
                         boost::asio::placeholders::bytes_transferred)
    );

    // schedule timeout
    (void) ResolveTimeoutTimer.expires_from_now(
                                        seconds(Config::ResolveTimeoutSeconds));
    ResolveTimeoutTimer.async_wait( boost::bind(
                                      &DnsResolver::handle_resolve_timeout,
                                      this, boost::asio::placeholders::error) );

    // send dns request
    dns_message.encode(RequestBuffer);
    size_t bytes_sent;
    try
    {
        bytes_sent = Socket.send_to(
                                boost::asio::buffer(RequestBuffer.get_array()),
                                NameServer );
    }
    catch (boost::system::system_error &err)
    {
        GlobalLogger.warning() << LogPrefix
                               << "Sending of DNS request message failed: "
                               << err.what();
        schedule_retry();
        return;
    }

    if ( bytes_sent == 0 )
    {
        GlobalLogger.warning() << LogPrefix << "Empty DNS request sent!";
        schedule_retry();
        return;
    }
}


void DnsResolver::handle_dns_result(const boost::system::error_code &error,
                                    const std::size_t bytes_transferred)
{
    if (error)
    {
        GlobalLogger.info() << LogPrefix << "DNS resolve resulted in error "
                            << error << " --> try again after a little while";
        schedule_retry();
        return;
    }
    else if ( OperationCancelled )
    {   // async_resolve was cancelled --> callbacks already called
        GlobalLogger.info() << LogPrefix
                            << "Ignoring DNS results since we were cancelled";
        return;
    }

    GlobalLogger.debug() << LogPrefix << "Handling DNS result ("
                         << bytes_transferred << " bytes transferred)";

    // next 3(+1) lines copied from boost/net/dns/resolver.hpp:
    // clamp the recvBuffer with the number of bytes transferred or decode buffr
    ReceiveBuffer.length(bytes_transferred);
    boost::net::dns::message result_message;
    result_message.decode( ReceiveBuffer );

    // check ID
    if (RequestId != result_message.id())
        GlobalLogger.warning() << LogPrefix << "Received answer for request ID "
           << std::showbase << std::hex << result_message.id()
           << " but expected ID " << RequestId;
    else
        GlobalLogger.debug() << LogPrefix << "Result has correct ID "
                                      << std::showbase << std::hex << RequestId;
    RequestId = 0;

    // loop over answers, remembering ips and cnames
    // work with a regular pointer to list of answers since result_message is
    //   owner of data and that exists until end of function
    // Items in answers list are shared_ptr to resource_base_t
    std::vector<host_addr_pair> result_ips;
    std::vector<src_cname_pair> result_cnames;
    std::vector<string_pair> result_name_servers;

    GlobalLogger.debug() << "Checking ANSWERS section of dns reply";
    gather_results(result_message.answers(), &result_ips, &result_cnames,
                                                          &result_name_servers);
    // results should have the logical order
    // Hostname [ --> cname1 --> cname2 --> ... --> cnameN ] [ --> ips ]

    // remember cname list (if there were any)
    BOOST_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
        ResolverBase::update_cache(host_and_cname.first, host_and_cname.second);

    if ( !result_ips.empty() )
        handle_ips( result_ips );
    else if ( !result_cnames.empty() )
        // no IPs but at least one cname --> find the "last" cname and
        // re-start resolving with that
        handle_cname(result_cnames);
    else
    {   // no answers --> cannot proceed
        GlobalLogger.warning() << LogPrefix << "No IP nor CNAME received! "
                               << "--> schedule retry";
        schedule_retry();
    }
}

/**
 * gather IPs, CNAMEs and name servers from list of resource records;
 *
 * can be run on anwers(), autorities() and additional() sections of dns reply
 * messages
 * 
 * @param rr_list: input list of resource records
 * @param result_ips: output vector of ips
 * @param result_cnames: output vector of cnames
 * @param result_name_servers: output vector of name servers
 */
void DnsResolver::gather_results(const boost::net::dns::rr_list_t *rr_list,
                                 std::vector<host_addr_pair> *result_ips,
                                 std::vector<src_cname_pair> *result_cnames,
                                 std::vector<string_pair> *result_name_servers)
                                                                           const
{
    using boost::net::dns::resource_base_t;
    boost::posix_time::ptime now =boost::posix_time::second_clock::local_time();
    BOOST_FOREACH( boost::shared_ptr<resource_base_t> rr_item, *rr_list )
    {
        boost::net::dns::type_t rr_type = rr_item->rtype();
        uint32_t ttl = rr_item->ttl();
        std::string domain = rr_item->domain();
        std::string expiry =
                        boost::posix_time::to_simple_string(now + seconds(ttl));

        if (rr_type == boost::net::dns::type_a)
        {    // 'A' resource records carry IPv4 addresses
            if (Protocol == DNS_IPv6)
            {
                GlobalLogger.info() << LogPrefix << "Ignoring IPv4 address "
                        << "because resolver was configured to only use IPv6.";
                continue;
            }
            boost::asio::ip::address_v4 ip =
                ( dynamic_cast<boost::net::dns::a_resource *> (rr_item.get()) )
                ->address();
            result_ips->push_back(host_addr_pair(domain, HostAddress(ip, ttl)));
            GlobalLogger.debug() << LogPrefix << domain << ": IPv4 " << ip
                                 << " with TTL " << ttl << "s (until "
                                 << expiry << ")";
        }
        else if (rr_type == boost::net::dns::type_a6)
        {   // 'AAAA' resource records carry IPv6 addresses
            if (Protocol == DNS_IPv4)
            {
                GlobalLogger.info() << LogPrefix << "Ignoring IPv6 address "
                        << "because resolver was configured to only use IPv4.";
                continue;
            }
            boost::asio::ip::address_v6 ip =
                ( dynamic_cast<boost::net::dns::a6_resource *> (rr_item.get()) )
                ->address();
            result_ips->push_back(host_addr_pair(domain, HostAddress(ip, ttl)));
            GlobalLogger.debug() << LogPrefix << domain << ": IPv6 " << ip
                                 << " with TTL " << ttl << "s (until "
                                 << expiry << ")";
        }
        else if (rr_type == boost::net::dns::type_cname)
        {   // 'CNAME' resource records that carry aliases
            std::string cname =
                (dynamic_cast<boost::net::dns::cname_resource *>(rr_item.get()))
                ->canonicalname();
            result_cnames->push_back( src_cname_pair(domain,
                                                     Cname(cname, ttl)) );
            GlobalLogger.debug() << LogPrefix << domain << ": CNAME to "
                                 << cname << " with TTL " << ttl << "s (until "
                                 << expiry << ")";
        }
        else if (rr_type == boost::net::dns::type_ns)
        {   // NS (name_server) resource records
            std::string name_server =
                (dynamic_cast<boost::net::dns::ns_resource *>(rr_item.get()))
                ->nameserver();
            result_name_servers->push_back( string_pair(domain, name_server) );
            GlobalLogger.debug() << LogPrefix << "NameServer " << name_server
                                 << " for " << domain << " with TTL " << ttl
                                 << "s (until " << expiry << ")";
        }
        else if (rr_type == boost::net::dns::type_soa)
            GlobalLogger.debug() << LogPrefix << "SOA resource";
        else if (rr_type == boost::net::dns::type_ptr)
            GlobalLogger.debug() << LogPrefix << "ptr resource";
        else if (rr_type == boost::net::dns::type_hinfo)
            GlobalLogger.debug() << LogPrefix << "hinfo resource";
        else if (rr_type == boost::net::dns::type_mx)
            GlobalLogger.debug() << LogPrefix << "mx resource";
        else if (rr_type == boost::net::dns::type_txt)
            GlobalLogger.debug() << LogPrefix << "txt resource";
        else if (rr_type == boost::net::dns::type_srv)
            GlobalLogger.debug() << LogPrefix << "srv resource";
        else if (rr_type == boost::net::dns::type_axfr)
            GlobalLogger.debug() << LogPrefix << "axfr resource";
        else
            GlobalLogger.debug() << LogPrefix << "unknown resource type: "
                                 << std::showbase << std::hex
                                 << static_cast<unsigned>(rr_item->rtype());
    }
}


void DnsResolver::handle_unavailable()
{
    // schedule new attempt in quite a while
    StaleDataLongtermTimer.expires_from_now(
                                     minutes(Config::StaleDataLongtermMinutes));
    StaleDataLongtermTimer.async_wait(
            boost::bind( &DnsResolver::wait_timer_timeout_handler,
                         this, boost::asio::placeholders::error
            )
    );

    // for now, admit failure
    bool was_success = false;
    finalize_resolve(was_success);
}


void DnsResolver::handle_ips(const std::vector<host_addr_pair> &result_ips)
{
    // received at least one IP which could be for the queried host name 
    // or the cname at the "end" of the cname list;
    // but all IPs should be for the same
    HostAddressVec addr_list;
    std::string only_host_for_ips = result_ips[0].first;
    BOOST_FOREACH( const host_addr_pair &host_and_addr, result_ips)
    {
        if ( host_and_addr.first != only_host_for_ips )
            GlobalLogger.warning() << LogPrefix
                << "Received IPs for different hosts " << only_host_for_ips
                << " and " << host_and_addr.first << " in one DNS result! "
                << "--> ignore second";
        else
        {
            GlobalLogger.notice() << LogPrefix << "Found IP "
                      << host_and_addr.second.get_ip() << " with TTL "
                      << host_and_addr.second.get_ttl().get_value() << "s";
            addr_list.push_back(host_and_addr.second);
        }
    }
    ResolverBase::update_cache( only_host_for_ips, addr_list );

    // clean up
    bool was_success = true;
    finalize_resolve(was_success);
}


void DnsResolver::handle_cname(const std::vector<src_cname_pair> &result_cnames)
{
    // find the "last" cname in the list
    // Hostname --> cname1 --> cname2 --> ... --> cnameN
    // We assume here that this list might not be in order but that all cnames
    //   form a single list (form one connected list and not several isolated)

    std::string last_cname = "";
    bool could_be_last;
    BOOST_REVERSE_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
    {
        could_be_last = true;
        BOOST_REVERSE_FOREACH( const src_cname_pair &other, result_cnames )
        {
            if (other.first == host_and_cname.second.Host)
            {   // found cname for current cname
                could_be_last = false;
                break;
            }
        }
        if (could_be_last)
        {
            last_cname = host_and_cname.second.Host;
            break;
        }
    }

    if (last_cname.empty())
    {
        GlobalLogger.error() << LogPrefix
            << "Could not identify \"last\" CNAME to handle -- "
            << "maybe we encountered a CNAME loop? Anyway, cannot proceed!";
        GlobalLogger.info() << LogPrefix << "Result CNAMEs were:";
        BOOST_FOREACH( const src_cname_pair &host_and_cname, result_cnames )
            GlobalLogger.info() << LogPrefix << host_and_cname.first << " --> "
                                             << host_and_cname.second.Host;
        handle_unavailable();
    }
    else
    {   // check cache for IP for this cname
        bool check_up_to_date = true;
        HostAddressVec cached_data = Cache->get_ips_recursive(last_cname,
                                                              check_up_to_date);
        if ( !cached_data.empty() )
        {
            bool was_success = true;
            int cname_count = 1;  // define cache access as only 1
            finalize_resolve(was_success, cname_count);
        }
        else
        {   // get resolver for canonical name
            ResolverItem resolver = DnsMaster::get_instance()
                                    ->get_resolver_for(last_cname, Protocol);
            callback_type callback = boost::bind(
                                           &DnsResolver::cname_resolve_callback,
                                           this, _1, _2 );
            resolver->async_resolve( callback );

            // treat a CNAME as a partial result: not enough to run callbacks
            // from finalize_resolve, but enough to stop timers and reset
            // RetryCount --> name resolution can take longer
            stop_trying();
        }
    }
}


void DnsResolver::cname_resolve_callback(const bool was_success,
                                         const int cname_count)
{
    if ( OperationCancelled )
    {   // async_resolve was cancelled --> callbacks already called
        GlobalLogger.info() << LogPrefix
                            << "Ignoring CNAME results since we were cancelled";
        return;
    }
    else if (was_success)
        GlobalLogger.debug() << LogPrefix << "CNAME resolution succeeded";
    else
        GlobalLogger.info() << LogPrefix << "CNAME resolution failed";
        // no use to schedule retry in this case since cname resolver must have
        // failed several times and we can only re-start the same procedure with
        // the same information

    // cname counts like one more recursion step ...
    finalize_resolve(was_success, cname_count+1);
}


void DnsResolver::finalize_resolve(const bool was_success,
                                   const int cname_count)
{
    // some consistency checks; failure might indicate a situation I had not
    // anticipated during programming but might not be harmfull yet
    if ( !IsResolving )
        GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
                                            << "not resolving any more!";
    if ( OperationCancelled )
        GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
                                            << " was cancelled!";
    if ( ResolverBase::CallbackList.empty() )
        GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
                                            << "no callbacks!";
    if ( RequestId != 0 )
        GlobalLogger.warning() << LogPrefix << "Consistency check failed: "
                                            << "waiting for DNS reply!";

    // stop timers
    stop_trying();

    // schedule callbacks, clearing callback list
    ResolverBase::schedule_callbacks(was_success, cname_count);

    // finalize
    GlobalLogger.notice() << LogPrefix << "finalized resolve"
                          << " with success = " << was_success
                          << " and cname_count = " << cname_count;
    IsResolving = false;
}


void DnsResolver::stop_trying()
{
    // cancel timers
    GlobalLogger.debug() << LogPrefix << "Cancelling timers";
    ResolveTimeoutTimer.cancel();
    PauseBeforeRetryTimer.cancel();
    StaleDataLongtermTimer.cancel();

    // clean up
    RetryCount = 0;
}


bool DnsResolver::is_resolving()
{
    return IsResolving;
}


/**
 * cancel a earlier call to async_resolve
 *
 * callbacks will be called with was_success=false; all internal operations
 * will be cancelled and internal callbacks (timers, dns results) have no
 * effect any more
 */
void DnsResolver::cancel_resolve()
{
    if ( !IsResolving )
    {
        GlobalLogger.info() << LogPrefix
               << "Cancel called on non-resolving resolver -- ignore";
        return;
    }
    else if (OperationCancelled)
    {
        GlobalLogger.info() << LogPrefix
               << "Cancel called on cancelled resolver -- ignore";
        return;
    }

    // set before finalize_resolve so can check in finalize_resolve that ID is
    //   always 0; ID is not used any more since handle_dns_result stops if
    //   OperationCancelled is true
    RequestId = 0;

    bool was_success = false;
    int cname_count = 1;
    finalize_resolve(was_success, cname_count);

    // set after finalize_resolve, so can check in finalize_resolve that 
    // OperationCancelled is never true
    OperationCancelled = true;

}


void DnsResolver::handle_resolve_timeout(const boost::system::error_code &error)
{
    if ( error ==  boost::asio::error::operation_aborted )   // cancelled
    {
        GlobalLogger.warning() << LogPrefix
                               << "Resolve timeout timer was cancelled!";
        return;
    }
    else if (error)
    {
        GlobalLogger.warning() << LogPrefix
                               << "resolve timeout handler received error "
                               << error << " --> retry";
        schedule_retry();
    }
    else if ( OperationCancelled )
    {   // async_resolve was cancelled --> callbacks already called
        GlobalLogger.info() << LogPrefix
                            << "Ignoring DNS timeout since we were cancelled";
        return;
    }
    else
    {
        GlobalLogger.notice() << LogPrefix << "DNS resolving timed out";
        schedule_retry();
    }
}


void DnsResolver::schedule_retry()
{
    // cancel timers
    ResolveTimeoutTimer.cancel();
    PauseBeforeRetryTimer.cancel();

    // increment timer
    ++RetryCount;

    if ( RetryCount > DnsMaster::get_instance()
                      ->get_max_address_resolution_attempts() )
    {   // too many re-tries
        GlobalLogger.info() << LogPrefix << "Not scheduling a retry since "
                            << "RetryCount " << RetryCount << " too high";
        handle_unavailable();   // will call stop_trying i.e. reset RetryCount
    }
    else
    {   // schedule retry
        GlobalLogger.info() << LogPrefix << "Scheduling a retry (RetryCount="
                            << RetryCount << ")";
        PauseBeforeRetryTimer.expires_from_now(
                seconds(Config::PauseBeforeRetrySeconds));
        PauseBeforeRetryTimer.async_wait(
                boost::bind( &DnsResolver::wait_timer_timeout_handler,
                             this, boost::asio::placeholders::error) );
    }
}

void DnsResolver::wait_timer_timeout_handler(
                                         const boost::system::error_code &error)
{
    if ( error ==  boost::asio::error::operation_aborted )   // cancelled
    {   // assume that our code cancelled this timer, so callbacks will be
        // taken care of!
        GlobalLogger.warning() << LogPrefix
                               << "Resolve wait timer was cancelled! ";
    }
    else if (error)
    {   // not sure what to do here, but callers waiting forever for a callback
        // is probably the worst thing to happen, so call finalize_resolve
        GlobalLogger.warning() << LogPrefix
                               << "resolve wait handler received error "
                               << error << "! Try to finalize resolve";
        bool was_success = false;
        finalize_resolve(was_success);
    }
    else if ( OperationCancelled )
    {   // async_resolve was cancelled --> callbacks already called
        GlobalLogger.info() << LogPrefix
                          << "Ignoring waiting timeout since we were cancelled";
        return;
    }
    else
    {
        GlobalLogger.info() << LogPrefix << "Done waiting --> re-try resolve";
        do_resolve();
    }
}


//==============================================================================
// RETRIEVAL
//==============================================================================

HostAddress DnsResolver::get_next_ip(bool check_up_to_date)
{
    // get cached data
    // (do not use arg check_up_to_date here in order to give NextIpIndex
    //  a chance to stay above number of outdate IPs)
    HostAddressVec cached_data = ResolverBase::get_cached_ips_recursively();

    // if no results cached, return default-constructed HostAddress (0.0.0.0)
    HostAddress return_candidate;
    if ( cached_data.empty() )
    {
        GlobalLogger.debug() << LogPrefix << "Get next IP: nothing cached";
        return return_candidate;
    }

    int n_iter = 0;
    std::size_t n_ips = cached_data.size();
    uint32_t ttl_thresh = static_cast<uint32_t>( DnsMaster::get_instance()
                                            ->get_resolved_ip_ttl_threshold() );

    GlobalLogger.info() << LogPrefix << "Get next IP from cached result of "
        << n_ips << " IPs; first index to consider is " << NextIpIndex
        << "; TTL thresh = " << ttl_thresh << " is used: " << check_up_to_date;

    // loop until we have found a cached result (that is up to date)
    //   or until we have tried all cached IPs
    while (true)
    {
        // check index since cache size may have changed since last call
        if (NextIpIndex >= n_ips)
            NextIpIndex = 0;
        else if ( n_iter >= n_ips)
            return HostAddress();   // have checked all candidates
        else
        {   // there are candidates left to consider
            return_candidate = cached_data[NextIpIndex++];
            if (!check_up_to_date)
                return return_candidate;
            else if (cached_data[NextIpIndex].get_ttl().get_updated_value()
                     > ttl_thresh)
                return cached_data[++NextIpIndex];
            else
                ++n_iter;
        }
    }
}

bool DnsResolver::have_up_to_date_ip()
{
    return get_resolved_ip_count() > 0;
}

int DnsResolver::get_resolved_ip_count()
{
    return ResolverBase::get_cached_ips_recursively("", true).size();
}

// (created using vim -- the world's best text editor)