root/trunk/pdns/pdns/recursor_cache.cc @ 934

#include "recursor_cache.hh"
#include "misc.hh"
#include <iostream>
#include <boost/shared_ptr.hpp>
#include "dnsrecords.hh"
#include "arguments.hh"
#include "syncres.hh"

using namespace std;
using namespace boost;

#include "config.h"

#ifdef GCC_SKIP_LOCKING
#include <bits/atomicity.h>
// This code is ugly but does speedup the recursor tremendously on multi-processor systems, and even has a large effect (20, 30%) on uniprocessor 
namespace __gnu_cxx
{
  _Atomic_word
  __attribute__ ((__unused__))
  __exchange_and_add(volatile _Atomic_word* __mem, int __val)
  {
    register _Atomic_word __result=*__mem;
    *__mem+=__val;
    return __result;
  }

  void
  __attribute__ ((__unused__))
  __atomic_add(volatile _Atomic_word* __mem, int __val)
  {
    *__mem+=__val;
  }
}
#endif

string simpleCompress(const string& label)
{
  typedef vector<pair<unsigned int, unsigned int> > parts_t;
  parts_t parts;
  vstringtok(parts, label, ".");
  string ret;
  ret.reserve(label.size()+4);
  for(parts_t::const_iterator i=parts.begin(); i!=parts.end(); ++i) {
    ret.append(1, (char)(i->second - i->first));
    ret.append(label.c_str() + i->first, i->second - i->first);
  }
  ret.append(1, (char)0);
  return ret;
}

void simpleExpandTo(const string& label, unsigned int frompos, string& ret)
{
  unsigned int labellen=0;
  while((labellen=label.at(frompos++))) {
    ret.append(label.c_str()+frompos, labellen);
    ret.append(1,'.');
    frompos+=labellen;
  }
}

DNSResourceRecord String2DNSRR(const string& qname, const QType& qt, const string& serial, uint32_t ttd)
{
  DNSResourceRecord rr;
  rr.ttl=ttd; 
  rr.qtype=qt;
  rr.qname=qname;

  if(rr.qtype.getCode()==QType::A) {
    uint32_t ip;
    memcpy((char*)&ip, serial.c_str(), 4);
    rr.content=U32ToIP(ntohl(ip));
  }
  else if(rr.qtype.getCode()==QType::CNAME || rr.qtype.getCode()==QType::NS || rr.qtype.getCode()==QType::PTR) {
    unsigned int frompos=0;
    unsigned char labellen;

    while((labellen=serial.at(frompos++))) {
      if((labellen & 0xc0) == 0xc0) {
        string encoded=simpleCompress(qname);
        uint16_t offset=256*(labellen & ~0xc0) + (unsigned int)serial.at(frompos++) - sizeof(dnsheader)-5;

        simpleExpandTo(encoded, offset, rr.content);
        //      cerr<<"Oops, fallback, content so far: '"<<rr.content<<"', offset: "<<offset<<", '"<<qname<<"', "<<qt.getName()<<"\n";
        break;
      }
      rr.content.append(serial.c_str()+frompos, labellen);
      frompos+=labellen;
      rr.content.append(1,'.');
    }
    if(rr.content.empty())
      rr.content=".";
  }
  else {
    shared_ptr<DNSRecordContent> regen=DNSRecordContent::unserialize(qname, qt.getCode(), serial);
    rr.content=regen->getZoneRepresentation();
  }
  rr.content.reserve(0);
  rr.qname.reserve(0);
  return rr;
}

string DNSRR2String(const DNSResourceRecord& rr)
{
  uint16_t type=rr.qtype.getCode();

  if(type==QType::A) {
    uint32_t ip;
    IpToU32(rr.content, &ip);
    return string((char*)&ip, 4);
  }
  else if(type==QType::NS) {
    NSRecordContent ar(rr.content);
    return ar.serialize(rr.qname);
  }
  else if(type==QType::CNAME) {
    CNAMERecordContent ar(rr.content);
    return ar.serialize(rr.qname);
  }
  else {
    string ret;
    shared_ptr<DNSRecordContent> drc(DNSRecordContent::mastermake(type, 1, rr.content));
    ret=drc->serialize(rr.qname);
  //  cerr<<"stored '"<<rr.qname<<" '"<<rr.qtype.getName()<<"' '"<<rr.content<<"' as "<<ret.size()<<" bytes"<<endl;
    return ret;
  }
}

unsigned int MemRecursorCache::size()
{
  return (unsigned int)d_cache.size();
}

unsigned int MemRecursorCache::bytes()
{
  unsigned int ret=0;

  for(cache_t::const_iterator i=d_cache.begin(); i!=d_cache.end(); ++i) {
    ret+=(unsigned int)i->d_qname.length();
    for(vector<StoredRecord>::const_iterator j=i->d_records.begin(); j!= i->d_records.end(); ++j)
      ret+=j->size();
  }
  return ret;
}

int MemRecursorCache::getDirect(time_t now, const char* qname, const QType& qt, uint32_t ttd[10], char* data[10], uint16_t len[10])
{
  if(!d_cachecachevalid || Utility::strcasecmp(d_cachedqname.c_str(), qname)) {
//    cerr<<"had cache cache miss for '"<<qname<<"'"<<endl;
    d_cachedqname=qname;
    d_cachecache=d_cache.equal_range(tie(qname));
    d_cachecachevalid=true;
  }
  else
    ;
  //    cerr<<"had cache cache hit!"<<endl;

  if(d_cachecache.first == d_cachecache.second) {
    g_stats.noShuntNoMatch++;
    return false;
  }

  pair<cache_t::iterator, cache_t::iterator> range = d_cachecache;
  
  unsigned int n=0;
  for(;range.first != range.second; ++range.first) {
    if(range.first->d_qtype == QType::CNAME) { // if we see a cname, we need the whole shebang (for now)
      g_stats.noShuntCNAME++;
      return false;
    }
    if(range.first->d_qtype != qt.getCode())
      continue;
    if(range.first->getTTD() < (unsigned int) now) {
      g_stats.noShuntExpired++;
      return false;
    }
    
    if(range.first->d_records.empty() || range.first->d_records.size() > 9 ) {
      g_stats.noShuntSize++;
      return false;
    }
    
    size_t limit=range.first->d_records.size();
    n=0;
    for(; n < limit; ++n) {
      data[n]=(char*)range.first->d_records[n].d_string.c_str();
      len[n]=range.first->d_records[n].d_string.length();
      ttd[n]=range.first->d_records[n].d_ttd;
    }
    if(n<10) {
      data[n]=0;
      typedef cache_t::nth_index<1>::type sequence_t;
      sequence_t& sidx=d_cache.get<1>();
      sequence_t::iterator si=d_cache.project<1>(range.first);
      sidx.relocate(sidx.end(), si); // move it in the LRU list 
      // can't yet return, need to figure out if there isn't a CNAME that messes things up
    }
    else
      return false;
  }
  if(!n)
    g_stats.noShuntNoMatch++;
  return n;

}

int MemRecursorCache::get(time_t now, const string &qname, const QType& qt, set<DNSResourceRecord>* res)
{
  unsigned int ttd=0;

  //  cerr<<"looking up "<< qname+"|"+qt.getName()<<"\n";

  if(!d_cachecachevalid || Utility::strcasecmp(d_cachedqname.c_str(), qname.c_str())) {
    //    cerr<<"had cache cache miss"<<endl;
    d_cachedqname=qname;
    d_cachecache=d_cache.equal_range(tie(qname));
    d_cachecachevalid=true;
  }
  else
    ;
  //    cerr<<"had cache cache hit!"<<endl;


  if(res)
    res->clear();

  if(d_cachecache.first!=d_cachecache.second) { 
    for(cache_t::const_iterator i=d_cachecache.first; i != d_cachecache.second; ++i) 
      if(i->d_qtype == qt.getCode() || qt.getCode()==QType::ANY) {
        typedef cache_t::nth_index<1>::type sequence_t;
        sequence_t& sidx=d_cache.get<1>();
        sequence_t::iterator si=d_cache.project<1>(i);
        
        for(vector<StoredRecord>::const_iterator k=i->d_records.begin(); k != i->d_records.end(); ++k) {
          if(k->d_ttd < 1000000000 || k->d_ttd > (uint32_t) now) {
            ttd=k->d_ttd;
            if(res) {
              DNSResourceRecord rr=String2DNSRR(qname, QType(i->d_qtype),  k->d_string, ttd); 
              res->insert(rr);
            }
          }
        }
        if(res) {
          if(res->empty())
            sidx.relocate(sidx.begin(), si); 
          else
            sidx.relocate(sidx.end(), si); 
        }
        if(qt.getCode()!=QType::ANY) // normally if we have a hit, we are done
          break;
      }

    //    cerr<<"time left : "<<ttd - now<<", "<< (res ? res->size() : 0) <<"\n";
    return (unsigned int)ttd-now;
  }
  return -1;
}
 
/* the code below is rather tricky - it basically replaces the stuff cached for qname by content, but it is special
   cased for when inserting identical records with only differing ttls, in which case the entry is not
   touched, but only given a new ttd */
void MemRecursorCache::replace(time_t now, const string &qname, const QType& qt,  const set<DNSResourceRecord>& content, bool auth)
{
  d_cachecachevalid=false;
  tuple<string, uint16_t> key=make_tuple(qname, qt.getCode());
  cache_t::iterator stored=d_cache.find(key);

  //  cerr<<"storing "<< qname+"|"+qt.getName()<<" -> '"<<content.begin()->content<<"'\n";

  bool isNew=false;
  if(stored == d_cache.end()) {
    stored=d_cache.insert(CacheEntry(key,vector<StoredRecord>(), auth)).first;
    isNew=true;
  }
  
  pair<vector<StoredRecord>::iterator, vector<StoredRecord>::iterator> range;

  StoredRecord dr;
  CacheEntry ce=*stored;

  if(qt.getCode()==QType::SOA || qt.getCode()==QType::CNAME)  // you can only have one (1) each of these
    ce.d_records.clear();

  if(auth && !ce.d_auth) {
    ce.d_records.clear(); // clear non-auth data
    ce.d_auth = true;
    isNew=true;           // data should be sorted again
  }

  if(!auth && ce.d_auth) {  // unauth data came in, we have some auth data, but is it fresh?
    vector<StoredRecord>::iterator j;
    for(j = ce.d_records.begin() ; j != ce.d_records.end(); ++j) 
      if((time_t)j->d_ttd > now) 
        break;
    if(j != ce.d_records.end()) { // we still have valid data, ignore unauth data
      return;
    }
    else {
      ce.d_auth = false;  // new data won't be auth
    }
  }

  for(set<DNSResourceRecord>::const_iterator i=content.begin(); i != content.end(); ++i) {
    dr.d_ttd=i->ttl;
    dr.d_string=DNSRR2String(*i);
    
    if(isNew) 
      ce.d_records.push_back(dr);
    else {
      range=equal_range(ce.d_records.begin(), ce.d_records.end(), dr);
      
      if(range.first != range.second) {
        for(vector<StoredRecord>::iterator j=range.first ; j!=range.second; ++j) {
          /* see http://mailman.powerdns.com/pipermail/pdns-users/2006-May/003413.html */
          if(j->d_ttd > (unsigned int) now && i->ttl > j->d_ttd && qt.getCode()==QType::NS && auth) // don't allow auth servers to *raise* TTL of an NS record
            continue;
          if(i->ttl > j->d_ttd || (auth && d_followRFC2181) ) // authoritative packets can override the TTL to be lower
            j->d_ttd=i->ttl;
        }
      }
      else {
        ce.d_records.push_back(dr);
        sort(ce.d_records.begin(), ce.d_records.end());
      }
    }
  }
  if(isNew) {
    sort(ce.d_records.begin(), ce.d_records.end());
  }

  if(ce.d_records.capacity() != ce.d_records.size())
    vector<StoredRecord>(ce.d_records).swap(ce.d_records);

  d_cache.replace(stored, ce);
}

int MemRecursorCache::doWipeCache(const string& name, uint16_t qtype)
{
  int count=0;
  d_cachecachevalid=false;
  pair<cache_t::iterator, cache_t::iterator> range;
  if(qtype==0xffff)
    range=d_cache.equal_range(tie(name));
  else
    range=d_cache.equal_range(tie(name, qtype));

  for(cache_t::const_iterator i=range.first; i != range.second; ) {
    count++;
    d_cache.erase(i++);
  }
  return count;
}

void MemRecursorCache::doDumpAndClose(int fd)
{
  FILE* fp=fdopen(fd, "w");
  if(!fp) {
    close(fd);
    return;
  }

  typedef cache_t::nth_index<1>::type sequence_t;
  sequence_t& sidx=d_cache.get<1>();

  time_t now=time(0);
  for(sequence_t::const_iterator i=sidx.begin(); i != sidx.end(); ++i) {
    for(vector<StoredRecord>::const_iterator j=i->d_records.begin(); j != i->d_records.end(); ++j) {
      try {
        DNSResourceRecord rr=String2DNSRR(i->d_qname, QType(i->d_qtype), j->d_string, j->d_ttd - now);
        fprintf(fp, "%s %d IN %s %s\n", rr.qname.c_str(), rr.ttl, rr.qtype.getName().c_str(), rr.content.c_str());
      }
      catch(...) {
        fprintf(fp, "; error printing '%s'\n", i->d_qname.c_str());
      }
    }
  }
  fclose(fp);
}

void MemRecursorCache::doSlash(int perc)
{
  doPrune();
}

void MemRecursorCache::doPrune(void)
{
  uint32_t now=(uint32_t)time(0);
  d_cachecachevalid=false;

  unsigned int maxCached=::arg().asNum("max-cache-entries");
  unsigned int toTrim=0;
  
  unsigned int cacheSize=d_cache.size();

  if(maxCached && cacheSize > maxCached) {
    toTrim = cacheSize - maxCached;
  }

  //  cout<<"Need to trim "<<toTrim<<" from cache to meet target!\n";

  typedef cache_t::nth_index<1>::type sequence_t;
  sequence_t& sidx=d_cache.get<1>();

  unsigned int tried=0, lookAt, erased=0;

  // two modes - if toTrim is 0, just look through 10000 records and nuke everything that is expired
  // otherwise, scan first 5*toTrim records, and stop once we've nuked enough
  if(toTrim)
    lookAt=5*toTrim;
  else
    lookAt=cacheSize/10;


  sequence_t::iterator iter=sidx.begin(), eiter;
  for(; iter != sidx.end() && tried < lookAt ; ++tried) {
    unsigned int ttd=iter->getTTD();
    if(ttd < now) { 
      sidx.erase(iter++);
      erased++;
    }
    else
      ++iter;

    if(toTrim && erased > toTrim)
      break;
  }

  //  cout<<"erased "<<erased<<" records based on ttd\n";
  
  if(erased >= toTrim)
    return;

  //  if(toTrim)
  //    cout<<"Still have "<<toTrim - erased<<" entries left to erase to meet target\n";


  eiter=iter=sidx.begin();
  std::advance(eiter, toTrim);
  sidx.erase(iter, eiter);
}