#include #include #include #include #include #include #include #include namespace llarp { namespace handlers { static void ExitHandlerRecvPkt(llarp_tun_io *tun, const llarp_buffer_t &buf) { static_cast< ExitEndpoint * >(tun->user)->OnInetPacket(buf); } static void ExitHandlerFlush(llarp_tun_io *tun) { auto *ep = static_cast< ExitEndpoint * >(tun->user); LogicCall(ep->GetRouter()->logic(), std::bind(&ExitEndpoint::Flush, ep)); } ExitEndpoint::ExitEndpoint(const std::string &name, AbstractRouter *r) : m_Router(r) , m_Resolver(std::make_shared< dns::Proxy >( r->netloop(), r->logic(), r->netloop(), r->logic(), this)) , m_Name(name) , m_Tun{{0}, 0, {0}, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr} , m_LocalResolverAddr("127.0.0.1", 53) , m_InetToNetwork(name + "_exit_rx", r->netloop(), r->netloop()) { m_Tun.user = this; m_Tun.recvpkt = &ExitHandlerRecvPkt; m_Tun.tick = &ExitHandlerFlush; m_ShouldInitTun = true; } ExitEndpoint::~ExitEndpoint() = default; util::StatusObject ExitEndpoint::ExtractStatus() const { util::StatusObject obj{{"permitExit", m_PermitExit}, {"ip", m_IfAddr.ToString()}}; util::StatusObject exitsObj{}; for(const auto &item : m_ActiveExits) { exitsObj[item.first.ToString()] = item.second->ExtractStatus(); } obj["exits"] = exitsObj; return obj; } bool ExitEndpoint::SupportsV6() const { return m_UseV6; } bool ExitEndpoint::ShouldHookDNSMessage(const dns::Message &msg) const { if(msg.questions.size() == 0) return false; // always hook ptr for ranges we own if(msg.questions[0].qtype == dns::qTypePTR) { huint128_t ip; if(!dns::DecodePTR(msg.questions[0].qname, ip)) return false; return m_OurRange.Contains(ip); } if(msg.questions[0].qtype == dns::qTypeA || msg.questions[0].qtype == dns::qTypeCNAME || msg.questions[0].qtype == dns::qTypeAAAA) { if(msg.questions[0].IsName("localhost.loki")) return true; if(msg.questions[0].HasTLD(".snode")) return true; } return false; } bool ExitEndpoint::HandleHookedDNSMessage( dns::Message &&msg, std::function< void(dns::Message) > reply) { if(msg.questions[0].qtype == dns::qTypePTR) { huint128_t ip; if(!dns::DecodePTR(msg.questions[0].qname, ip)) return false; if(ip == m_IfAddr) { RouterID us = GetRouter()->pubkey(); msg.AddAReply(us.ToString(), 300); } else { auto itr = m_IPToKey.find(ip); if(itr != m_IPToKey.end() && m_SNodeKeys.find(itr->second) != m_SNodeKeys.end()) { RouterID them = itr->second; msg.AddAReply(them.ToString()); } else msg.AddNXReply(); } } else if(msg.questions[0].qtype == dns::qTypeCNAME) { if(msg.questions[0].IsName("random.snode")) { RouterID random; if(GetRouter()->GetRandomGoodRouter(random)) msg.AddCNAMEReply(random.ToString(), 1); else msg.AddNXReply(); } else if(msg.questions[0].IsName("localhost.loki")) { RouterID us = m_Router->pubkey(); msg.AddAReply(us.ToString(), 1); } else msg.AddNXReply(); } else if(msg.questions[0].qtype == dns::qTypeA || msg.questions[0].qtype == dns::qTypeAAAA) { const bool isV6 = msg.questions[0].qtype == dns::qTypeAAAA; const bool isV4 = msg.questions[0].qtype == dns::qTypeA; if(msg.questions[0].IsName("random.snode")) { RouterID random; if(GetRouter()->GetRandomGoodRouter(random)) { msg.AddCNAMEReply(random.ToString(), 1); auto ip = ObtainServiceNodeIP(random); msg.AddINReply(ip, false); } else msg.AddNXReply(); reply(msg); return true; } if(msg.questions[0].IsName("localhost.loki")) { msg.AddINReply(GetIfAddr(), isV6); reply(msg); return true; } // forward dns for snode RouterID r; if(r.FromString(msg.questions[0].Name())) { huint128_t ip; PubKey pubKey(r); if(isV4 && SupportsV6()) { msg.hdr_fields |= dns::flags_QR | dns::flags_AA | dns::flags_RA; } else if(m_SNodeKeys.find(pubKey) == m_SNodeKeys.end()) { // we do not have it mapped, async obtain it ObtainSNodeSession( r, [&](std::shared_ptr< exit::BaseSession > session) { if(session && session->IsReady()) { msg.AddINReply(m_KeyToIP[pubKey], isV6); } else { msg.AddNXReply(); } reply(msg); }); return true; } else { // we have it mapped already as a service node auto itr = m_KeyToIP.find(pubKey); if(itr != m_KeyToIP.end()) { ip = itr->second; msg.AddINReply(ip, isV6); } else // fallback case that should never happen (probably) msg.AddNXReply(); } } else msg.AddNXReply(); } reply(msg); return true; } void ExitEndpoint::ObtainSNodeSession(const RouterID &router, exit::SessionReadyFunc obtainCb) { ObtainServiceNodeIP(router); m_SNodeSessions[router]->AddReadyHook(obtainCb); } llarp_time_t ExitEndpoint::Now() const { return m_Router->Now(); } bool ExitEndpoint::VisitEndpointsFor( const PubKey &pk, std::function< bool(exit::Endpoint *const) > visit) { auto range = m_ActiveExits.equal_range(pk); auto itr = range.first; while(itr != range.second) { if(visit(itr->second.get())) ++itr; else return true; } return false; } void ExitEndpoint::Flush() { m_InetToNetwork.Process([&](Pkt_t &pkt) { PubKey pk; { auto itr = m_IPToKey.find(pkt.dstv6()); if(itr == m_IPToKey.end()) { // drop LogWarn(Name(), " dropping packet, has no session at ", pkt.dstv6()); return; } pk = itr->second; } // check if this key is a service node if(m_SNodeKeys.find(pk) != m_SNodeKeys.end()) { // check if it's a service node session we made and queue it via our // snode session that we made otherwise use an inbound session that // was made by the other service node auto itr = m_SNodeSessions.find(pk); if(itr != m_SNodeSessions.end()) { if(itr->second->QueueUpstreamTraffic(pkt, routing::ExitPadSize)) return; } } auto tryFlushingTraffic = [&](exit::Endpoint *const ep) -> bool { if(!ep->QueueInboundTraffic(ManagedBuffer{pkt.Buffer()})) { LogWarn(Name(), " dropped inbound traffic for session ", pk, " as we are overloaded (probably)"); // continue iteration return true; } // break iteration return false; }; if(!VisitEndpointsFor(pk, tryFlushingTraffic)) { // we may have all dead sessions, wtf now? LogWarn(Name(), " dropped inbound traffic for session ", pk, " as we have no working endpoints"); } }); { auto itr = m_ActiveExits.begin(); while(itr != m_ActiveExits.end()) { if(!itr->second->Flush()) { LogWarn("exit session with ", itr->first, " dropped packets"); } ++itr; } } { auto itr = m_SNodeSessions.begin(); while(itr != m_SNodeSessions.end()) { // TODO: move flush upstream to router event loop if(!itr->second->FlushUpstream()) { LogWarn("failed to flush snode traffic to ", itr->first, " via outbound session"); } itr->second->FlushDownstream(); ++itr; } } m_Router->PumpLL(); } bool ExitEndpoint::Start() { // map our address const PubKey us(m_Router->pubkey()); const huint128_t ip = GetIfAddr(); m_KeyToIP[us] = ip; m_IPToKey[ip] = us; m_IPActivity[ip] = std::numeric_limits< llarp_time_t >::max(); m_SNodeKeys.insert(us); if(m_ShouldInitTun) { auto loop = GetRouter()->netloop(); if(!llarp_ev_add_tun(loop.get(), &m_Tun)) { llarp::LogWarn("Could not create tunnel for exit endpoint"); return false; } llarp::LogInfo("Trying to start resolver ", m_LocalResolverAddr.ToString()); return m_Resolver->Start(m_LocalResolverAddr, m_UpstreamResolvers); } return true; } AbstractRouter * ExitEndpoint::GetRouter() { return m_Router; } huint128_t ExitEndpoint::GetIfAddr() const { return m_IfAddr; } bool ExitEndpoint::Stop() { for(auto &item : m_SNodeSessions) item.second->Stop(); return true; } bool ExitEndpoint::ShouldRemove() const { for(auto &item : m_SNodeSessions) if(!item.second->ShouldRemove()) return false; return true; } bool ExitEndpoint::HasLocalMappedAddrFor(const PubKey &pk) const { return m_KeyToIP.find(pk) != m_KeyToIP.end(); } huint128_t ExitEndpoint::GetIPForIdent(const PubKey pk) { huint128_t found = {0}; if(!HasLocalMappedAddrFor(pk)) { // allocate and map found.h = AllocateNewAddress().h; if(!m_KeyToIP.emplace(pk, found).second) { LogError(Name(), "failed to map ", pk, " to ", found); return found; } if(!m_IPToKey.emplace(found, pk).second) { LogError(Name(), "failed to map ", found, " to ", pk); return found; } if(HasLocalMappedAddrFor(pk)) LogInfo(Name(), " mapping ", pk, " to ", found); else LogError(Name(), "failed to map ", pk, " to ", found); } else found.h = m_KeyToIP[pk].h; MarkIPActive(found); m_KeyToIP.rehash(0); assert(HasLocalMappedAddrFor(pk)); return found; } huint128_t ExitEndpoint::AllocateNewAddress() { if(m_NextAddr < m_HigestAddr) return ++m_NextAddr; // find oldest activity ip address huint128_t found = {0}; llarp_time_t min = std::numeric_limits< llarp_time_t >::max(); auto itr = m_IPActivity.begin(); while(itr != m_IPActivity.end()) { if(itr->second < min) { found.h = itr->first.h; min = itr->second; } ++itr; } // kick old ident off exit // TODO: DoS PubKey pk = m_IPToKey[found]; KickIdentOffExit(pk); return found; } bool ExitEndpoint::QueueOutboundTraffic(const llarp_buffer_t &buf) { return llarp_ev_tun_async_write(&m_Tun, buf); } void ExitEndpoint::KickIdentOffExit(const PubKey &pk) { LogInfo(Name(), " kicking ", pk, " off exit"); huint128_t ip = m_KeyToIP[pk]; m_KeyToIP.erase(pk); m_IPToKey.erase(ip); auto range = m_ActiveExits.equal_range(pk); auto exit_itr = range.first; while(exit_itr != range.second) exit_itr = m_ActiveExits.erase(exit_itr); } void ExitEndpoint::MarkIPActive(huint128_t ip) { m_IPActivity[ip] = GetRouter()->Now(); } void ExitEndpoint::OnInetPacket(const llarp_buffer_t &buf) { m_InetToNetwork.EmplaceIf( [b = ManagedBuffer(buf)](Pkt_t &pkt) -> bool { return pkt.Load(b); }); } bool ExitEndpoint::QueueSNodePacket(const llarp_buffer_t &buf, huint128_t from) { net::IPPacket pkt; if(!pkt.Load(buf)) return false; // rewrite ip pkt.UpdateIPv6Address(from, m_IfAddr); return llarp_ev_tun_async_write(&m_Tun, pkt.Buffer()); } exit::Endpoint * ExitEndpoint::FindEndpointByPath(const PathID_t &path) { exit::Endpoint *endpoint = nullptr; PubKey pk; { auto itr = m_Paths.find(path); if(itr == m_Paths.end()) return nullptr; pk = itr->second; } { auto itr = m_ActiveExits.find(pk); if(itr != m_ActiveExits.end()) { if(itr->second->PubKey() == pk) endpoint = itr->second.get(); } } return endpoint; } bool ExitEndpoint::UpdateEndpointPath(const PubKey &remote, const PathID_t &next) { // check if already mapped auto itr = m_Paths.find(next); if(itr != m_Paths.end()) return false; m_Paths.emplace(next, remote); return true; } bool ExitEndpoint::SetOption(const std::string &k, const std::string &v) { if(k == "type" && v == "null") { m_ShouldInitTun = false; return true; } if(k == "exit") { m_PermitExit = IsTrueValue(v.c_str()); return true; } if(k == "local-dns") { std::string resolverAddr = v; uint16_t dnsport = 53; auto pos = v.find(":"); if(pos != std::string::npos) { resolverAddr = v.substr(0, pos); dnsport = std::atoi(v.substr(pos + 1).c_str()); } m_LocalResolverAddr = Addr(resolverAddr, dnsport); LogInfo(Name(), " local dns set to ", m_LocalResolverAddr); } if(k == "upstream-dns") { std::string resolverAddr = v; uint16_t dnsport = 53; auto pos = v.find(":"); if(pos != std::string::npos) { resolverAddr = v.substr(0, pos); dnsport = std::atoi(v.substr(pos + 1).c_str()); } m_UpstreamResolvers.emplace_back(resolverAddr, dnsport); LogInfo(Name(), " adding upstream dns set to ", resolverAddr, ":", dnsport); } if(k == "ifaddr") { if(!m_OurRange.FromString(v)) { LogError(Name(), " has invalid address range: ", v); return false; } auto pos = v.find("/"); if(pos == std::string::npos) { LogError(Name(), " ifaddr is not a cidr: ", v); return false; } std::string nmask_str = v.substr(1 + pos); std::string host_str = v.substr(0, pos); // string, or just a plain char array? strncpy(m_Tun.ifaddr, host_str.c_str(), sizeof(m_Tun.ifaddr) - 1); m_Tun.netmask = std::atoi(nmask_str.c_str()); m_IfAddr = m_OurRange.addr; m_NextAddr = m_IfAddr; m_HigestAddr = m_OurRange.HighestAddr(); LogInfo(Name(), " set ifaddr range to ", m_Tun.ifaddr, "/", m_Tun.netmask, " lo=", m_IfAddr, " hi=", m_HigestAddr); } if(k == "ifname") { if(v.length() >= sizeof(m_Tun.ifname)) { LogError(Name() + " ifname '", v, "' is too long"); return false; } strncpy(m_Tun.ifname, v.c_str(), sizeof(m_Tun.ifname) - 1); LogInfo(Name(), " set ifname to ", m_Tun.ifname); } if(k == "exit-whitelist") { // add exit policy whitelist rule // TODO: implement me return true; } if(k == "exit-blacklist") { // add exit policy blacklist rule // TODO: implement me return true; } return true; } huint128_t ExitEndpoint::ObtainServiceNodeIP(const RouterID &other) { const PubKey pubKey(other); const PubKey us(m_Router->pubkey()); // just in case if(pubKey == us) return m_IfAddr; huint128_t ip = GetIPForIdent(pubKey); if(m_SNodeKeys.emplace(pubKey).second) { auto session = std::make_shared< exit::SNodeSession >( other, std::bind(&ExitEndpoint::QueueSNodePacket, this, std::placeholders::_1, ip), GetRouter(), 2, 1, true, false); // this is a new service node make an outbound session to them m_SNodeSessions.emplace(other, session); } return ip; } bool ExitEndpoint::AllocateNewExit(const PubKey pk, const PathID_t &path, bool wantInternet) { if(wantInternet && !m_PermitExit) return false; auto ip = GetIPForIdent(pk); if(GetRouter()->pathContext().TransitHopPreviousIsRouter(path, pk.as_array())) { // we think this path belongs to a service node // mark it as such so we don't make an outbound session to them m_SNodeKeys.emplace(pk.as_array()); } m_ActiveExits.emplace(pk, std::make_unique< exit::Endpoint >( pk, path, !wantInternet, ip, this)); m_Paths[path] = pk; return HasLocalMappedAddrFor(pk); } std::string ExitEndpoint::Name() const { return m_Name; } void ExitEndpoint::DelEndpointInfo(const PathID_t &path) { m_Paths.erase(path); } void ExitEndpoint::RemoveExit(const exit::Endpoint *ep) { auto range = m_ActiveExits.equal_range(ep->PubKey()); auto itr = range.first; while(itr != range.second) { if(itr->second->LocalPath() == ep->LocalPath()) { itr = m_ActiveExits.erase(itr); // now ep is gone af return; } ++itr; } } void ExitEndpoint::Tick(llarp_time_t now) { { auto itr = m_SNodeSessions.begin(); while(itr != m_SNodeSessions.end()) { if(itr->second->IsExpired(now)) itr = m_SNodeSessions.erase(itr); else { itr->second->Tick(now); ++itr; } } } { // expire auto itr = m_ActiveExits.begin(); while(itr != m_ActiveExits.end()) { if(itr->second->IsExpired(now)) itr = m_ActiveExits.erase(itr); else ++itr; } // pick chosen exits and tick m_ChosenExits.clear(); itr = m_ActiveExits.begin(); while(itr != m_ActiveExits.end()) { // do we have an exit set for this key? if(m_ChosenExits.find(itr->first) != m_ChosenExits.end()) { // yes if(m_ChosenExits[itr->first]->createdAt < itr->second->createdAt) { // if the iterators's exit is newer use it for the chosen exit for // key if(!itr->second->LooksDead(now)) m_ChosenExits[itr->first] = itr->second.get(); } } else if(!itr->second->LooksDead( now)) // set chosen exit if not dead for key that doesn't // have one yet m_ChosenExits[itr->first] = itr->second.get(); // tick which clears the tx rx counters itr->second->Tick(now); ++itr; } } } } // namespace handlers } // namespace llarp